Anders Storm

$SEYE 30% Down in one Day — The Brutal Market of 2026 - but IFRS in Q3 might transform H1 2026 to positive EBIT

Anders Storm — Wed, 26 Aug 2026 20:39:33 GMT

We’ve now seen Klarna and Smart Eye both fall brutally after reports that were quite good.

This kind of drop used to only happen when a company announced a share issue with a 35% discount, or when a fraud was uncovered, or when a major customer walked away. Not when companies deliver double-digit organic growth, expanding margins, and/or reaffirmed guidance (Klarna did not, that was their issue) .

But apparently in 2026, if you have a challenging history of missed promises, you get hit hard even when you deliver good numbers. Even if you have not promised anything, SEYE had not guide any numbers, but the market expected more.

Klarna certainly has that history. Peak valuation of $45.6 billion in 2021, collapsed 85% to $6.7 billion by 2022. IPO delayed multiple times. Listed in September 2025 at $15.1 billion, 69% below peak. Still unprofitable through H1 2025 despite years of promising the turn to sustained profitability. When Klarna beat every Q2 estimate on August 18 and trimmed 2026 guidance by 5-6% due to German consumer weakness, the stock fell 22% in a day and 30% for the week, the worst week since IPO.

Smart Eye has its own version of that history. Analysts published aggressive base case targets of SEK 155-160 in 2020-2022 that did not materialize. Affärsvärlden projected SEK 1 billion in revenue for 2026 back in January 2024. The stock fell from over SEK 250 in 2020-2021 to below SEK 50 in 2024. Repeated share issues to fund the gap between contract wins and revenue realization. A market that had been trained to distrust the timing of the inflection.

I want to walk through Smart Eye’s actual Q2 report using the slides the company presented today and explain why I’ve mostly been shaking my head and laughing at the market’s reaction while increasing my position on the dip.

I’m long SEYE, with a 3–6x share-price target in 2026. I’m not particularly concerned that roughly SEK 30 million in revenue was “delayed.” But let’s dig into today’s report.

Slide 1: The Q2 Summary the Market Ignored

Look at this slide honestly and tell me this looks like a damaged business breaking down:

Net Sales SEK 141M with +54% organic growth
EBITDA SEK 27M, improved by SEK 26M year-over-year
Automotive organic growth +128%
Automotive License growth +200%
Car models reached production: 175 (up from 155)
OEMs reached production: 20 (up from 15)
Free cash flow improved by SEK 14M

The two negatives on the slide: Behavioral Research -7% (management said this stabilizes and Q4 seasonally stronger) and OPEX excluding amortization -17 MSEK (planned investment in sales & marketing plus salary revisions).

This is a company at operational inflection. The market chose to focus on the fact that Redeye had estimated Q2 Automotive at SEK 122M and it came in at SEK 93M. The delta between “excellent” and “aggressive-sell-side-fantasy” got priced as failure.

My favorite picture to use in these cases is this one:

379 Design Wins

This slide is central to the long-term thesis and it hasn’t changed today.

The yellow line is total design wins accumulated since 2016, at 379. The green line is design wins that have entered production, currently at 175. The red line at the bottom is exited wins, only 15 out of 379, a 4% attrition rate.

The remaining SEK 7.6 billion in future order value from these design wins is real. It’s already contracted. The question was never whether these wins would deliver revenue, the question was always about the timing of the ramp.

175 in production today. Management confirmed all active design-wins will be in production before end of 2030. That’s what the CEO said explicitly on the call.

If you take that guidance at face value and I do, because it’s a stated commitment against a contractual backlog, not a hope then between now and 2030 an additional 204 design wins enter production. That’s 51 additional wins per year on average, which is consistent with the acceleration slope we’ve seen since 2023.

Where the Automotive Ramp Actually Stands

This is the most important slide from today for understanding the near-term trajectory. 175 models in production at end of H1 2026. Guidance for full year 2026: 200-250 (250 is the stretch target, 200 is base). But here’s what the CEO said on the call that’s not visible on the slide:

“80% of the 175 models currently in production are actively producing at full rate today. The remaining 20% will reach full production during Q3”. My added comments is that Automotive always has a much stronger volumes in Q4 every year.

“25+ accelerating in 2H 2026”. Minimum 25 additional model starts in H2 2026, then acceleration in 2027.

“Most is coming in before end of 2030”. The remaining 204 design wins (379 total minus 175 in production) enter production between now and end of 2030.

Think about what this means for Q3 alone. If 35 models transition from partial to full production, and 12-15 new models start production in Q3, the aggregate production volume increases roughly 30-35% sequentially. The Q3 report on November 17 has structural tailwinds that Q2 didn’t. There guidance hence looks quite good for rest of 2026.

The Per-Vehicle Pricing Story — SEK 7.6B Order Value

SEK 7.6 billion in future order value from the current design wins. Let me walk through what the CEO said about per-vehicle economics because this changes how you should think about future revenue.

Current per-vehicle royalty: 3-10 Euro per car. That’s SEK 34-113 per car depending on contract vintage.

Why the current blended average is at the lower end: contracts signed in 2020-21 during the DMS market ramp had lower pricing to secure market share. These contracts are what’s driving production volumes right now, hence the lower blended ASP that Redeye estimated at SEK 42.

Why low-cost cars pull the average down: for low cost vehicle models, GSR compliance is a legal requirement, not a value driver. OEMs pay minimum viable royalty to check the regulatory box.

Why future cars carry higher per-vehicle value: future models don’t just have DMS. They have DMS + Interior Sensing + posture detection + face ID + iris ID + intoxication detection. Each function adds royalty layer. The CEO said future cars have “high value” per vehicle from this bundle of features.

This is the pricing expansion story that Redeye’s SEK 42 estimate captures for 2026 but doesn’t capture for 2028-2030 when Interior Sensing and adjacent features become standard.

If per-vehicle royalty expands from SEK 42 in 2026 toward SEK 70-90 in 2028-2030 as newer contracts with higher feature content dominate the mix, plus vehicle production volume grows 3x and royalty revenue expansion on ASP0 This is what I like to hear and that support my long term estimates.

The Financial Trend the Market Missed

R12 Net Sales: SEK 488.8M, up from SEK 361.6M, that’s +35% year-over-year on trailing twelve months.

The rolling 12-month chart on the left tells the real story. Net sales R12 turned the corner in Q4 2025 and has been accelerating each quarter. Q2 2026 at SEK 140.9M is the highest single quarter in company history.

The waterfall on the right shows how Q2 2026 built to SEK 141M from Q2 2025’s SEK 92M:

Automotive contributed +SEK 53M (the main growth engine)
Behavioral Research contributed -SEK 4M (the temporary headwind)
Sightic acquisition contributed +SEK 3M
FX effect contributed -SEK 3M

The organic Automotive contribution alone was +58% of the total sales base from a year ago. This is not a company decelerating.

R12 revenue growth of +35% with +43% organic. Q2 organic growth of +54%. That’s an accelerating pattern.

EBITDA Trajectory and the IFRS Question

This slide might be the most important one for the medium-term thesis.

Q2 2024: EBITDA -22.5M
Q2 2025: EBITDA +3.6M
Q2 2026: EBITDA +31M (adjusted for one-time items)

The rolling 12-month EBITDA line is now trending upward at approximately +80 MSEK annualized run-rate. This is the operational leverage the company has been promising for years finally showing up in the numbers. How can this warrant a 30% drop in share price. This is hilarious, who is it drivning such a sell off based on this kind of fantastic development?

Now to the note the market completely missed: “Q3 IFRS?”

On the earnings call, the CFO confirmed again that Smart Eye is transitioning from K3 to IFRS accounting starting Q3 2026. This is not a cosmetic change. It fundamentally changes how the reported P&L looks.

The IFRS Impact That Changes Everything on EBIT

Look at the Q2 report’s balance sheet:

Goodwill on the balance sheet at end 2025: SEK 453.2 million.

Under Swedish K3 accounting, goodwill is amortized systematically over its useful life, typically 5-10 years. Smart Eye amortized SEK 85 million of goodwill in 2025 alone. That’s SEK 21 million per quarter.

Under IFRS, goodwill is not amortized. Instead, it’s tested annually for impairment. So starting Q3 2026, that SEK 21 million per quarter charge disappears from the reported EBIT line. This is also a big deal for 1H and full 2026. When Smart Eye reports Q3 on November 17, the retroactive IFRS restatement will show that H1 2026 was already profitable on an EBIT basis, not the loss that was reported using K3.

H1 2026 EBIT under IFRS (estimate):

H1 2026 K3 EBIT: -SEK 36,5M
Plus H1 goodwill amortization (SEK 21,4M × 2 Quarters): +SEK 42,8M
H1 2026 IFRS EBIT: approx.: +SEK 6,3M positiv

Q3 has a genuine chance to show positive reported EBIT for the first time in company history. Not only because operations changing. Also because the accounting standard changed to match global software peers.

This matters since many institutional funds have mandate restrictions against companies reporting operating losses regardless of underlying economics. Smart Eye transitioning to reported operating profit unlocks a broader institutional investor base.

Also, comparison to peers becomes cleaner. ARM Holdings, Cadence, Synopsys, none of them amortize goodwill because US GAAP doesn’t require it. When Smart Eye moves to IFRS, apples-to-apples comparison finally works.

What I Got Wrong in My August 11 Post

Before continuing I want to be transparent about two errors in my original piece.

First, I anchored the Q2 estimate on Redeye’s aggressive number published three days before I wrote. Actual Q2 Automotive came in at SEK 93M vs Redeye’s SEK 122M. The peer read-through from Seeing Machines Redeye used didn’t translate mechanically to Smart Eye. I should have been more sceptical of that.

Corrected 2026 full year estimate: SEK 640-650M, still +55-60% year-over-year growth, still Rule of 40 above 70. The market is not inline with this is clear, since the share dropped 30% in a single-day drop on a report that showed +128% organic Automotive growth.

Valuation After the Drop

At SEK 66.40 today Smart Eye’s market cap is SEK 2.6 billion. On R12 revenue of SEK 488.8M growing 35%, that’s 5.3x P/S. On corrected 2026 estimate of SEK 645M, forward P/S is 4x.

Peer benchmarks:

Cerence (Rule of 40 = 3, declining revenue): 3-4x P/S
Synopsys (Rule of 40 = 49): 12x P/S
Cadence (Rule of 40 = 51): 15-17x P/S
ARM Holdings (Rule of 40 = 71): 25-30x P/S

Smart Eye on Q2 numbers (+53% revenue growth + 19.5% EBITDA margin = Rule of 40 = 72) trades at Cerence multiples despite ARM-level fundamentals.

a) Either something is structurally broken in the business that management is not disclosing, b) I am as insane as the X account insane_analyst or c) the market has totally lost confidence in SEYEs and it is a truly damage stock not a damaged company, as the market value it as of now. I bet on the last c)..but it can be b) the as well :-). Time will tell.

Why I Bought More

I bought more today for four specific reasons:

First, the operational data confirmed the thesis. +128% organic Automotive growth. +200% royalty acceleration. 175 models in production with 80% actively producing today. This is not a business breaking down.

Second, the 2030 visibility got clearer. Management said all design wins will be in production by end of 2030. That’s not a hope. That’s a stated commitment against a SEK 7.6 billion contractual backlog. The cohort thesis just got stronger.

Third, IFRS transition changes the Q3 optics dramatically. The K3 goodwill amortization of ~SEK 21M per quarter goes away. Q3 has a real chance to show positive reported EBIT for the first time ever. Psychological threshold that unlocks broader institutional investors.

Fourth, the math got better. From SEK 66.40 to my 2030 target of approximately SEK 300-600 (bear/bull) is roughly 9-10x return, or approximately 74% IRR over four years. From the original SEK 97.96 entry it was 6x, or 58% IRR. Same target, better entry.

What Would Change My Mind

I want to be honest about this because publishing an investment thesis means holding it accountable.

If Q3 on November 17 shows material deceleration in royalty growth, that changes the picture. If design wins in production come in below 200 at year-end, the ramp curve is broken rather than delayed. If management guidance for 2027 comes in dramatically below current cohort trajectory, the long-term thesis needs recalibration.

None of these happened today. What happened today was that Q2 came in below aggressive estimates while confirming the underlying trajectory. That’s what got sold at 30% in a day.

Closing Thought

I don’t have confidence about the next few weeks. Stocks that drop 30% in a day often keep drifting lower as momentum unwinds and forced selling completes. Smart Eye may trade lower before it trades higher.

What I have confidence about is the operational trajectory and the 2030 target. If management delivers on 200-250 models in production by year-end, IFRS-adjusted positive EBIT in Q3 and all design-wins-in-production by 2030, then today’s SEK 66.40 is going to look like a gift in hindsight.

Klarna beat every estimate on August 18 and fell +30% in a week. Smart Eye delivered +128% organic Automotive growth today and fell 30% in a day. This is what the 2026 market does to companies with a history of missed promises even when they deliver.

I’ve done my homework. I bought more at these levels. I’m targeting approximately SEK 600 by 2030.

The market can do whatever the market does between now and November 17.

Anders Storm
Disclosure: The author holds positions in stocks discussed in this publication, including Smart Eye. Position sizes and timing are not disclosed. This analysis is for informational purposes only and does not constitute investment advice. Readers should conduct their own due diligence and consult licensed advisors for personal investment decisions.

AXT: The Fundamental Material/Substrate Player That is the Base for Any InP Laser. It's Also the Base in the $LAZR ETF

Anders Storm — Sat, 22 Aug 2026 16:39:12 GMT

Back in May, I wrote about the almost-unknown materials company, at least in Sweden. AXT had 300 owners then, today that has doubled August 2026, still small numbers. AXT was up roughly 5,000% in a year, sitting at what I called a hidden bottleneck in AI infrastructure.

That piece “held up”: the stock kept running to an all-time high above $143 later that month, however it now sitting at 50% of ATH. The business itself has since delivered a genuinely stunning Q2, with record revenue of $47.6 million, up 164% year-over-year, gross margin more than doubling to 45%, and indium phosphide revenue alone hitting a record $30.7 million.

Needham upgraded the stock to Buy. Management now guides for InP revenue to triple by the end of this year and reach $130 million a quarter by the end of 2027.

But there's things that needs clarification and more depth. Not just describing AXT as sitting downstream in the supply chain, supplying substrate to companies like Lumentum to grow their lasers on. This is still correct, but with China/US relations being strained, AXT China dependency needs to be reviewed.

Background

One thing worth clearing up before going further: LAZR doesn't only hold AXT at the substrate layer. The also owns LandMark Optoelectronics, IntelliEPI, IQE, and Soitec, all frequently described as "substrate" or "materials" names in the same breath as AXT. But it's worth being precise here, because most of them aren't actually InP alternatives.

IntelliEPI is primarily a gallium arsenide specialist, not an InP grower. Soitec works in an entirely different substrate technology silicon-on-insulator, commanding the vast majority of global supply which is super important for Silicon Photonics (SiPho). But that is on the Silicon side of the ecosystem not the laser side. Of the four, only LandMark Optoelectronics and IQE actually produce InP alongside AXT (and Sumitomo and JX, more about them later).

So while $LAZR does hold real diversification at the broader "compound semiconductor substrate" layer (see pictures above from SemiAnalysis). Top 3 is making +90% of the substrates.

This information show that th InP bottleneck is a much narrower club than the ETFS substrate-layer holdings might suggest at a glance. I will also explain, why.

Where AXT Actually Makes

Lumentum: "The Second Act"

Anders Storm — Sat, 15 Aug 2026 08:26:34 GMT

Introduction

Lumentum is the poster child of this whole optical boom. It feels like every photonics newsletter, every sell-side note, every X thread on AI infrastructure has already written the Lumentum piece the sold-out lasers, the Nvidia stake, the stock chart that looks like a hockey stick. I'm not going to write that piece again. Instead, I want to look at how Lumentum actually got to this position, because once you see the path, the current numbers make a lot more sense than they do as a standalone headline.

And the numbers, to be clear, are extraordinary. Just this week, Lumentum reported fiscal Q4 2026 revenue of $1.01 billion, the first time the company has crossed a billion dollars in quarterly net sales, up 109% year-over-year, with non-GAAP gross margin crossing 50% a full quarter ahead of management's own timeline. There is no serious debate left about whether Lumentum is delivering on the expectations that have been piled onto it. It clearly is. That's exactly why it's the natural, obvious #1 holding in the $LAZR ETF, not because it's the most exciting story to tell, but because the numbers have simply earned the position. What I'm more interested in, is how a company gets to be this unambiguously the leader in a market this hot. It's a 45-year-old photonics lineage that has already been through one spectacular boom and bust on a single giant customer, an insatiable order book, and then the whiplash when that customer's needs shifted. Understanding that first act is what makes the second one, the one playing out right now, making more sense.

Act One: Two lens grinders and a laser company

Lumentum's lineage traces back to two separate pioneering photonics businesses: Uniphase Corp, founded in 1979, and JDS Fitel, founded in 1981 in Canada. Uniphase built its name in commercial lasers and optical transmission equipment; JDS Fitel pioneered components for fiber-optic networking. The two merged in 1999 to form JDS Uniphase JDSU, under founder Jozef Straus, who built an engineering-first culture around what the company called a "design-in" strategy: get your components embedded deep into a customer's own engineering roadmap, and you become very hard to displace (sounds familiar?).

By 2014, JDSU's Communications and Commercial Optical Products division CCOP, was doing roughly $800 million a year, sitting inside a larger company that also ran a test-and-measurement business and a specialty pigments and holographic-labels unit. In August 2015, JDSU split itself in two. The test-and-measurement and specialty-materials businesses kept the JDSU name, later rebranding as Viavi Solutions. CCOP, plus JDSU's WaveReady product line, became a new standalone public company: Lumentum Holdings, trading on NASDAQ under LITE, led by Alan Lowe.

Act Two: The iPhone that made Lumentum famous

Lumentum's first real taste of hypergrowth had nothing to do with data centers. In November 2017, Apple launched the iPhone X with Face ID and the vertical-cavity surface-emitting lasers (VCSELs) powering the TrueDepth camera system. The company became Apple's dominant VCSEL supplier, at one point commanding roughly 68% of that market, with Apple alone accounting for something like a quarter of Lumentum's total revenue. Over five years, Lumentum sold more than $1.5 billion worth of these tiny laser arrays. The stock, which had traded in the twenties post-spinoff, rocketed into the mid-eighties.

Then, in November 2018, it broke the other way just as fast. Lumentum disclosed that one of its largest customers unnamed, but unmistakably Apple had asked to materially cut back shipments of laser diodes that had already been ordered for that quarter. The stock fell 28% in a single morning, and dragged down every other name in the VCSEL supply chain with it, including Apple itself.

That episode is worth sitting with, because it's the exact structural risk that might shows up again in the bear case AI story.

Act Two-and-a-Half: Buying the laser moat

Here's the part of the story that rarely gets told alongside the Apple drama, even though it's arguably more important to where Lumentum sits today. In 2018, the same year the Face ID cycle was peaking and then breaking, Lumentum quietly closed a $1.8 billion acquisition of Oclaro, a photonics company with more than three decades of laser innovation behind it. Industry commentary at the time, called it close to a merger of equals on the telecom and datacom side, because Oclaro actually had the stronger position in 400G indium phosphide technology at that moment, not Lumentum.

That deal is the real foundation of the laser business pepole are writing about. It's what gave Lumentum dominant InP wafer growth, EML fabrication, and photonic integration capability, the "Laser Moat," as one investor write-up bluntly calls it, rather than something Lumentum built organically inside JDSU. Every subsequent deal (NeoPhotonics for coherent tunable lasers, CloudLight for transceiver assembly) was built on top of the InP foundation Oclaro provided.

Oclaro itself traced back further still, to a UK company called Bookham Technology, founded by physicist and entrepreneur Dr. Andrew Rickman, reportedly Britain's first internet billionaire and company named after his hometown of Bookham, Surrey. Bookham merged with Avanex in 2009 to form Oclaro, which Lumentum then bought a decade later.

A fun, small-world fact: I actually met Dr. Rickman in person, in London around 2017 in a long meeting, in a Sivers/Rockley after our aquestion of CST Global early 2017. Probebly right around the time Lumentum was closing the Oclaro deal that would go on to define its laser business for the next decade. At the time, Rickman was running his next venture, Rockley Photonics, pursuing silicon photonics for data communications. Rockley later pivoted hard into wearable health sensors, went public via a SPAC in 2021 at a valuation north of $1 billion (with partnerships alongside Apple and Medtronic), and then collapsed just as fast, the stock fell from over $15 to under $0.20 a share, and the company filed for Chapter 11 bankruptcy in January 2023.

There's a lesson in there worth sitting with, and it isn't really about Rockley's healthcare pivot going wrong. Rockley's original idea, highly integrated silicon photonics for data communications was directionally correct. It just arrived roughly a decade before the market needed it. In 2013, when Rickman founded the company, there was no AI infrastructure buildout demanding hundreds of millions of laser chips a year. There wasn't a Nvidia writing $2 billion checks to secure laser supply. The demand curve that would have justified Rockley's original bet simply didn't exist yet, and a photonics company burning venture capital in search of a market that's still ten years out eventually has to pivot toward something that pays the bills here ans now. Healthcare wearables in Rockley's case, whether or not that pivot ever worked out.

That's the part of this story that should stay with you longer than the Oclaro deal itself: Lumentum did not create the AI infrastructure boom. It didn't invent the demand for hundreds of millions of laser chips, and it didn't drive the hyperscalers into needing optical interconnects at unprecedented scale. It simply happened to already be standing in exactly the right spot, with exactly the right decade-old acquisition already digested, when a demand wave arrived that nobody in 2207-18, not Lumentum, not Rickman, not anyone else in the industry had fully priced. The same underlying technology, the same photonics expertise, sat in two different companies. One was too early and ran out of runway waiting for its market. The other was, through some mix of luck and a decade of unglamorous integration work, still standing when the market finally showed up. Being right about the technology was necessary for both companies. It was nowhere near sufficient. Timing did the rest.

Act Three: A failed acquisition that quietly shaped today's competitive map

Here's a detail most people miss entirely. In January 2021, Lumentum agreed to acquire the laser company Coherent, Inc. for $5.7 billion in cash and stock, a deal meant to diversify Lumentum beyond telecom and consumer optics into industrial and scientific lasers. It didn't stay uncontested for long. A three-way bidding war broke out between Lumentum, II-VI Incorporated, and MKS Instruments, and by late March 2021 the price had been bid up past $7 billion. Lumentum made a final offer of $230 in cash plus 0.67 shares of Lumentum stock per Coherent share. Coherent's board chose II-VI instead, a decision Lumentum publicly called inferior on pure value terms and Lumentum collected a $217.6 million termination fee for its trouble.

II-VI completed that acquisition later in 2021 and then did something clever with the spoils: it renamed the combined company Coherent Corp, adopting the more recognizable brand of the business it had just bought. The Coherent that sits in LAZR's portfolio today, and that I will write about later, the one Lumentum tried and failed to buy.

Having lost Coherent, Lumentum pivoted to a smaller, more targeted deal: it acquired NeoPhotonics, a photonics component maker focused specifically on cloud computing and optical networking, for roughly $900 million, completing in 2022. Rather than diversifying broadly into industrial lasers the way the Coherent deal would have, Lumentum doubled down on exactly the optical networking and coherent transmission technology that now sits at the center of the AI data center buildout. In hindsight, losing Coherent may have forced Lumentum onto the more focused path that positioned it well for what came next.

Act Four: New leadership, right as the AI wave arrived

Alan Lowe, who had run the CCOP business since before the spinoff and then led Lumentum as an independent company for its first decade, stepped down as CEO in February 2025. His successor, Michael Hurlston, came in with a resume built almost entirely inside the optical components world: CEO of Finisar (another major optical/VCSEL player, later acquired by II-VI) through 2019, then CEO of Synaptics, before returning to photonics to lead Lumentum. The timing of that handoff early 2025, just as hyperscaler AI infrastructure spending was accelerating into the current supercycle, turned out to be close to perfect.

Act Five: The current claim to fame

Which brings us to now. Lumentum's fiscal Q4 2026 revenue hit $1.01 billion — the first billion-dollar quarter in the company's history — up 109% year-over-year, with non-GAAP EPS of $3.23 against a $2.97 consensus, and non-GAAP gross margin at 50.4%, a threshold management hadn't expected to hit until a $2 billion quarterly run rate. This was the eighth consecutive quarter of sequential growth, the third straight quarter with sequential growth above 20%. Guidance for the current quarter (fiscal Q1 2027) sits at $1.225-1.275 billion, implying year-over-year growth above 130%. Management says the company is sold out of key components through 2028.

The demand driving that isn't diversified consumer electronics anymore — it's concentrated, structural, and coming from the handful of companies building AI infrastructure. Hurlston himself has framed the scale shift bluntly: telecom customers used to order lasers in the hundreds; hyperscalers are now asking for hundreds of millions. He's called the resulting indium phosphide shortage potentially worse than the memory crunch that's rattled the rest of the semiconductor industry this year.

Unlike the Apple VCSEL episode, where Lumentum was a supplier at the mercy of one customer's changing order pattern, this time the largest customer has taken a direct stake in solving the supply problem: Nvidia has invested $2 billion in Lumentum equity, the same playbook it used with TSMC, specifically to help fund and de-risk the capacity expansion. That's a structurally different relationship than "we hope Apple keeps ordering."

By independent market-share data (Nomura, sourcing Yuanjie Technology's own public offering document), Lumentum is now the largest laser chip supplier globally, at roughly 17% share, ahead of Broadcom, and ahead of longtime Japanese photonics incumbents Mitsubishi Electric and Sumitomo Electric. The company runs five InP fabs and is still shipping more than 30% below what customers want; its next US facility, in Greensboro, isn't expected online until early 2028.

The stock reflects all of this: up dramatically over the trailing year, with market cap surging past $75 billion in the days following the Q4 print — now a member of both the S&P 500 and the Nasdaq-100 — a company that spent much of 2022 and 2023 as a mid-cap photonics name now sitting among the more valuable companies in the US market. Worth flagging plainly: a stock moving this fast means any specific market cap figure is stale within days, if not hours — treat the number as a snapshot of momentum, not a fixed data point.

Why the history actually matters for the thesis

The bear case on Lumentum today is almost entirely a valuation argument, how much of the next several years of growth is already priced into a stock up more than sixfold in a year. Nobody serious argues the underlying demand isn't real.

But the company's own history is what should calibrate how you think about the risk that matters most: customer concentration. Lumentum has been here before, at smaller scale, with Apple. The difference this time is structural, not just larger. Apple never took an equity stake in Lumentum to keep VCSEL supply flowing — it just placed orders, and then un-placed them. Nvidia, by contrast, has put $2 billion of its own balance sheet directly into ensuring this supply chain doesn't break. That's a meaningfully different arrangement than the one that ended in a 28% single-day drop in 2018, even though the surface-level pattern — one customer category driving an enormous share of growth — looks similar on paper.

The company that lived through the first version of this story is now running the second one with a lot more institutional memory, a much larger and more diversified customer base within AI infrastructure itself, and a customer who's financially committed to the outcome. That's not a guarantee. But it's a genuinely different setup than the one that preceded it.

Worth remembering, too, that none of this makes Lumentum a visionary in the way the stock chart might suggest. It didn't foresee the AI buildout any better than Rockley Photonics did when it made almost the identical technology bet a decade too early. Lumentum's real achievement was more modest and, in some ways, more valuable: it kept its InP capability alive and integrated through years when nobody was paying up for it, so that when the demand curve finally arrived, it was already standing in the right place. That's not nothing. But it's a different story than "Lumentum built the AI boom." Lumentum is a supplier that happened to be positioned exactly right when someone else's boom needed exactly what it had spent a decade quietly building.

How long does this actually run?

That question, how long is really the only one that matters from here, and it's worth answering with something more durable than a stock chart. BloombergNEF's own data center demand forecast puts US data center power demand at roughly 194 gigawatts by 2035, up from a small fraction of that today, with the growth curve only steepening past 2025 rather than leveling off. That's not a photonics forecast or a Lumentum forecast. It's a physical infrastructure build-out, measured in power plants and grid interconnects, on a timeline that stretches a full decade out.

I read the chart as the actual duration of this cycle, more than any single quarter's guidance. We're in an AI supercycle, and the data center buildout underneath it looks like at least a five-year growth runway from here, arguably longer given where that curve is still headed by 2035. That doesn't mean a straight line, there will be air pockets, digestion quarters, and probably several sharp drawdown along the way, the same way there was in late July of this year when AI infrastructure sentiment wobbled hard for a few weeks. Lumentum's own history, the Apple cycle in particular, is a reminder that concentrated demand curves don't move in a straight line even when the underlying trend is real.

But the mechanism that matters here is simple: as long as AI infrastructure keeps delivering on its own promises, as long as hyperscalers keep needing more compute, more bandwidth, more optical interconnect to tie it all together, Lumentum sits in almost exactly the spot it needs to sit in to keep benefiting from it. That's the whole bet, stated as plainly as I can state it. Not that Lumentum is a visionary, not that the stock can't have a bad quarter or a bad year along the way. Just that the multi-year infrastructure build-out underneath this entire thesis is still, by the best available forecasts, in its early-to-middle innings and Lumentum, for reasons that go back to a $1.8 billion acquisition in 2018 and a decade of unglamorous integration work before that, is one of the best-positioned companies to keep riding it.

*Positions and disclosures: [insert your standard disclosure language here]. This is not investment advice.*

Introduction

And the numbers, to be clear, are extraordinary. Just last week, Lumentum reported fiscal Q4 2026 revenue of $1.01 billion, the first time the company has crossed a billion dollars in quarterly net sales, up 109% year-over-year, with non-GAAP gross margin crossing 50% a full quarter ahead of management's own timeline. There is no serious debate left about whether Lumentum is delivering on the expectations that have been piled onto it. It clearly is. That's exactly why it's the natural, obvious #1 holding in the LAZR ETF, not because it's the most exciting story to tell, but because the numbers have simply earned the position. What I'm more interested in is how a company gets to be this unambiguously the leader in a market this hot. It's a 45-year-old photonics lineage that has already been through one spectacular boom and bust on almost exactly this pattern a single giant customer, an insatiable order book, and then the whiplash when that customer's needs shifted. Understanding that first act is what makes the second one the one playing out right now make sense.

Act One: Two lens grinders and a laser company

Act Two: The iPhone that made Lumentum famous

Lumentum's first real taste of hypergrowth had nothing to do with data centers. In November 2017, Apple launched the iPhone X with Face ID — and the vertical-cavity surface-emitting lasers (VCSELs) powering that TrueDepth camera system came overwhelmingly from Lumentum. The company became Apple's dominant VCSEL supplier, at one point commanding roughly 68% of that market, with Apple alone accounting for something like a quarter of Lumentum's total revenue. Over five years, Lumentum sold more than $1.5 billion worth of these tiny laser arrays. The stock, which had traded in the twenties post-spinoff, rocketed into the mid-eighties.

That episode is worth sitting with, because it's the exact structural risk that shows up again in the AI story: a single category of customer demand growing so fast that a component supplier reorganizes its whole cost structure around it and then finding out how fast that can reverse when the customer's own product cycle shifts.

Act Two-and-a-Half: Buying the laser moat

Here's the part of the story that rarely gets told alongside the Apple drama, even though it's arguably more important to where Lumentum sits today. In 2018, the same year the Face ID cycle was peaking and then breaking, Lumentum quietly closed a $1.8 billion acquisition of Oclaro, a photonics company with more than three decades of laser innovation behind it. Industry commentary at the time called it close to a merger of equals on the telecom and datacom side, because Oclaro actually had the stronger position in 400G indium phosphide technology at that moment, not Lumentum.

That deal is the real foundation of the laser business we've been writing about. It's what gave Lumentum dominant InP wafer growth, EML fabrication, and photonic integration capability, the "Laser Moat," as one investor write-up bluntly calls it, rather than something Lumentum built organically inside JDSU. Every subsequent deal (NeoPhotonics for coherent tunable lasers, CloudLight for transceiver assembly) was built on top of the InP foundation Oclaro provided.

Oclaro itself traced back further still, to a UK company called Bookham Technology, founded by physicist and entrepreneur Dr. Andrew Rickman, reportedly Britain's first internet billionaire and named after his hometown of Bookham, Surrey. Bookham merged with Avanex in 2009 to form Oclaro, which Lumentum then bought a decade later.

A fun, small-world fact: I actually met Dr. Rickman in person, in London around 2017, in a Sivers/Rockley meeting after our aquestion of CST Global early 2017. Probebly right around the time Lumentum was closing the Oclaro deal that would go on to define its laser business for the next decade. At the time, Rickman was running his next venture, Rockley Photonics, pursuing silicon photonics for data communications. Rockley later pivoted hard into wearable health sensors, went public via a SPAC in 2021 at a valuation north of $1 billion (with partnerships alongside Apple and Medtronic), and then collapsed just as fast, the stock fell from over $15 to under $0.20 a share, and the company filed for Chapter 11 bankruptcy in January 2023.

Act Three: A failed acquisition that quietly shaped today's competitive map

Act Four: New leadership, right as the AI wave arrived

Act Five: The current claim to fame

Why the history actually matters for the thesis

How long does this actually run?

Anders Storm

Disclosure: The author is Advisor to Tema ETFs $LAZR Photonics and Optical EFT and holds positions in stocks discussed in this publication, including Lumentum. Position sizes and timing are not disclosed. This analysis is for informational purposes only and does not constitute investment advice. Readers should conduct their own due diligence and consult licensed advisors for personal investment decisions.

The Hidden Facts in Smart Eye’s Design Wins Chart - Proposing a Rule of 40 valuation for the future

Anders Storm — Tue, 11 Aug 2026 20:21:34 GMT

Prologue: Why a Wireless and Optical Tech Guy Is Writing About $SEYE

I spend most of my professional life thinking about technology and products. As former CEO of Sivers Semiconductors and now leading Dirac Research, I have spent nearly a decade in the wireless and optical semiconductor. Over the last 2 years I entered into automotive. So far my analytical work on Substack and X has focused primarily on photonics, mmWave semiconductors, and the AI infrastructure buildout. Smart Eye is neither. So why am I writing about them?

I have followed Smart Eye in parallel with Sivers Semiconductors for years on the Swedish small-cap market, and the structural similarities between the two companies are difficult to ignore. In 2016 SEYE Net Sales was SEK 40m (SIVE app SEK 20m). In 2025 both approx. 10-12x Net Sales, done some buy side M&As and are now about to enter their inflection points.

Both went public (Nasdaq First North 2016/2017) years before their addressable markets fully materialized. Both raised capital repeatedly through share issues to fund the gap between contract wins and revenue realization. Both accumulated design wins from major customers that took far longer than expected to convert into recognized revenue. Both had been characterized by financial media as companies that can not execute. Both are now in the early phase of what appears to be a mechanical revenue ramp from already-signed contracts, with inflection catalysts accelerating revenue in the coming years.

Building Sivers in parallel 2016-2024 has been instructive for how I read Smart Eyes position today. The pattern of skepticism, delayed inflection, and then starting to see the acceleration is quite exacting for two so different but yet “Sister Compnies” for many years at the same First North list.

My current role at Dirac Research adds a second lens. Dirac operates in the automotive supply chain as a software provider to OEMs and Tier 1 suppliers. I now spend meaningful time in the same industrial ecosystem where Smart Eye competes. I have direct operational familiarity with automotive design win cycles, the mechanics of Tier 1 supplier relationships, royalty-per-vehicle contract structures, and the multi-year lead times between contract award and volume production. These are not abstract concepts to me, they are the operational realities I encounter in Dirac’s own commercial pipeline.

This dual perspective matters for how I read Smart Eye. When Smart Eye reports 125 design wins in production, I understand what that number represents in terms of OEM commitment, integration work, and locked-in future revenue. When management guides toward 200-250 wins in production by year-end 2026, I can assess whether that pace is plausible against automotive industry norms. When the company reports a Tier 1 software supplier deal with a Japanese OEM, I recognize the structural significance of that positioning.

I am not a Smart Eye insider. I have no unique access to their commercial pipeline or customer relationships. But the analysis that follows draws on operational familiarity with automotive design win economics that most financial analysts covering this stock do not have, and on the pattern recognition of watching a structurally similar Swedish small-cap company go through the same inflection cycle in parallel.

The company I have led as CEO and the company I now lead both operate in industries where design wins create backlog long before they create revenue, where regulatory catalysts drive multi-year adoption curves, and where the gap between contract and cash flow is often mistaken for weakness rather than pipeline. That is exactly the situation I see in Smart Eye today.

What follows is my read of the data.

Stock Information (11 August 2026):

Ticker: SEYE.ST (First North Stockholm) / SMTEF (OTC Pink)

Share Price: SEK 97.95

Shares Outstanding: 39,198,762

Market Cap: SEK 3.84 billion (~$365M USD)

P/S: 8.61x

Gross Margin: 87.7%

Q1 2026 Revenue: SEK 126.5M (+40.4% YoY)

Q2 2026 Report: 26 August 2026 (15 days away)

EU GSR full activation: 7 July 2026 (active for 5 weeks)

About Smart Eye: Six Years of Frustration, Now Convergence

Smart Eye AB is a Swedish software company founded in 1999 and headquartered in Gothenburg. The company builds AI-powered systems that analyze human attention, eye movement, and behavior, initially for academic research, later for the automotive industry, and increasingly for adjacent verticals including impairment detection and commercial fleet safety.

The company has approximately 270 employees and operates across three main business areas: Automotive Solutions (driver monitoring, interior sensing, authentication, intoxication detection), Behavioral Research (eye tracking instruments for academic and commercial research, including the iMotions software platform), and a smaller but rapidly growing Aftermarket Intelligent Systems business serving commercial fleets.

To understand today’s situation, it helps to understand the history. Smart Eye went public on Nasdaq Stockholm in 2016. Between 2018 and 2022, the company won hundreds of design wins from major automotive OEMs, anticipating that the EU General Safety Regulation and Euro NCAP rating changes would create a structural market for DMS technology by mid-decade. Analysts including Redeye published bullish base case targets of SEK 155-160. Affärsvärlden, in early 2024, projected SEK 1 billion in revenue for 2026. Both proved too optimistic on timing.

What followed was four years of frustration. Chip shortages delayed production starts. COVID disrupted automotive supply chains. The transition to software-defined vehicles created integration complexity OEMs had not anticipated. Design wins that should have entered production in 2022 where pushed to 2024 or 2025. The market increasingly characterized Smart Eye as a company that won contracts but could not convert them to revenue and the share price reflected that skepticism, falling from peaks above SEK 250 in 2020-2021 to lows below SEK 50 in 2024.

The competitive landscape during this period also evolved. Seeing Machines, an Australian peer, accumulated more vehicles on the road through earlier-stage deployments. Tobii acquired a US DMS business to compete more directly. Mobileye signaled intent to integrate DMS into its EyeQ system-on-chip platform, threatening to bundle the functionality at scale. Cipia targeted the low-cost segment with Israeli engineering and SoC partnerships. Smart Eye’s premium positioning came under question.

What changed in Q1 2026 is that the production ramp finally arrived. Automotive Solutions grew 122% organically year-over-year. Royalty revenue specifically grew over 200%. EBITDA improved by SEK 44.8 million. The company guided for 200-250 vehicle models in production by year-end 2026, double the count at year-end 2025. And critically, this acceleration occurred before the EU GSR mandate took full effect on July 7, 2026.

We are now five weeks into EU GSR implementation, and the Q2 2026 report is 15 days away. Independent sell-side data from Redeye and Seeing Machines’ own Q4 FY2026 report published earlier today have both validated the acceleration thesis. The rest of this piece walks through the framework for interpreting what the Q2 report will reveal.

There is a chart in Smart Eye’s 2025 Annual Report that contains more information than any analyst report I have read on the company. It is buried on page 13. It shows the number of design wins in production by year, with a single forward-looking estimate for 2026.

Design Wins in Production chart from Smart Eye Annual Report 2025:

The chart says this: Smart Eye expects 200-250 design wins in production by the end of 2026, up from 125 at year-end 2025. That is a doubling, in one year, of the number of vehicle programs generating royalty revenue.

This is not an analyst projection. It is management’s own guidance, published in an official annual report filed with shareholders. And it is the central piece of evidence in what I believe is one of the most clearly mispriced Swedish small-cap stories in the market today.

This piece walks through five facts that the design wins chart reveals, facts that are individually visible to anyone reading the annual report, but collectively underappreciated by the analyst community covering the stock. The stock has moved from SEK 89.7 at Q1 disclosure to SEK 97.96 today, but the Q2 report on August 26 will be the next major test. This means readers discovering this analysis now have approximately two weeks before the market receives fresh quantitative validation or challenge to the cohort thesis.

My core argument of this piece is that Smart Eye is transitioning into a Rule of 40+ software business, and Rule of 40+ software businesses trade at premium multiples. The market has not yet fully priced this transition.

Fact One: The Production Acceleration Is Convex, Not Linear

Design wins in production at year-end: 250

This is not steady growth. It is acceleration. The annual addition is growing both in absolute terms and as a percentage of the prior year base. Wins added in 2024 (+35) were 75% more than 2023. Wins added in 2025 (+50) were 43% more than 2024. Wins guided for 2026 (+75-125) represent 50-150% more than 2025.

This pattern is characteristic of an S-curve transition the period when accumulated contract wins begin materializing in production at a pace that compounds. In automotive software, design cycles run 5-7 years from nomination to production start. Smart Eye won the bulk of its current production wins between 2019 and 2022. The acceleration we see in 2024-2026 is the result of contract wins that were already documented four to seven years ago.

The wave hence is not speculative. It is scheduled.

Fact Two: The Accumulated Order Value Is SEK 14.2 Billion

The single most important data point in the entire annual report appears in a small chart on page 7, easily missed:

Estimerat ordervärde från Design Wins chart from Smart Eye Annual Report 2025;

Compare this to 2025 revenue of SEK 404 million.

Smart Eye has an accumulated design win order value equivalent to 35x annual revenue. This is not speculation about future contracts, it is already contracted order value to be realized over each vehicle model’s production lifecycle, typically seven to ten years.

With a conservative seven-year average lifecycle:

Confirmed value of SEK 8,725M ÷ 7 = ~SEK 1,246M/year at full ramp

Potential value of SEK 5,505M ÷ 7 = ~SEK 786M/year at full ramp

Total Automotive run-rate at full backlog realization: ~SEK 2,032M/year

This figure includes only wins already secured. Any new wins won between now and 2030 add on top of this baseline.

Beyond DMS: The Non-Automotive Business Lines That Are Often Overlooked

Most analyses of Smart Eye focus on automotive driver monitoring because that is where the dominant revenue acceleration comes from. But the company has three additional business lines that collectively represented approximately 55% of 2025 revenue and provide meaningful diversification plus optionality for 2030.

Behavioral Research (~52% of 2025 revenue):

This is Smart Eye’s original business, dating to the early 2000s. The company sells eye-tracking instruments and software to academic researchers, marketing firms, neuroscience labs, and government agencies. Customers include Harvard, NASA, Boeing, Airbus, and over 1,300 research institutions globally. Through the iMotions software platform, acquired in 2021, the business expanded into multimodal human behavior research — combining eye tracking, facial expression analysis, biometric sensors, and emotional state recognition.

This segment generated SEK 208.7 million in 2025, down slightly from SEK 220.6 million in 2024, essentially flat with currency headwinds masking modest underlying activity. Redeye estimates SEK 203M for 2026 (-3% y/y).

For modeling purposes, I assume Behavioral Research grows at 3-5% annually through 2030 — reaching approximately SEK 250 million.

Aftermarket Intelligent Systems (AIS):

This segment supplies retrofit DMS systems for commercial vehicles — trucks, buses, mining equipment, and fleet vehicles that were not equipped with DMS at manufacture. The business operates a different model from the automotive royalty business: hardware-plus-software sales rather than per-vehicle royalties, with margins in the 50-55% range rather than 85-95%.

The AIS segment had two important developments in 2025-2026. First, an order from Spanish company Optix in October 2025 to retrofit 4,000 commercial vehicles in Europe — a meaningful proof point that fleet operators are voluntarily adopting DMS ahead of regulatory mandates. Second, expansion into the heavy commercial vehicle segment where EU GSR mandates DMS for new trucks and buses sold in Europe from July 2026 onward — now in effect. The aftermarket opportunity to retrofit existing fleets is substantial — approximately 6 million trucks and buses operate in Europe that could potentially be retrofitted over the next decade.

Sightic Analytics (acquired February 2026):

Sightic is the smallest but potentially most strategically important addition. The company, acquired for SEK 60.5 million in February 2026, specializes in alcohol and drug impairment detection using camera-based behavioral analysis. Sightic had already secured initial design wins with Japanese OEMs for alcohol detection systems with a combined lifetime value of approximately SEK 200 million.

Post-acquisition, Smart Eye signed a multi-year contract with a European police agency for drug impairment detection systems used in roadside testing — a different customer segment from automotive but using similar underlying technology.

The strategic logic is that impairment detection is becoming the next regulatory frontier after attention monitoring. The US has been debating federal mandates for alcohol detection in new vehicles for several years. If any of these materialize, Sightic positions Smart Eye to capture a share without significant additional development cost.

For modeling purposes, my base case assumes Sightic plus related impairment detection revenue reaches SEK 100-200 million by 2030. This is the segment with the widest range of potential outcomes.

The combined non-DMS picture:

These three segments collectively account for approximately SEK 220 million of 2025 revenue. By 2030, they may contribute SEK 500-650 million combined, against SEK 2.0 billion in automotive royalty. The proportion changes dramatically as automotive scales, but the absolute contribution roughly doubles.

This matters for two reasons. First, it provides downside protection, if automotive DMS materializes more slowly than the cohort model suggests, the non-DMS business lines still provide a meaningful revenue base. Second, it provides upside optionality — if Sightic or impairment detection regulation accelerates, the non-DMS contribution could exceed base case estimates substantially.

The thesis is not “Smart Eye is purely a DMS bet.” It is “Smart Eye is a human insight AI platform with DMS as the largest and most predictable growth driver, plus several adjacent revenue streams that provide diversification and optionality.”

The Moat: Why Smart Eye Keeps Winning the Design Wins

Before extrapolating revenue, the underlying question must be addressed: in a market with multiple competent competitors: Mobileye, Cipia, Seeing Machines, Tobii, plus in-house OEM development teams, why does Smart Eye continue accumulating design wins?

The honest answer is that the moat is not a single feature. It is a combination of factors that compound over time, and each on its own would not be sufficient. Together, they create a structurally favorable position that is difficult to displace.

The hardware-agnostic software layer position. The software runs on NVIDIA DRIVE, Qualcomm Snapdragon Cockpit, Texas Instruments, Ambarella, and others. This matters because the modern automotive industry is consolidating around software-defined vehicles where the OEM chooses the SoC late in the design cycle. An OEM that picks NVIDIA today and Qualcomm tomorrow for a different platform can use Smart Eye in both. A vendor locked to one SoC ecosystem cannot.

A decade of production-grade edge cases. Computer vision software for DMS sounds straightforward until you encounter the actual edge cases that production deployment requires. Sunglasses with reflective coatings. Drivers with eye conditions. Lighting transitions from tunnels to bright sun. Cultural variation in head movement patterns. Camera placement variations between vehicle models. They are operational realities that automotive QA processes test against, and that disqualify systems which have not encountered them at production volume. Smart Eye has been deploying DMS software in serial production since 2018 over six years of accumulated edge case learning that new competitor entering today cannot match for several years.

The switching cost that compounds each production year. Once an OEM has Smart Eye software in serial production on a vehicle platform, the cost of switching is substantial. The platform-level integration work, the safety certification process, the validation testing across edge cases, the supplier qualification documentation, all of this must be redone with a new vendor. For a vehicle platform with 5-7 years of production ahead of it, switching makes no economic sense unless the incumbent fails catastrophically. The company reports only 12 exited wins out of 372 over recent years — a 3% attrition rate. The backlog is not just a sales pipeline; it is a structural commitment locked in by switching costs.

The Tier 1 software supplier position. Traditional automotive supply chains place software vendors as Tier 2 or Tier 3 suppliers, they sell software to Tier 1 hardware vendors (Bosch, Continental, Aptiv) who integrate and sell complete systems to OEMs. Smart Eye announced its first Tier 1 software supplier deal with a major Japanese OEM in 2025, valued at SEK 200 million for two vehicle models. This is structurally different from traditional positioning. As a software-only Tier 1, Smart Eye captures more of the value chain, has direct relationships with OEM engineering teams, and is involved earlier in vehicle development cycles.

The major competition

Seeing Machines is the most direct peer and a genuinely strong competitor. Their Q4 FY2026 report published this morning (August 11, 2026) shows 2.11M vehicles produced in calendar Q2 2026 (+64% q/q, +333% y/y), full-year FY2026 production of 4.49M vehicles (+195% y/y), and automotive royalty revenue of USD 33.9M (+135% y/y). Cars on road with Seeing Machines technology rose 120% y/y to 8.2M. This is genuinely strong data and validates the DMS market thesis from a competitor’s perspective.

Redeye’s read on the competitive positioning is worth quoting directly: “We judge Smart Eye to be tracking just a notch behind Seeing Machines in cars produced per quarter, but will likely overtake Seeing Machines as it has been more successful in securing design wins outside of the EU that will enter production during 2027, for example with Toyota.”

This is significant, projecting that Smart Eye will surpass Seeing Machines in production volume during 2027 — driven by non-EU design wins including Toyota. The moat analysis has argued that Smart Eye’s Tier 1 software supplier positioning with Japanese OEMs would prove structurally valuable. Redeye’s commentary supports this thesis with a specific customer name.

Mobileye is a stronger structural threat than is sometimes acknowledged. Their EyeQ6L SoC integrates DMS directly with ADAS perception in a single chip, eliminating the need for a separate DMS ECU. For OEMs prioritizing bill-of-materials cost, this is attractive. Mobileye recently won a major US OEM program expected to span millions of vehicles starting in 2027. The mitigating factor for Smart Eye is that Mobileye competes on architecture and bundling rather than on best-in-class DMS algorithms, and their threat is to new wins from 2027 onward rather than to Smart Eye’s existing backlog already locked into Tier 1 contracts.

In-house OEM development is a structural long-term risk for all external DMS vendors. Which I am not going to spend time on now.

Per-Vehicle Pricing and growth

The per-vehicle royalty assumption is critical to any revenue model for Smart Eye. Redeye’s analysis published this morning provides specific data points:

Smart Eye estimated ASP: SEK 42 per vehicle (Redeye characterizes their own estimate as “slightly on the conservative side”)

Seeing Machines H1 2026 ASP: SEK 70 per vehicle (down from SEK 79 in H2 2025)

Seeing Machines incremental royalties added H1 2026: SEK 65 per vehicle

The gap between Smart Eye (SEK 42) and Seeing Machines (SEK 65-70) is real and requires explanation.

I think that the the most likely explanation is volume-based pricing, which is industry standard in automotive software royalty contracts. An OEM guaranteeing 500,000+ vehicles per year over a seven-year program negotiates lower per-vehicle pricing than an OEM guaranteeing 50,000 vehicles per year. The total contract value scales with volume; the per-unit price scales inversely.

Smart Eye and Seeing Machines have pursued different commercial strategies that produce comparable total economics in different forms:

Smart Eye has won contracts with high-volume OEMs. The blended per-vehicle royalty is lower, but total program value is higher.

The implication for valuation is that volume scaling, not price scaling, is the primary revenue driver through 2030. Investors expecting per-vehicle pricing to converge with Seeing Machines should not expect that, the difference reflects portfolio mix, not pricing power weakness.’

Year-over-year quarterly growth has progressed from +18% in Q1 2025 to +110% in Q1 2026 and Redeye’s revised Q2 estimate implies +194% acceleration. This is not steady growth. This is a step change.

Critically, the acceleration through Q2 2026 occurred before the EU General Safety Regulation mandate took full effect on 7 July 2026. Redeye’s own analysis: “Q3 2026 looks to be the first quarter with full EU penetration” — meaning the largest catalyst is still ahead.

The Q1 2026 report disclosed that royalty revenue specifically grew over 200% year-over-year. Automotive Solutions organic growth reached 122% in the quarter. Redeye’s revised Q2 estimate of SEK 122.1M represents +49% sequential growth from Q1’s SEK 82.1M, an extraordinary quarterly ramp for any company at this stage.

The historic pattern of delayed inflection points has clearly broken. The royalty engine is operational, the production curve is steepening, and the regulatory tailwind has activated.

Royalty per vehicle produced: Recurring revenue over the production lifecycle. Gross margin approximately 85-95%.

The royalty share of total Automotive revenue is increasing rapidly. The Q2 2026 EBITDA-CAPEX transition to positive territory, Redeye’s estimate is SEK 19.9M at 12% margin is a structurally significant milestone. This is the first quarter in company history where the business generates cash after all investment costs. This transition typically triggers meaningful multiple expansion in growth software businesses.

EBIT margin trajectory:

2025: -40.6%

2026e (Redeye): 37% EBITDA margin

2027e (extrapolated): 42% EBIT margin

2028e (extrapolated): 44-46% EBIT margin

2029e (extrapolated): 45-48% EBIT margin

2030e (extrapolated): 45-48% EBIT margin

These margins are not aspirational. They are characteristic of software businesses at maturity when royalty represents a clear majority of revenue.

Assumptions:

Why I am bullish the AI super cycle: Google just posted its first negative free cash flow quarter in 22 years. Here's why that's the wrong number to panic about

Anders Storm — Sat, 25 Jul 2026 10:11:39 GMT

Alphabet's Q2 2026 print gave the market exactly the headline it wanted to be scared by: negative free cash flow, first time since the 2004 IPO. Cue the "AI bubble" posts. Cue the "even Google can't afford this" takes.

Here's the bear case, stated as strongly as it deserves to be stated, before I take it apart.

The bear case, in full

Capex doubled year over year to $44.9 billion in a single quarter — more than double what Alphabet spent in “any single quarter” over the past two years. Management didn't hold guidance steady; they raised it for the second time this year, to $195–205 billion for 2026, with 2027 flagged to go higher still. Alphabet is now renting Nvidia chips from SpaceX at roughly $920 million a month just to cover a supply gap. It has taken on close to $100 billion in new debt this year, plus a large equity raise explicitly earmarked for AI infrastructure. A company that has never once needed outside capital to fund its own growth is now doing exactly that. If the returns on this spend don't show up, this is how the AI capex cycle ends — good businesses bleeding cash to keep pace in a race none of them can afford to lose or leave.

That's a coherent case. It is not the case I hold.

What the bears are pricing off the wrong line

Free cash flow answers one question: is the company generating more cash than it's spending, “right now, this quarter”. It's the correct number for a mature, low-capex business. It's close to the wrong number for a company mid-buildout of an asset base that will earn for the next 5-10 years.

Operating cash flow answers a different, better question for this moment: is the underlying business throwing off more cash than it used to. Alphabet's answer is unambiguous — OCF grew 41% year over year to $39.1 billion. That is not a company running out of gas. That is the core engine getting stronger while a separate, temporary line item (capex) grows even faster on top of it.

Split those two apart and the picture changes completely:

- Revenue: $119.8B, +24% YoY, one of the strongest quarters Alphabet has ever posted by conventional income-statement measures.

- Operating margin: 34%.

- Google Cloud: +82% YoY to $24.8B, backlog at $514B.

- Operating cash flow: $39.1B, +41% YoY.

- Capex: $44.9B, roughly double a year ago.

- Quarterly FCF: -$5.9B. Trailing-twelve-month FCF: still +$53B.

Negative FCF here isn't the business weakening. It's the business strengthening faster than management can build capacity for it, and choosing to fund the gap externally rather than throttle growth to protect a single quarter's cash flow statement.

The important number that actually tells you if this is fine

It isn't the FCF sign. It's the relationship between OCF growth and capex growth, and what's funding the difference.

Debt and equity raises to fund AI infrastructure are a red flag when they're propping up a business whose operating cash generation is flat or shrinking. They are a rational, even conservative, capital allocation choice when as here, the operating engine is growing faster than it ever has, and the capital is chasing demand that's already contracted (a $514B Cloud backlog is not speculative capacity; it's revenue you can already see coming).

Alphabet also isn't spending blind. It's leaning on CoreWeave and Nebius as a bridge while its own data centers come online, a tell that this is a supply-constrained sprint, not a "spend and pray" strategy. And it's sitting on roughly $240 billion in cash and marketable securities. This is a company financing a land grab from a position of strength, not covering a shortfall from a position of weakness.

The track record you're actually betting on

There's a version of this argument that doesn't need a single line item from this quarter's cash flow statement. Chamath Palihapitiya made it recently on All-In: Alphabet's average return on invested capital since its IPO is around 32%, compounded over roughly two decades. A management team with that track record spending aggressively isn't "flying fast and loose" — it's doing, at greater scale, the exact thing that produced a 32% ROIC for 20+ years. The base rate on Alphabet capital allocation is not neutral. It's about as good as it gets in public markets.

That matters for how you read a $195-205B capex number. The question isn't "is this a lot of money", it obviously is. The question is whether the team spending it has earned the benefit of the doubt. On this specific team, the answer is yes, and it's an answer built on two decades of evidence, not this quarter's guidance call.

It also reframes the "500 models" fragmentation risk that worries some observers about AI. If the ecosystem fractures into many competing model providers rather than consolidating around one or two winners, that's actually good for Alphabet on two fronts at once: TPU/silicon capacity providers and cloud infrastructure providers don't need to bet on which model wins — they get paid regardless, at both the compute layer and the hosting layer. Alphabet is positioned on both.

What would actually make me bearish

I want to be honest about what would flip this. If, over the next 2-3 quarters, OCF growth decelerates while capex guidance keeps climbing , that's the combination that says the spend has stopped tracking demand. If Cloud backlog growth stalls while capex accelerates, that's the same signal from a different angle. Neither of those has happened yet. What's happened is a lagging accounting measure (FCF) reacting to a leading operational one (capex ramp) faster than depreciation can catch up — which is mechanically guaranteed to happen in any year where capex roughly doubles, regardless of whether the underlying investment is good.

The single-quarter FCF print is the least informative number in the entire earnings release, and it's the one the headlines led with. The two numbers worth actually watching quarter over quarter are OCF growth and Cloud backlog growth. Both are still accelerating.

That's the story, I would take with me from Alphabet’s Q2 2026.

There for I am bullish on the AI SuperNova cycle and see a very long term positive trend for Hyperscaler and the AI semiconductor stack. Hyperscalers Net Income is estimated to grow.+20% CAGR to 2031.

Have a great weekend.

Anders Storm

What We Learned From Computex Part 2: The AI Compute Is Becoming one Large AI Compute Cluster

Anders Storm — Sat, 04 Jul 2026 11:17:59 GMT

For decades, the definition of a computer was simple A processor - Memory - Storage - Networking. All inside one box.

The data center was simply a warehouse filled with thousands of those boxes. But artificial intelligence is changing that. The bottleneck is no longer only the chip. The bottleneck is interconnect and the need for a optical fabric.

The next step in Scale-up is to link racks together and unlock shared compute efficiencies to avoid GPU/CPU/XPU idle time among other benefits. These AI compute clusters will soon be scaling beyond 1,000+ accelerators.

Above picture: Wiwynn rack-scale multiple rack scale-up fabric 8 compute and 2 switch nodes

To fully utilize these accelerator clusters, high bandwidth and low latency are needed. That is why next‑generation AI systems are increasingly shifting to co-packaged optics (CPO) or Optical IO. Traditional copper interconnects increasingly constrain performance, system growth, and power efficiency. At the speed these interconnects operate, copper can only reach a few meters without signal degradation. Conventional pluggable optics, found in data centers today, are too inefficient to be viable as scale-up fabrics push toward 1,000+ accelerators and ever-higher bandwidth density

The Copper Wall

Currently AI compute had been improved by adding more accelerators, racks, and networking (Scale-out). But processors became faster than the infrastructure connecting them.

Compute ↓ Memory ↓ Interconnect

Copper built modern computing, but at extreme AI bandwidth, copper becomes increasingly difficult to scale efficiently.

Marvell CEO at Computex 2026

Distance: AI wants thousands of accelerators acting together.

Density: More bandwidth means more cables and complexity.

Heat: Moving data creates thermal problems.

Everyone Expected a “Slow” Optical Transition

Copper → Linear optics → Near-package optics → Co-packaged optics → Optical I/O

If you remember the right hand box that Goldman have not yet calculated in the Optical Networking report from April 2026.

But what if optics IO is already available sooner? Was this what Wiwynn and Ayar shared at Computex 2026? Could hyperscalers move faster than expected by designing new AI campuses around optical fabrics from the beginning? I.e. for completely new data centers, can they us the L11 rack that has now been presented?

Wiwynn + Ayar Labs: Scale-up scaling beyond 1,000 accelerators

They are not building a faster cable. They are building a new AI compute optical fabric.

Old world: Server = computer. Rack = collection of computers.

Wiwynn AI compute: Large Rack clusters one large compute node. At Computex they show the future here and now.

A photonics startup demonstrating a chip is one thing. A hyperscaler manufacturer like Wiwynn demonstrating a full L11 rack system is something completely different. Wiwynn solves manufacturing, cooling, serviceability, and deployment. They don't develop this on their own initiative, they are not a rack product company, they are a contract manufacturer. That opens up for three questions:

Which Hyperscaler is behind this?
When will they start deployments?
How many External Laser Small form Factor Pluggable (ELSFP) can we expect in such a scenario?

The Optical Tray and Rack

The full rack solution featured:

32 AI compute trays
64 AI ASICs
More than 800 TB/s optical bandwidth
TeraPHY optical engines
Publicly presented system diagrams indicate up to 16 SuperNova light source assemblies per compute tray. Applied across 32 trays (L11 configuration), this implies a theoretical maximum of 512 SuperNova assemblies within the compute rack architecture.
Rack-scale optical connectivity designed for hyperscale AI infrastructure.

The full compute cluster and the ELS economics per AI compute datacenter

Let’s assume an AI scale-up cluster supporting 1,024 accelerators. 64 ASICs per rack = 16 Racks in one cluster.

16 Racks × 512 ELSFP => 8192 ELSFP

This gives high‑bandwidth, ultra‑low‑latency optical links between AI accelerators across systems and racks. This lets accelerators in different racks communicate, enabling accelerators across racks to operate as one tightly coupled AI cluster, enabling efficient scale‑up for thousands of GPUs.

The unit of AI infrastructure shifts from “server” to “1,024-accelerator optical computer.” A 100,000-accelerator AI factory could contain roughly 100 of these compute clusters. Which is approximately 820.000 ELSFP.

Connecting this to a single hyperscaler AI campus: Power: 1 GW. Accelerators: 200,000–500,000+. Optical clusters: 200–500 Ayar/Wiwynn-style domains. This would be approximately 2-5x in size.

The Breakthrough — External Laser Sources

What We Learned From Computex 2026 — And How It Connects To Goldman Sachs' Optical Networking Report (CPO Scale Up)

Anders Storm — Sat, 06 Jun 2026 12:13:56 GMT

In April 2026, Goldman Sachs published one of the most important reports yet on Optical Networking for AI infrastructure:

Optical Networking: The Next Mega Trend in AI Infrastructure.

The headline numbers immediately grabbed investors’ attention:

Optical networking TAM growing from roughly $15 billion to $154 billion by 2028.
Scale-up networking representing approximately $106 billion of that opportunity.
Co-Packaged Optics (CPO) potentially contributing roughly $91 billion.

Most discussions since then have focused on the size of the market.

But after reviewing the demonstrations presented at Computex 2026, I believe the most interesting takeaway from Goldman’s report may not be the TAM itself.

It may be the part of the roadmap Goldman explicitly acknowledges but does not yet know how to model. The below picture shows the pluggable (Scale Out) and CPO (Scale Up) for Switches and that is what the Goldman report is focusing on. However they do not mention any data about CPO for XPU (or as Ayar Labs have called it Optical I/O):

Goldman’s Three-Step Optical Roadmap

The Goldman report effectively describes three stages of AI networking evolution:

Stage 1: Pluggable Optics

This is the world we know today.

800G modules
1.6T modules
EML-based architectures
Silicon photonics gradually entering the market

This is largely an evolution of traditional networking.

Stage 2: Switch-Centric CPO

The next transition is moving optics closer to the switch ASIC.

Examples include:

NVIDIA Spectrum-X Photonics
NVIDIA Quantum-X Photonics
Broadcom CPO platforms
Marvell optical switch architectures
GlobalFoundries SCALE

This is where Goldman spends most of its effort.

The report estimates optical engine counts, external laser source (ELS) counts and rack-level dollar content because the architecture is reasonably well understood.

This is also where the famous $91 billion CPO opportunity originates.

Stage 3: Optical I/O

I find the the final box in Goldman’s roadmap quite interesting especially after Wiwynn Computex announcement of the Rack scale Co-packaged optics for AI scale-up. Placing optics next to the XPU.

Goldman labels this:

CPO with XPU

And then adds a simple note:

“Adoption schedule still unclear.”

That may be the most important sentence in the entire report.

Because Goldman is effectively saying:

We know where the industry is going.
We do not know when it arrives. However, the Wiwynn and Ayar Labs Computex demonstration suggests the industry is progressing beyond theoretical concepts and toward deployable Optical I/O architectures.

Important to note Wiwynn is an ODM (Original Design Manufacturer), it is a company that designs, develops, and manufactures products that are ultimately branded and sold by another company under that buyer's brand name. They almost never publicly announces customers by name because it’s an ODM. That’s the business model. But the industry has a very good idea who the major customers are.

The strongest publicly supported names are:

Meta Platforms
Microsoft
Amazon Web Services

Multiple sources describe Wiwynn as a major supplier to hyperscale cloud providers, especially Meta and Microsoft.

What Computex 2026 Revealed

At Computex 2026, Wiwynn and Ayar Labs (and many other partners) demonstrated something that appears remarkably close to Goldman’s third box.

The joint full rack solution featured:

32 AI compute trays
64 AI ASICs
More than 800 TB/s optical bandwidth
TeraPHY optical engines
Publicly presented system diagrams indicate up to 16 SuperNova light source assemblies per compute tray. Applied across 32 trays, this implies a theoretical maximum of 512 SuperNova assemblies within the compute rack architecture.
Rack-scale optical connectivity designed for hyperscale AI infrastructure.

A link to the full video about the solution you can find here.

This is not simply a photonic switch demonstration. This is a rack-scale optical fabric supporting full optical I/O next to the XPU/ASIC. This may ultimately prove to be the most important architectural shift in AI networking. Ayar Labs claim staggering improvements:

Why This Matters

Goldman’s report is primarily modeling a switch-centric world. The architecture generally looks like:

Switch ASIC
↓
Optical Engine
↓
External Laser Source

This is a networking-centric deployment model.

The Ayar/Wiwynn demonstration move optics much closer to compute.

The architecture instead becomes:

AI ASIC
↓
Optical Engine
↓
Remote Light Source
↓
Optical Fabric

This is a compute-centric deployment model. That distinction may sound subtle. It isn’t, this is the end game of the AI Photonics.

The Scale-Up Opportunity

One of the most important conclusions in the Goldman report is that scale-up networking becomes larger than scale-out networking. Goldman estimates that scale-up networking could represent approximately $106 billion of the total $154 billion TAM. This is a critical shift. Historically, investors thought about optical networking primarily as connections between racks (scale out). Increasingly, the largest opportunity appears to be inside the AI cluster itself.

Between accelerators.

Between racks.

Across large unified AI systems.

That is precisely the problem optical I/O is attempting to solve and the end-game for AI Photonics.

The Most Interesting Unanswered Question

The Goldman report successfully models:

Pluggables
Switch-CPO
Optical engines
External laser sources
Rack-level content

What Goldman does not yet model is the eventual transition to optical I/O at scale.

Questions remain:

When will hyperscalers deploy optical I/O?
How quickly can reliability be proven?
How will system architectures evolve?
How much optical content will be required at the compute layer?

The guess has been 2028-2030 with ramp start in 2027.

We have a model and a TAM for Goldman’s switch-CPO assumptions. For CPO they suggest approximately: 30–60 External light sources per rack. Which contributes to Goldman Sachs' projected $154 billion optical networking TAM.

However with Wiwynn and Ayar Labs Computex 2026 announcement we get the first glims of the total number of External light sources per rack in a CPO XPU/XPU architecture, which shows 512 units per rack.

Conclusion

The headline takeaway from Goldman’s report was a potential $154 billion optical networking market. After Computex 2026 takeaway may be something else. Goldman’s roadmap acknowledges a future where optics move beyond switches and become part of the compute fabric itself.

At Computex 2026, Wiwynn and Ayar Labs may have provided one of the clearest public demonstrations yet of what that future could look like.

The question is no longer:

Will optical networking become critical to AI infrastructure?

Goldman has already answered that.

The question now is:

How quickly will the industry move from only switch-centric CPO to also start using optical I/O?

This transition may ultimately define the next phase of AI networking, since the Wiwynn and Ayar Labs Computex 2026 announcement offers a first understanding of the total number of External light sources per rack in a CPO XPU/XPU architecture. If the SuperNova assembly count can be compared directly with Goldman’s ELS assumptions, the resulting light-source density would be materially higher (8-15x) than the switch-centric CPO architectures modeled today. Further work is needed to determine whether these metrics are directly comparable. This would be replacing old copper NVLink type solutions and hence will be a 100% add-on to the current TAM.

Interestingly, this possibility aligns with NVIDIA's recent expansion of the NVLink Fusion ecosystem, which now includes optical-interconnect specialists such as Ayar Labs, LightMatter. While NVIDIA has not publicly disclosed its long-term Optical I/O roadmap, the ecosystem being assembled today appears increasingly consistent with the type of "CPO with XPU" architecture Goldman identified but could not yet model.

Goldman explicitly models the transition from pluggables to switch-centric CPO. Computex 2026 may have shown the industry's first glimpse of what comes next: Optical I/O at the compute layer. The question is no longer whether optics will become critical to AI infrastructure. The question is how quickly optics move from the switch to the accelerator itself.

Anders Storm

Disclosure: This analysis is for informational purposes only and does not constitute investment advice.

The author has positions in stocks that might be related to companies in this article Sivers, GlobalFoundries. Readers are highly recommended to conduct their own due diligence on any investment.

Golden Dome: Inside the Stack One of the most ambitious DoD programs ever undertaken

Anders Storm — Sat, 23 May 2026 09:18:45 GMT

Background

On May 12, 2026, the Congressional Budget Office (CBO) released a report quantifying what defense industry insiders had suspected for months. Trump's Golden Dome, the most ambitious homeland missile defense program since Reagan's SDI, will cost approximately $1.2 trillion over 20 years. That's seven times the White House's initial figure of $175 billion, and substantially higher than the Pentagon's own $185 billion architecture number.

For investors, the headline cost matters less than what it implies about scale. CBO's estimate assumes up to 7,800 space-based interceptors, an expanded sensor layer, a new battle management architecture, and a tactical communications network spanning every branch of the US military. The Missile Defense Agency has already approved 2,440 vendors under the SHIELD IDIQ contract vehicle, with a ceiling of $151 billion over ten years.

This is not one program. It is an industrial mobilization on the scale of the Apollo program or the Manhattan Project, executed through a fragmented multi-agency vendor ecosystem rather than a single prime contractor. The investment question is: who will be the actual winners?

Leonardo DRS a leading U.S. defense contractor and the primary American subsidiary of the global Italian aerospace and defense group, Leonardo S.p.A. Operating out of Arlington, Virginia describes the Golden Dome in this picture:

In February 2026, Leonardo DRS was selected as one of the prime contractors for the Missile Defense Agency's (MDA) Scalable Homeland Innovative Enterprise Layered Defense (SHIELD)

Who are obvious winners

Most defense investors are positioning around the obvious primes: Lockheed Martin, Northrop Grumman, RTX (Raytheon), L3Harris, Leonardo. They will undoubtedly capture significant value. But another interesting question is which maybe less obvious participants in the stack will see disproportionate value creation. History suggests the largest percentage returns in defense build-outs go to specialized component suppliers two or three layers below the primes, companies whose technology becomes designed-in at a level the market hasn't yet recognized.

This is an analysis of the Golden Dome stack from satellite bus down to RF chip, where the winners likely sit at each layer.

The Five-Layer Architecture

Before identifying winners, it's worth understanding what Golden Dome actually is. Public coverage focuses on the political narrative; the actual architecture is more interesting.

Layer 1 — Space-based sensors. Hundreds of LEO satellites with infrared sensors detecting missile launches within seconds and tracking exhaust plumes during the boost phase, before warheads can deploy decoys. The Space Development Agency is already building this constellation as the Transport Layer and Tracking Layer. Hypersonic and Ballistic Tracking Space Sensor (HBTSS) satellites are a key component.

Layer 2 — Space-based interceptors. The genuinely new capability. Twelve companies received OTA agreements worth up to $3.2 billion combined in late 2025 and early 2026 to develop SBI prototypes: Anduril, Booz Allen Hamilton, General Dynamics Mission Systems, GITAI USA, Lockheed Martin, Northrop Grumman, Quindar, Raytheon, SciTec, SpaceX, True Anomaly, and Turion Space Corp. First integrated flight test target: 2028. CBO estimates the SBI fleet alone could reach 7,800 units at full deployment.

Layer 3 — Ground-based strategic interceptors. Existing GBIs in Alaska and California expanded with Next-Generation Interceptor, plus THAAD, Patriot PAC-3 MSE, SM-3, SM-6, Glide-Phase Interceptor, and Aegis Ashore installations. This layer is the "known" part of US missile defense, but it's being substantially modernized.

Layer 4 — Command, control, communications, computers, and battle management (C5ISR). AI-driven decision systems operating in milliseconds. The Pentagon has described this as "data centers in space" — integrating data from infrared satellites, ground radars, and interceptor seekers into unified tracking and engagement decisions.

Layer 5 — Transport and tactical communications. The resilient, jam-resistant communications backbone linking everything together. This includes inter-satellite optical links, ground stations, and tactical SATCOM terminals deployed across Army, Navy, Air Force, and Marine Corps units.

The instinct of most investors is to focus on Layer 2 (the interceptors are the visible, glamorous part) or Layer 3 (the existing defense primes). But the most underappreciated layers — economically and strategically — are 4 and 5. Without resilient communications and AI battle management, every other layer is blind. Adversaries will not waste resources attacking interceptors directly when they can simply jam or blind the architecture that coordinates them.

Who Wins at Each Layer

Layer 1 — Sensors and Tracking

The winners here are the satellite bus manufacturers and infrared sensor specialists already executing on the Space Development Agency's Transport and Tracking Layers. L3Harris Technologies is the dominant prime, with multiple HBTSS contracts and Tracking Layer satellites in production. Northrop Grumman holds significant work on the OPIR (Overhead Persistent Infrared) program, the inheritor of legacy DSP/SBIRS missile warning capabilities. Leonardo DRS as already mentioned

Lower in the stack, infrared focal plane array suppliers — Teledyne Technologies, BAE Systems (through its US subsidiary) — provide the sensors themselves. These are not pure-play Golden Dome investments but represent solid medium-term exposure.

The dark horse: York Space Systems (NYSE: YSS). York IPO'd in January 2026 and has emerged as a credible small-satellite prime. The company has delivered 21 LEO satellites to SDA under a 2022 Transport Layer contract, won a $237 million IDIQ from US Space Force in May 2025 explicitly naming "Golden Dome demonstrations" among supported missions, and added multiple additional IDIQ awards in May 2026 explicitly tied to Golden Dome. York's vertical integration strategy (bus + ground operations + battle management + tactical terminals) is positioned for the kind of mission-package contracts SHIELD is structured to issue.

Layer 2 — Space-Based Interceptors

This is the most contested layer. SBI prototype contracts are funded. The 2028 flight test will determine which of the twelve OTA holders advance.

Possible winners: Lockheed Martin (deepest experience in missile interceptor production), Northrop Grumman (strong sensor-to-shooter integration capability), Raytheon / RTX (existing interceptor production base. Anduril is the newcomer to watch, they've moved aggressively into space systems and have political alignment with the current administration.

The lower-stack winners depend on which prime wins. Propulsion systems (Aerojet Rocketdyne, now part of L3Harris), guidance computers (specialized rad-hard processors from BAE, Cobham, Mercury Systems), and seeker electronics will all see substantial volume if the program produces at the 7,800-unit CBO estimate.

Layer 3 — Ground-Based Interceptors

Possible the most predictable layer. Lockheed Martin dominates THAAD. RTX dominates Patriot and SM-series. L3Harris holds significant ground radar work.

This is where existing defense majors capture the most committed dollars, but it's also where multiples are already extended and the surprise upside is limited. The long case for the primes is volume and longevity, possibly not multiple expansion.

Layer 4 — AI Battle Management and C5ISR

An interesting layer for technology investors, because it's the one most likely to disrupt traditional defense procurement patterns. AI-driven battle management at the speed required, millisecond decisions across thousands of potential threats, is not something traditional primes have demonstrated.

The natural winners are Palantir Technologies (Mission Maven and AIP platform are already deployed in DoD contexts), Anduril Industries (Lattice software platform), and emerging firms like Shield AI and Scale AI. Larger players include Booz Allen Hamilton for systems integration work.

The hardware enablers for AI in space, radiation-hardened GPUs, secure compute modules, represent a smaller but interesting sub-segment. Mercury Systems, Kratos Defense, and Curtiss-Wright all have positions here.

Layer 5 — Tactical Communications

Here is where the analysis gets genuinely differentiated. Tactical SATCOM terminals are the "boring" component some investors skip, except they're the component without which the entire architecture is inert.

Viasat, Hughes (EchoStar), L3Harris, and General Dynamics Mission Systems are the incumbent prime suppliers. They will all win significant volume. Iridium Communications has unique low-latency Ku-band coverage that complements the architecture.

But the technology shift in tactical terminals favors a smaller, newer player. The traditional terminals use mechanically-steered antennas. Modern tactical SATCOM demands electronically steered arrays (ESAs), flat-panel antennas with no moving parts, capable of tracking multiple satellites simultaneously and providing "make-before-break" connectivity as satellites pass overhead.

York Space acquired All.Space

ALL.SPACE is the leader in multi-orbit, multi-link SATCOM solutions. They have successfully completed testing under the U.S. Army’s Next Generation Tactical Terminal (NGTT) program. The company’s Hydra MAX terminal has been validated as a first-of-its-kind solution capable of delivering seamless, simultaneous, and resilient connectivity across Low Earth Orbit (LEO), Medium Earth Orbit (MEO), and Geostationary Orbit (GEO) satellite networks – all while on the move.

“With this acquisition, York is creating a complete communications ecosystem that operates in contested environments across commercial and government networks,” said Dirk Wallinger, CEO of York. “The addition of ALL.SPACE brings a world-class team whose proven leadership and technical excellence will be key as we scale these capabilities for our customers.”

All.space ESA terminal is purpose-built to align with the Department of Defense’s Joint All-Domain Command and Control (JADC2) objectives, Hydra MAX supports a Modular Open Systems Approach (MOSA), enabling rapid integration with future constellations, including Telesat Lightspeed once Hydra MAX is certified for the network. With multi-beam transport diversity, integrated SD-WAN resilience

What major technology is behind a ESA terminal

ESA terminals require massive numbers of beamforming RFICs, specialized chips that electronically steer the radio beam. This is a small, technically demanding market dominated by perhaps five suppliers globally. And one of them, almost no defense investor has heard of.

The Component Layer in terminals

Look at any modern high-end phased-array antenna and you'll find roughly 1,000 to 2,000 beamforming chips arranged in a grid pattern, each contributing a small piece of the steering signal. The chip count scales with antenna size and capability. A high-end military ESA terminal capable of simultaneous multi-orbit connectivity contains around 1,500 - 2,000 beamforming ICs.

The market for Ka-band beamforming RFICs is dominated by three companies: Qorvo (acquired Anokiwave in 2022), Renesas Electronics (through its IDT acquisition), and Sivers Semiconductors (NASDAQ Stockholm: SIVE). Each has different strengths. Anokiwave/Qorvo has been the volume incumbent and started SATCOM BFIC development long before SIVE. Sivers has captured a position in the most high-performance defense segment that few DoD investors has noticed. Via the acquisition of Mixcomm.

Sivers Semiconductors is a Sweden based fabless chip designer. Its Wireless division produces Ka/Ku-band beamforming RFICs and integrated ESA panels. Its Photonics division produces InP DFB lasers for AI data center optical interconnects, a separate business, see my other substack on this link .

Important fact: Sivers is the single-source supplier for the beamforming chips in the All.Space Hydra tactical SATCOM terminal.

All.Space, formerly Isotropic Systems, builds what is generally considered the most advanced military SATCOM terminal in the world. The Hydra connects to four satellites simultaneously across different orbits and bands. It is jam-resistant by design. It already has contracts with US Army, US Navy, and NATO, and has a mass production facility in Alabama. On April 28, 2026, York Space Systems announced the acquisition of All.Space for $355 million.

On York's Q1 2026 earnings call, CEO Dirk Wallinger described Hydra as "a phenomenal fit for Golden Dome. You know, assured communications that are far more resistant to jamming is definitely required for Golden Dome specifically."

This connects three things that few investor have connected:

York is a Golden Dome contractor with $237M+ in IDIQs explicitly naming the program

All.Space is now York's tactical communications subsidiary

Sivers is single-source supplier inside All.Space's flagship terminal

The unit economics

Sivers' own investor presentations confirm the chip economics:

Approximately $5 per chip for Ka-band. 1,800 chips per Hydra terminal, $9,000 of Sivers content per terminal

The new dual-band Ka/Ku chip will have a higer ASP raising per-terminal content. The dual-band capability is significant it expands Hydra's addressable market to include legacy Ku-band constellations including portions of the Viasat fleet, older WGS, and most maritime SATCOM systems.

Single-source positioning in DoD-qualified terminals is not a casual vendor relationship. Re-qualifying an alternative beamforming RFIC requires complete RF front-end redesign, re-certification against MIL-STD-810, MIL-STD-461, and MIL-STD-704, plus re-testing across Army, Navy, and NATO programs that already have Hydra in service. Industry-standard timeline: 18 to 36 months. Industry-standard cost: $5 to $20 million per re-qualification. Defense terminal production lifecycles run 15 to 20 years.

All.Space is, in practical terms, economically locked into Sivers for the foreseeable production lifecycle of Hydra.

Why "Swedish supplier" is the wrong framing

The natural objection is: "How does a Swedish company become single-source in a DoD-critical terminal?" The answer reveals one of the most underappreciated facts about Sivers' actual structure.

Sivers Wireless is fabless. The chips are designed by engineers in Sweden and in New York (acquired through MixComm in 2022), but they are manufactured by GlobalFoundries at facilities in Vermont and upstate New York. GlobalFoundries (NASDAQ: GFS) is one of the few DoD-designated "trusted foundries" certified for sensitive defense electronics, and was a minority investor in MixComm before the Sivers acquisition.

From a DoD supply chain perspective, Sivers' chips are produced in the United States by a trusted American foundry, packaged and tested in American facilities, with design IP partially held by US-based engineers in New York. The "Swedish" label refers to the parent company's listing venue and headquarters — not the production trust chain.

This is precisely the supply chain structure the CHIPS and Science Act was designed to support. And the Pentagon has noticed.

The Second Path — CHIPS Act

Sivers may also have a second, independent path into Golden Dome that has received even less investor attention.

In October 2024 and January 2025, Sivers signed contracts with the Northeast Microelectronics Coalition Hub under the US CHIPS and Science Act's Microelectronics Commons program. The contracts, executed through the Naval Surface Warfare Center Crane Division, fund two parallel chip development

programs:

Electronic Warfare full-duplex arrays, chips capable of transmitting and receiving simultaneously on the same frequency through ultra-precise self-interference cancellation

5G/6G FR3 beamformer ICs — mid-band beamforming for both commercial and defense applications

First-year funding: $11.6 million. Total program value if renewed across three years: approximately $30 million. Link to deep dive here.

The partner list, named explicitly in Sivers' own press releases, is the interesting part: BAE Systems, Raytheon, MIT Lincoln Laboratory, Columbia University, and Ericsson.

The second round of the awards was released 19th of May: $6.6M Year 2 Program Award through NEMC Rewards Strong Execution and Reinforces Growing Momentum in Modernizing U.S. Defense Infrastructure Using Sivers Technology.

The CHIPS Act designation matters beyond the funding itself. Sivers has been assessed as strategically important enough that the US government channeled federal money through a Northeast hub to a Swedish-parented company.

Sivers now has two independent paths into Golden Dome. Path A: single-source supplier to All.Space, acquired by York, building tactical terminals for the program's communications layer. Path B: CHIPS Act-validated component supplier.

Conclusion

The Golden Dome stack has obvious winners and hidden ones. . The hidden winners are smaller, more specialized, and harder to identify because they sit two or three layers below the primes in technology stacks that defense generalists don't analyze.

Among these hidden winners sit the component suppliers behind tactical SATCOM terminals, specifically the beamforming RFICs that make modern ESA terminals possible. York's acquisition of All.Space was not random. The acquisition gives York vertical integration into the communications layer of the very architecture the program needs. And All.Space, in turn, is single-source dependent on Sivers Semiconductors for the chips that make Hydra work.

Golden Dome applications also effect the next layer like chip manufacturing through GlobalFoundries.

The thesis is not "AI infras”

The thesis is not "AI infrastructure will continue" or "Golden Dome will be funded." Both are true and neither is unique insight. The thesis is that the connections between three concurrent macro trends, hyperscaler AI capex, missile defense modernization, and US semiconductor reshoring, flow through a single Swedish-listed microcap whose dual-listing on Nasdaq is on the way.

On May 22, 2026 — one day before this is published — Sweden and the United States signed a new memorandum of understanding on technology cooperation covering AI, defense innovation, space, and quantum technology. It is a strong signal that the two countries are formalizing strategic alignment in precisely the domains where Sivers operates.

This is what asymmetric opportunity looks like before consensus arrives.

Anders Storm

23 of May 2026

Disclosure: This analysis is for informational purposes only and does not constitute investment advice.

The author has positions in Sivers, YSS and GlobalFoundries. Readers are highly recommended to conduct their own due diligence on any investment.

The AI Interconnect Layer: Where GPU, ASIC, and CPU Meet AI Photonics — and Where Most Investors Haven't Looked

Anders Storm — Sat, 09 May 2026 13:53:57 GMT

"Follow the money." It's the most useful piece of investigative advice ever offered, and it works just as well for AI infrastructure as it did for political reporting.

It's old advice, but it works. In the previous post, The AI Compute Stack, we mapped Goldman's 5 years $7.6T AI capex framework, and how that connects to GPUs, custom ASICs, and CPUs (first part of this series: read more here). Today the compute layer, is the most visible, most discussed, and most understood part of the stack.

But compute is where the AI infrastructure money story starts. Not where it ends.

Compute is just the beginning. Once you've assembled 100,000 GPUs in a hyperscaler campus, you have a fundamentally new problem: connecting them. The interconnect layer downstream of compute is where AI infrastructure becomes physically real, and it's where the next architectural battle is being fought.

This is also where Swedish investor attention falls off a cliff. To illustrate this let’s take a look at Avanza, Sweden's largest retail trading platform. NVIDIA: ~98,000 Swedish shareholders. AMD: ~28,000. These are widely held names. Then drop one layer downstream into custom ASICs and networking: Broadcom ~10,000 holders. Marvell ~2,000. And drop one more layer into the optical interconnect names that actually capture the dollar flow when AI clusters scale: Lumentum ~800 holders. Coherent ~800 holders.

The discovery curve is steep. Most retail investors stop tracking at the brand-name compute layer. The interconnect layer, where the bandwidth, the power efficiency, and arguably the larger long-term economics live, remains mostly invisible to non-specialists.

That's an opportunity!

Why Compute Without Interconnect Is Useless

A 100,000-GPU cluster is just 100,000 islands of computation if you can't move data between them. AI training workloads are bandwidth-bound long before they're compute-bound. The economics flow accordingly.

Three constraints make optical interconnect non-optional at AI cluster scale.

Bandwidth density. Each generation of AI accelerator demands roughly 2x interconnect bandwidth. NVIDIA's NVLink 5 in Blackwell delivers 1.8 TB/s per GPU. The next generation targets 3.6 TB/s. Copper traces at these speeds become unmanageable in PCB routing within just a few inches.

Power per bit. At 200 Gbps per lane, electrical SerDes consume approximately 5 picojoules per bit. Optical equivalents target 1 pJ/bit. Across a 100,000-GPU cluster running continuously, the aggregate power difference reaches hundreds of megawatts — which is why hyperscaler power budgets increasingly drive networking architecture decisions.

Distance. Copper performance degrades sharply beyond ~1 meter at AI cluster speeds. Optical maintains signal integrity across racks, rows, and entire data center halls without amplification.

The optical transition isn't optional. It's an engineering inevitability for AI clusters scaling toward 100,000+ GPU deployments. The contested questions are when each architectural layer transitions, who wins each transition, and what counts as the addressable market.

Two Markets, Not One

The first thing to understand about optical interconnect is that the analyst forecasts disagree by a factor of 65. Mordor Intelligence sees a $2 billion CPO market by 2030. DigiTimes projects $130 billion. That's not methodological imprecision. It's that "CPO market" actually means two distinct architectures with different economics, different vendors, and different timing.

Coherent investors presentation OFC

Scale-out optics — rack-to-rack networking. Already a mature market dominated by pluggable transceivers. This is where 800G and 1.6T pluggables live, where Coherent and Lumentum compete with Innolight and Hisense, and where Broadcom's Tomahawk switch ASICs are increasingly winning silicon-photonics integration deals.

Scale-up optics — chip-to-chip within a rack. The contested frontier. This is where Ayar Labs, Marvell's Photonic Fabric (post-Celestial AI acquisition), POET Technologies' Optical Interposer, and external light source vendors like Lumentum, Coherent, Sivers Semiconductors compete for what could become a fundamentally larger market by 2030.

These two markets share underlying physics, both move bits via photons rather than electrons — but the economics, the vendor landscape, and the deployment timing are genuinely different. Most retail and even institutional commentary conflates the two.

The Scale-Out Story: Pluggables and Switch ASICs

Scale-out optics is the more mature market, and the more concentrated. Three players matter most.

Coherent Inc. is the merchant pluggable transceiver leader. Their Q1 2026 earnings disclosed total optical SAM of $50 billion in their existing markets, with $20 billion incremental SAM from emerging AI applications by 2030. That $70 billion total SAM by 2030 is a useful anchor for evaluating other forecasts.

Coherent's strategic position is strong because they have InP fabrication capability (acquired through II-VI). InP wafer supply is a structural bottleneck for AI optics, and Coherent controls a meaningful share of merchant InP capacity.

Lumentum is Coherent's primary competitor in pluggables. Lumentum has historically been stronger in DSP-based pluggables and has deep relationships with NVIDIA on indium phosphide laser supply for both pluggable and emerging CPO applications.

Broadcom sits at a different layer but increasingly captures pluggable economics through their switch ASICs. The Tomahawk 5-Bailly platform integrates silicon photonics directly into the switch, displacing pluggable transceivers in some hyperscaler deployments. Broadcom shipped 50,000+ Bailly units in 2025 and is ramping capacity for 2026 deployments at Google, Microsoft, and Meta.

Three drivers worth tracking:

800G and 1.6T pluggables ramp. LightCounting projects pluggables retain 80%+ market share through 2030, even as CPO grows.

LPO/LRO emergence. Power-optimized pluggable variant that competes with CPO at the rack level. AOI, Jabil and Hisense are leading here.

LPO (Linear Pluggable Optics): Removes the Digital Signal Processor (DSP) entirely from the transceiver, relying on the switch ASIC for signal equalization. This offers the lowest power but has higher interoperability risks.

LRO (Linear Receive Optics): Also known as "Half Retimed Optics," it keeps a DSP on the transmitter side (for signal stabilization) while removing it from the receiver side. It offers roughly half the power savings of LPO, but with superior link stability and easier adoption. (Jabil latest solution under development).

Tomahawk-style switch ASIC integration. Broadcom's silicon photonics integration displaces some merchant pluggable revenue but creates new architectural dependencies on Broadcom-designed clusters.

Scale-out is real, growing, and increasingly dominated by the Broadcom-Coherent-Lumentum triangle.

The Scale-Up Story: Where the Real New Money Could Be!

Scale-up optics is the new frontier and the larger long-term opportunity.

The architectural premise: as accelerator clusters scale, the bottleneck shifts from rack-to-rack bandwidth (where pluggables work fine) to chip-to-chip bandwidth within the rack. Once you need 10+ TB/s of bandwidth between accelerators in a single chassis, copper electrical interconnect becomes physically impossible. Optical chip-to-chip is the only path forward.

Several vendors are competing for this scale-up layer.

Ayar Labs has shipped TeraPHY optical I/O chiplets and SuperNova external light sources. They're funded by Intel, NVIDIA, and the US Department of Defense, with deployments at AWS Trainium and Intel's accelerator programs. Ayar Labs' architecture co-packages an optical I/O chiplet directly with the accelerator die, with light sources provided externally for thermal isolation.

Marvell acquired Celestial AI for $3.25 billion in December 2025. Celestial AI's Photonic Fabric technology is direct competition with Ayar Labs in the chip-to-chip optical layer. The acquisition gives Marvell a complete custom ASIC + optical interconnect platform, a structural advantage versus Broadcom, which would need to build optical interconnect capability separately.

POET Technologies offers the Optical Interposer platform, a different architectural approach that integrates lasers, modulators, photodetectors, and CMOS electronics on a single substrate.

NVIDIA's Spectrum-X and Quantum-X CPO switches were announced for 2026 deployment. NVIDIA's approach integrates silicon photonics directly into their networking ASICs, similar to Broadcom's Bailly platform but for AI clusters specifically.

External Light Sources (ELS), the laser layer, is a separate but adjacent market. Most scale-up optical architectures use externally located lasers (rather than integrating laser sources directly with the accelerator) for thermal management reasons. This creates a market for high-power, narrow-linewidth InP lasers from vendors including Lumentum, Coherent, MACOM, Sivers Semiconductors etc.

The scale-up market is more contested, less mature, and harder to size. But it's where the larger long-term economics could land if hyperscalers genuinely transition to optical chip-to-chip at the volumes some analysts project (which the market now believes will happen).

The Sizing Disagreement Decoded

Now the analyst forecast spread makes sense.

Optimistic forecasts ($50-130B by 2030):

DigiTimes: 142% CAGR (excluding ELS)

Yole Group: $8.1B by 2030, 137% CAGR

Coherent's own SAM: $70B + total by 2030

These forecasts include scale-up optics, ELS, and the implicit assumption that chip-to-chip optical reaches significant deployment by 2030.

Conservative forecasts ($2-8B by 2030):

Mordor Intelligence: 36% CAGR

IDTechEx: 37% CAGR

These forecasts focus narrowly on "true CPO", defined as silicon photonics modules co-packaged with switch or accelerator ASICs. They exclude pluggables, exclude ELS, and apply more conservative assumptions about scale-up adoption timing.

Both lenses are correct. They're measuring different things. The 65x spread reflects definitional differences, not analytical error.

What Avanza Numbers Actually Tell Us

Back to the Swedish retail discovery curve.

NVIDIA's ~98,000 Swedish shareholders on Avanza reflect what works at the brand layer: a name everyone has heard of, a stock everyone has read about, a chart everyone has watched go vertical. AMD's ~28,000 holders reflect the second-mover effect, the credible challenger, recognizable enough to be a "second AI bet."

Then the layered drop. Broadcom's ~10,000 Swedish holders is interesting because Broadcom genuinely is one of the largest AI infrastructure beneficiaries, by some measures more than AMD. Broadcom captures roughly 60-70% of the custom ASIC market (anchored by Google TPU and Meta MTIA), runs the dominant networking switch ASIC franchise (Tomahawk 5-Bailly with 50,000+ units shipped in 2025), and is increasingly capturing pluggable transceiver economics through silicon photonics integration. Yet its Swedish retail holder base is roughly one-third of AMD's, despite arguably broader AI exposure.

Then another step down. Marvell at ~2,000 holders reflects the gap between merchant GPU recognition and custom ASIC understanding. Marvell is the second major custom ASIC design house, anchored by AWS Trainium and Microsoft Maia programs. Its custom silicon business runs at $1.5 billion annualized revenue, and its December 2025 acquisition of Celestial AI for $3.25 billion gave it integrated optical interconnect capability. By any analytical measure, Marvell is more central to the AI infrastructure story than its Swedish retail visibility suggests.

Lumentum at ~800 holders. Coherent at ~800 holders. These are the merchant pluggable leaders directly capturing optical interconnect dollars as AI clusters scale. The combined Avanza holder count for Lumentum and Coherent (~1,600) is 5% of AMD's Swedish retail base — and 1.6% of NVIDIA's.

The discovery curve compresses dramatically as you move downstream:

NVIDIA: ~98,000 holders (compute, brand layer)

AMD: ~28,000 (compute, second source)

Broadcom: ~10,000 (custom ASIC + networking)

Marvell: ~2,000 (custom ASIC second mover)

Lumentum: ~800 (optical interconnect)

Coherent: ~800 (optical interconnect)

The gap matters because it tells you something about market efficiency. Brand-layer compute names trade on broad investor attention. Interconnect-layer names trade on a much narrower base of specialists. That's not necessarily mispricing, sometimes the broader investor base is right and the specialists are over-confident. But it's a structural feature worth knowing about.

Worth noting is the share price development over the last year

AI Photonics layer: Lumentum +1200% , Coherent +400%

Upstream: Broadcom +100%, Marvell 200%, NVIDIA +80%, AMD 350%.

The later group for example NVIDIA have had an early revaluation over 3 years at +600%.

For a Swedish investor specifically: the discovery curve from NVIDIA → Lumentum is a 99% reduction in retail competition.

What This Means for Vendor Selection

The architectural split clarifies vendor strategy.

For scale-out exposure (more concentrated, more mature):

Coherent — InP capability + pluggable leadership

Lumentum — DSP pluggables, NVIDIA relationships

Broadcom — switch ASIC integration capturing pluggable economics

AOI, Innolight — pluggable competition

For scale-up exposure (more contested, more upside):

Ayar Labs — direct chip-to-chip optical I/O

Marvell (Celestial AI) — integrated ASIC + optical platform

POET Technologies — Optical Interposer architecture

NVIDIA — Spectrum-X/Quantum-X CPO switches

Laser/ELS/Materials suppliers: Lumentum, Coherent, MACOM, Sivers, AXT, IQE.

An AI bet on Macro level with a diversified portfolio across all layers captures the architectural transition without taking a position on which scale-up architecture wins, also including Hyperscalers could be even better. A concentrated bet on scale-up specifically requires a higher-conviction but gives a higher-variance.

Hence Macro plays in all layers in the stack could be a possible path, including: Hyperscaler - GPU/CPU - ASIC - AI Photonics - Lasers - Materials (to simplify the stack).

What to Watch in 2026-2027

Several deployment data points will resolve much of the architectural uncertainty over the next 24 months:

1. NVIDIA Spectrum-X / Quantum-X production deployments. Confirmed for 2026 but with limited disclosed deployment scale. AMD 2026 data on CPU to GPU ratio change.

2. Marvell Photonic Fabric customer disclosures. Post-Celestial AI integration, Marvell needs to disclose hyperscaler design wins for the integrated ASIC + optical platform. Without disclosed wins by H2 2026, the $3.25B acquisition thesis weakens.

3. Ayar Labs production scale. Ayar Labs has shipped engineering samples but commercial deployment scale remains unclear. Production ramps in 2026 would validate the chip-to-chip optical architecture broadly.

4. Broadcom Bailly volume trajectory. Broadcom shipped 50,000+ units in 2025. The 2026 trajectory determines whether silicon photonics integration in switch ASICs becomes the dominant scale-out architecture or remains a niche optimization.

5. InP lasers and wafer supply expansion. Multiple vendors (Lumentum, Coherent, MACOM, Sivers, AXT, IQE) are expanding InP capacity. The pace of capacity addition relative to demand growth determines whether laser supply becomes a structural bottleneck.

The Architectural Takeaway

Optical interconnect isn't one market. It's two markets with shared physics but distinct economics.

Scale-out optics is real, growing, and concentrating around Broadcom-Coherent-Lumentum. The economics are good but the structural growth is moderate for the incumbents, but could be huge for new players.

Scale-up optics is contested, still less mature, and harder to size. But it's where the larger long-term economic opportunity lives if hyperscalers genuinely transition to optical chip-to-chip at the volumes some analysts project.

The dollar flow inside Goldman's networking line item depends critically on which scale-up architecture wins, when it wins, and at what deployment volume. Analyst spreads of 65x reflect that genuine uncertainty — not analytical sloppiness.

The vendors building credibly in scale-up, Ayar Labs, Marvell post-Celestial AI, POET, NVIDIA's CPO switches, and the external light source ecosystem, are competing for what could be the most valuable layer of the AI infrastructure stack.

Watch the deployment data over 2026-2027. The architectural question resolves in the actual chips that ship, not in the analyst forecasts. And while the broad retail attention concentrates on the brand-layer compute names, the interconnect layer is quietly capturing an increasing share of the actual AI capex dollars.

Anders Storm

Disclosure: I am a shareholder in AMD, Marvell and Sivers and the former CEO of Sivers Semiconductors (2017-2024). This commentary reflects publicly available information and the author's analytical framework. Nothing in this post constitutes investment advice or a recommendation to buy or sell any security. Avanza shareholder figures are approximate retail holdings on the Swedish trading platform Avanza Bank.

The AI Compute Stack: How Hyperscalers Tie Into GPUs, ASICs, and CPUs

Anders Storm — Wed, 06 May 2026 06:29:39 GMT

The AI capex narrative is dominated by a single name “NVIDIA”. NVIDIA’s GPU dominance is real, but that’s not the full story.

The reality is more interesting and structurally important. Goldman Sachs Global Institute’s recent framework projects $7.6 trillion in AI capex from 2026-2031. Within that universe, the silicon stack is genuinely contested. Custom ASICs designed by Broadcom, Marvell, and Alchip are scaling at 50% CAGR. AMD is growing share at the merchant GPU layer (SemiAnalysis projects AMD share doubling to ~10% by mid-2026). CPU-to-GPU ratios are shifting as agentic AI workloads change the compute mix over time.

The “who builds what” question matters because it determines where the actual dollars flow. The answer is more diverse than the current NVIDIA-centric narrative suggests.

The Capex Universe

Goldman's framework breaks the $7.6T into three layers:

- Compute: $5.1T (~67%)

- Data Centers: $2.1T (~28%)

- Power: $358B (~5%)

Hyperscaler concentration is striking. Combined hyperscaler capex reaches $660-690 billion in 2026 alone, with Google, Microsoft, Amazon, and Meta each planning $60-80 billion individually. Approximately 75% of this is directed at AI-specific infrastructure. The buyers are concentrated. The sellers, less so.

Goldman's analysis identifies four assumptions that materially move the total: silicon useful life, data center cost per megawatt, chip architecture mix, and build-out elongation. The third assumption, chip architecture mix is where the multi-architecture reality lives.

The Merchant GPU Layer

NVIDIA dominates current AI compute. The H100, H200, and Blackwell B100/B200 generations have captured roughly 85-90% of the AI training and inference market. NVIDIA's gross margins run approximately 75% on data center silicon, reflecting both architectural advantages and CUDA software lock-in. But the merchant GPU layer is not single-vendor. AMD has emerged as a credible challenger with the Instinct MI300X, MI325X, and the upcoming MI350/MI400 series. AMD's strategy is straightforward: hyperscalers want a second source, and AMD provides one with reasonable performance at meaningful cost discount.

The MI300X has won deployments at Microsoft, Meta, and Oracle. The MI350 ramps in 2026. The MI400 series, planned for 2027, targets aggressive performance parity with NVIDIA's then-current generation. AMD's challenge isn't hardware, it's ROCm software ecosystem maturity. Until ROCm credibly matches CUDA, AMD remains the second source rather than a true competitor.

Realistic merchant GPU share by 2028: NVIDIA 70-80%, AMD 15-25%, others <5%. The duopoly serves hyperscalers' interest in optionality, but the larger structural shift is happening at a different layer.

Realistic 2028 merchant GPU share projections vary by analyst. SparkCo’s analysis of IDC and Gartner baselines provides three scenarios for AMD: bull case 25-30% share with $18-22 billion revenue (driven by aggressive hyperscaler diversification), base case 15-20% share with $10-14 billion revenue (aligned with current MI300 deployments scaling), and bear case <10% share if NVIDIA’s H200/H300 generation maintains exclusivity.

The base case scenario implies an NVIDIA-AMD merchant duopoly of roughly NVIDIA 70-80%, AMD 15-20%, others <5% by 2028. AMD’s own forecasts from Lisa Su’s Advancing AI presentation are more aggressive ($500B+ TAM, 60% CAGR), though analysts including The Next Platform argue both AMD and Gartner projections may be miscalibrated in different directions.

The duopoly serves hyperscalers' interest in optionality for GPUs. But the larger structural shift is happening at a different layer: custom ASICs designed by Broadcom, Marvell, and Alchip.

Custom ASICs

Custom ASICs are projected to surpass merchant GPU shipments by volume in 2028.

James Carter at Oplexa Insights writes: “NVIDIA built a $4.6 trillion empire selling one product: the world’s most powerful AI chip. But the companies that made Jensen Huang a legend are now building their own silicon, and the Custom ASIC market 2026 is the weapon they are using to do it.

Google does not buy NVIDIA GPUs to run Gemini. Amazon does not pay NVIDIA margins to train its AI models. Meta is not using general-purpose accelerators to serve 3 billion users daily. They are all running custom Application-Specific Integrated Circuits, chips designed from scratch for one purpose, executing at maximum efficiency, at a fraction of the power cost.”

The New Goldman Sachs report: "Tracking the Trillions": What Goldman's AI Capex Framework Tells Us About the Photonics Layer

Anders Storm — Sat, 02 May 2026 18:17:45 GMT

The Goldman Framework

AXTI: Why is this almost unknown materials company up +5,500% in one year — and still barely known in Sweden?

Anders Storm — Fri, 01 May 2026 11:29:14 GMT

Summary

AXT Inc. is up roughly 5000% in a year.

Not because of earnings. Not because of profitability. But because the market believes it sits at a hidden bottleneck in AI infrastructure:

Indium Phosphide (InP) — the material enabling high-speed optical networking.

AXT has 366 owners on Avanza trading platform (which I guess has 30-50% of the retail traders in Sweden). Congratulations to you all, I missed this one.

Why This Matters for Swedish Investors

If you follow NVIDIA or AMD, you already understand how fast AI compute is scaling. Driven by hyperscalers that will invest $1 trillion according to Dagens Industi (DI)

And if you know Sivers Semiconductors, you’ve seen how InP is used in photonics laser arrays further down the stack from the mega companies. $SIVE has approximately 17.500 Avanza shareholders to compare with ATX.

AXT sits downstream for example Lumentum. It supplies the raw substrate there lasers are grown on. As far as I know Sivers doesn’t use ATX for substrates for 3-4” production today, they historical use IQE for EPI waffers and substrates. Possibly WIN Semiconductors could, Sivers volume production partner for AI Photonics, I have not investigated this.

The Story Behind the 5000% Move

As AI clusters scale, data movement becomes the bottleneck. Copper interconnects stop scaling efficiently.

So the industry shifts to optical networking.

Companies like Marvell and Lumentum are driving the transition from 800G to 1.6T and CPO and beyond. The optical market could grow from roughly $18B to $90B+ by 2030.

At the same time, the underlying material is changing.

Indium Phosphide is going from around 40% of optical lanes today to ~90% by 2030.

This is not just growth — it’s a technology shift and big future bottleneck.

Where AXT Sits in the Stack

AI Compute
NVIDIA
AMD

Networking / Optical Chips
Marvell

Optical Components including InP
Lumentum & Coherent

InP Lasers and arrays

Sivers Semiconductors

Epitaxy
IQE

Substrate
AXT
Sumitomo Electric

Raw Materials

InP supply

Key Insight

AXT sits at the deepest layer — where supply can be bottleneck if upstream grows quickly. Traditionally this layer has been challenged with lower gross margins.

What the Market Is Pricing In

At roughly a $3.5B valuation, the market assumes:

InP demand explodes
AXT scales production successfully
Gross margins expand from current 29% (Q1 2026) toward 40–45%
Supply becoming even more constrained

In the 2026 Q1 report, 30th April 2026. The company shared that InP backlog has climbed to “a new high of over $100 million” and that AXT is “running ahead of our plan to double our Indium Phosphide capacity this year from Q4 of 2025 levels.”

They also outlined additional expansion: “Beyond our 2026 capacity expansion, we’re planning to double our Indium Phosphide capacity again in 2027 with a new facility near our current one that we will be dedicated to Indium Phosphide wafer production.”

In the Q&A, they provided a revenue-based capacity framing, saying AXT expects to reach “about $35 million per quarter capability by the end of 2026” for InP, after having increased capacity about 25% in Q4 2025 and aiming to double that in 2026. VP of Business Development Tim Bettles added that after completing the next phase of expansion, “by the end of 2027, maybe early 2028,” AXT expects to be “somewhere in the region of $65 million-$70 million of capacity per quarter.” They also said the company is also planning for 2028 expansion and noted that “scale matters” in the InP market and that barriers to entry are high

In short, AXT is being priced for extreme growth and a bottleneck supplier to the AI Photonics economy where the need for InP materials and lasers will be key. No InP - No Lasers -No Photons - NoAI.

Final Take

AXT is in the right place at the right time. The Avanza pepole who found it earlier or around a year ago. At 1st of May 2025 AXT was $1.35 → now at +$70, they have had a crazy ride. Congratulations to all 366 that might have been on that journey. I missed that journey, but I should have paid attention.

Anders Storm

Disclaimer: This newsletter is NOT investment advice. The aim of this newsletter is to provide subscribers with basic and simplified insights about the semiconductor industry and its participants. Disclosure: I have no position in AXTI.

Articles will be published semi-frequently.

6 weeks that change $SIVE

Anders Storm — Thu, 23 Apr 2026 03:45:39 GMT

While sentiment around Sivers Semiconductors ($SIVE) in Swedish media has largely been negative, in some cases with reason, however the reality beneath the surface looks different.

In just six weeks, a major fundamental shifted has happened and the company is now seen with new eyes, as a major player in the AI photonics bottleneck.

A decade of growth

Sivers has grown from:

• 2016: From $2m to $41M (DNB Carnegie Access estimate) for 2026

That’s:

• ~20x revenue growth
• ~35.5% CAGR over 10 years

That should stand tall in almost any comparison.

Understanding the business mix

2025 revenue:

• 72% NRE (Non-Recurring Engineering)
• 28% product sales

This matters.

NRE means customers are paying upfront to develop laser solutions they intend to deploy at scale.

In other words:

You don’t fund development unless you plan to buy volume.

When volume ramps, revenue mix flips and scales fast.

The re-rating: 6 weeks

• March 11, 2026 (Q4 report):
→ Market cap $120M

• April 22, 2026:
→ Market cap $1B

~8.5x in 6 weeks

What actually happened?

Photonics AI happened. But also a rediscovered now by US investors, see more below.

The stack the US market understand

Very simplified:

OpenAI
→ Hyperscalers (Amazon, Microsoft)
→ NVIDIA / AMD
→ Marvell
→ Celestial AI / Ayar Labs
→ POET
→ Lumentum / Sivers
→ WIN Semi / GlobalFoundries
→ IQE

This mapping is not obvious.

It requires deep understanding of the semiconductor and photonics ecosystem.

US started the work

About a year ago, this mapping was already done by specialists in the US.

They weren’t looking at:

• last twelve months
• current earnings

They were looking:

3–5 years forward
→ identifying bottlenecks

And they found a few but in this case:

The laser layer and as you can see 52 week stock returns has been 500-6800%. Name in the picture below.

Picture from https://www.substack.com/@damnang

The trigger for Sivers

While this was already happening in the US:

The US semiconductor researcher Moody and the Swedish Siliconmatters, published work on Sivers 11th and 13th March 2026. Serenity published 16th of March investments in $SIVE.

That research reached:

• Engaged US semi investors including some names om WSB and X (Serenity)
• deep tech communities

They saw something unexpected:

A ~$120M company in Sweden
positioned similarly to multi-billion dollar photonics leaders

Same ecosystem
Same partners
Same end market

Then the catalyst

NVIDIA made a decisive move:

• ~$2B into Lumentum
• ~$2B into Coherent
• ~$2B into Marvell

Why?

Because they see:

A laser bottleneck in AI photonics (InP)

On top of this, AI Photonics is also hot in not listed companies

• Celestial AI → acquired by Marvell (~$3.2B)
• Ayar Labs → ~$3.75B valuation, $500M funding round

Both:

• early-stage
• pre-scale
• positioned for 2027–2028 ramp

Is this hype?

That’s for each investor to decide, there are big risk with qualification and yield still to be aware of.

But this is not a typical story.

• AI is the new use cases
• demand seems to be real and growing
• infrastructure is still being built

The overlooked timing

Two years before the AI acceleration (around 2022):

Sivers was already:

• engaging with key ecosystem players
• aligning with photonics leaders
• positioning in laser technology

Alongside:

• Lumentum
• Coherent
• Ayar Labs

Coincidence?

Unlikely.

The fundamentals

Over the last 12–18 months, Sivers has announced:

All public PRs

• Ayar Labs → optical I/O / CPO manufacturing
• POET → optical engines + external light sources
• Tier-1 (undisclosed) → high-volume laser arrays MoU
• WIN Semiconductors → InP manufacturing scale
• O-Net / Enablence → ELS integration
• Jabil ecosystem → 1.6T hyperscaler pathway
• LiDAR → volume ramp (~$56M over 4 years from Q4 2026)

Additionally:

• institutional funding secured
• refinancing completed
• pathway opened for potential US dual listing

What this really shows

Individually: Small headlines.

Together: A position across the emerging AI photonics stack bottleneck

The shift

AI infrastructure is evolving:

800G → 1.6T (SIVE inside) → 3.2T (TBD)

CPO (SIVE Inside)

And the bottleneck is moving:

Compute → optics → lasers

Sivers is not betting on one architecture.

It is:

• supplying the laser layer
• participating across ecosystems
• positioned where all architectures converge

Final Thoughts

What ever you think about SIVE, this is quite an exiting story, for any company. I wish more of Swedish media and people in general would be looking into the interesting side of this story and try to be happy for a possible new international success from our little country, in the Semiconductors industry this time. Which I guess, is quite rare even if Norway has done that with Nordic Semiconductors already. More research are welcome even in Sweden.

It's been a few interesting weeks looking forward to see what comming next.

Anders Storm

This is not investment advice.

SiVE vs POET — Two Ways to Play the Optical AI Bottleneck

Anders Storm — Wed, 22 Apr 2026 12:37:54 GMT

Two partners, both in AI Photonics, market is exploding — what’s the difference

The optical layer in AI datacenters is moving from electrical → optical → co-packaged optics (CPO).

That shift is creating a massive opportunity, but also a clear divide between business models.

On one side:
Sivers Semiconductors (SiVE) → shipping, scaling, already generating revenue (mainly NRE still, and are Pre-CPO photonics ramp.

On the other:
POET Technologies → pre-scale, high upside, maybe higher uncertainty, but also closer to Nvidia, just Marvell in between then in the stack. Bigger TAM in this area than SiVE.

Both are exposed to the same megatrend:
→ The AI datacenter becoming optical
→ The optical layer becoming laser-constrained

But the way they monetize that trend same but still very different. SiVE is more a Swiss Army Knife and POET the Single Blade.

The numbers (2025 snapshot)

What the market is actually pricing

This isn’t about “which is better” — it’s about what you’re paying for and what risks you might see and total upside.

SiVE

You’re paying for:

Design-in with Ayar / Jabil /POET
Participation in 1.6T and CPO
A path to scale with real customers via Win Semiconductors fab
Also has traction within LiDAR, SATCOM, 5G/ 6G, DoD, biometrics

Risk: Qualification, Yield (gross margins), slower ramp than expected
Upside: operating leverage, many verticals, FAB less with GlobalFoundries and WIN that can volumes scale quickly.

POET

You’re paying for:

Optical integration platform 10x ASP vs SIVE InP laser arrays.
Potential socket wins in future architectures
Early validation from Marvell (Celestial AI) → supplier to Nvidia ($2B investment)
Foxconn / Luxshare (pipeline)

Risk: commercialization delays
Upside: massive if platform becomes standard via Celestial AI Photonic Fabric. Acquired by Marvell for +£3,2B

TAM — same market, different part of the stack

The big total number: ~$70B Optical AI opportunity

CPO (Co-Packaged Optics)

~$12-15B SAM

800G and 1.6T Pluggables

~$35-40B SAM

Full stack.

Where each company plays

SiVE

Laser arrays inside Ayar Labs SuperNova
Positioned at the light source bottleneck
Content: ~$50–150 per unit

TAM exposure = laser share of CPO (multi-billion approx B1.5-2B SAM). Total TAM ~$15B. from all veriticals Wireless, Lidar etc

POET

Optical engines / integration platform
Potential to simplify or replace parts of optical stack

TAM exposure = $11B system-level photonics 2028 (10x larger in AI photonics vs SIVE). Due to POET is down streaming SIVE.

The real difference

SiVE is selling the constraint InP
POET is selling the solution.

SiVE → you need lasers no matter what architecture wins
POET → you win if your architecture becomes the standard via $MRLV and Nvidia.

Market psychology

SiVE trades more like a scaling busy, but ramp soon
POET trades like a venture-style option, but also ramps very soon

They will also ramp together with POET 7-9x ASP.

What to watch

SiVE

Volume ramp with Ayar Labs, Jabil, POET, SATCOM and LiDAR.
Hyperscaler design wins
Revenue scaling

POET

PO → revenue conversion
Celestial AI ramp
Marvell relationship

Summary

Sivers Semiconductors:

• Laser arrays (InP, DFB) – core to CPO and external light sources

• mmWave / RF – wireless backhaul, satellite communications

• Photonics + Wireless – multiple verticals, multiple markets

→ A portfolio with several legs to stand on

→ Exposure to both AI data centers and communications infrastructure and DoD

POET Technologies:

• Optical Interposer – integration platform

• Infinity – optical engines (800G / 1.6T)

• Starlight / Blazer – external light sources

→ A concentrated bet on next-generation optical architecture connection directly to Marvell and Nvidia

→ All optionality depends on that architecture winning

The AI datacenter is becoming optical.
The optical layer is becoming laser-constrained.

$SIVE – the Swiss Army knife vs $POET – the single blade. What would you prefer to invest in one of them or both?

Anders Storm

The $70B Optical AI Opportunity — And the Hidden Laser Bottleneck

Anders Storm — Tue, 21 Apr 2026 09:25:28 GMT

📌 Table of Contents

The shift investors are missing
The $70B photonics market
Scale-out vs scale-up: the architectural break
Why CPO changes everything
The hidden multiplier: lasers
GPU math → real TAM
Coherent confirms the bottleneck
Where the real leverage sits
Final takeaway

1. The shift investors are missing

AI is no longer just a compute story.

It’s a connectivity story.

At its 2026 Technology Innovation Briefing, Coherent laid out something very important:

Photonics is becoming the fabric of AI datacenters.

This is not incremental. This is a full infrastructure transition.

2. The $70B photonics market (“module” TAM)

Coherent breaks the market into two parts:

Existing growth engines (~$50B SAM)

Pluggable transceivers (800G → 1.6T → 3.2T → 6.4T)
DCI (ZR/ZR+)
Transport & optical components

~$50B+ SAM by 2030

New growth engines (+$20B SAM)

Co-packaged optics (CPO/NPO)
Optical circuit switching
Multi-rail systems
Thermal solutions

+$20B incremental SAM

3. Scale-out vs scale-up: where things break

This is the most important structural shift.

Scale-out (today)

Pluggables dominate
Flexible architecture
10m → 10km links

Scale-up (future)

Copper hits limits
Power becomes critical
Density becomes everything

→ This forces a transition to CPO

4. Why CPO changes everything

This is not just a new product. It’s a system redesign. Coherent states it clearly:

“CPO is a SAM expansion as optics replace copper”

And quantifies it:

$15B+ CPO SAM by 2030

But the key insight is deeper: CPO increases component density per system. Especially for lasers

The hidden multiplier: laser Arrays

Look at what CPO introduces:

External laser sources (ELS)
InP CW lasers

What changes structurally?

Architecture - Laser demand - Pluggables:
1 per module

CPO
Shared + distributed - AI clusters - Arrays with 16 wavelengths.

Laser unit demand scales faster than optics revenue, 64x SuperNova light sources + heatsink assemblies” (Ayar Labs / Jabil rack design). Each array has 16 InP lasers.

6. GPU/CPO math → real TAM/SAM - InP SAM 2030

Using realistic assumptions (please note these are all

Deep dive Sivers SATCOM BFICs

Anders Storm — Sun, 19 Apr 2026 08:15:37 GMT

Introduction

One thing I’ve learned is to avoid going too deep into the technology stack when talking to investors.

Why? Because you scare investor off.

It’s a natural reaction — if you don’t understand what you’re investing in, the instinct is to say no.

This post is primarily for the tech crowd.

And for non-technical investors, instead look at who made that decision to use Sivers wireless BFICs:

Ericsson
Raytheon
All.Space
And many more..

These players have evaluated multiple suppliers — and still picked Sivers. They know better than us why. Hence focus your analysis on execution and market risk instead.
Biggest risk will be qualification (Photonics mostly) and yield (both BUs), which is directly connected to gross margins.

Anyway, here we go. ⤵️

The Real Question

Why are Sivers’ SATCOM beamforming ICs (BFICs) competitive — and in many cases preferred?

This is not about one feature. It’s about a system-level advantage and the new SATCOM constellations that Sivers can adress

1. Power: The Core Constraint in SATCOM

In satellite communications, everything starts with power.

Longer distance
Better link margin
More reliable connection

Sivers BFICs deliver high output power per channel. (based on best in class design and RF chips technology and Fab partner Global Foundries - GF)

This allows:

Fewer antenna elements
Smaller terminals
Lower total system cost

More power per chip = less hardware per system

2. Efficiency: The Real Differentiator

Power alone is not enough.

Efficiency determines whether the system actually works in real-world conditions.

Sivers focuses on:

High RF efficiency
Lower power consumption
Reduced thermal load

This matters in:

Mobile terminals
Airborne systems
Defense platforms

High efficiency = lower energy, less heat, better reliability

3. Integration: Reducing System Complexity

Traditional RF systems require multiple discrete components.

Sivers integrates:

Power amplifiers
Low-noise amplifiers
Phase shifters
Beam control
Digital control

All into a single BFIC.

Result:

Fewer components
Lower cost
Faster development cycles
Higher reliability

4. Beamforming: The Actual Value Driver

Beamforming is the core capability in modern SATCOM.

Sivers enables:

Narrow, directed beams
Higher spectral efficiency
Reduced interference
Improved security
Market ledning bandwidth (Ka-band and Ku-band) in single chip

5. Multi-Beam Capability

Modern satellite systems require:

Tracking multiple satellites
Switching between orbits
Maintaining continuous connectivity

Sivers supports:

Multi-beam operation
Seamless handovers

Critical for multi-band simultaneously GEO, MEO, LEO constellations and next-gen networks for DoD secure make before break communication.

6. Manufacturability (Underrated)

Great tech is useless if it can’t scale.

Sivers BFICs are:

Built on RF-SOI platforms
Calibrated at chip level
Designed for high-volume production
Qualification on standard technologies form GF

This is not just lab performance — it’s deployable at scale

7. Deep-Tech Origin

The BFIC technology originates from:

MixComm
Advanced mmWave research at Columbia University
Many years of R&D
DARPA-backed programs

This is not the best of the best technology -state of the art technology.

The Key Insight

Sivers’ advantage is not one metric.

It is the combination:

High power
High efficiency
Deep integration
Beamforming performance - broad bandwidth
Highly integrated BFIC
Scalable manufacturing

That combination is what wins real-world SATCOM programs.

Final Thoughts

If you are not technical: Don’t try to reverse-engineer the chip. Instead, look at who is choosing it. Because in markets like this: The best validation for deep tech is not theory — it’s adoption and to look at who has adopted it.

Anders Storm

Sivers Semiconductors $SIVE - SATCOM Timeline: From RFQ Loss to Ramp 2020-2027

Anders Storm — Sat, 18 Apr 2026 10:59:38 GMT

Table of Contents

The Trigger: Losing the RFQ
Strategic Response: MixComm Acquisition
NRE Phase Begins
First System Validation: All.Space
European Traction: Thorium Space
Institutional Backing: ESA Programs
Expanding NRE Pipeline
U.S. Positioning: DoD & Ecosystem
The Current Phase: Pre-Ramp
Key Takeaway

1. The Trigger (2020–2021): Losing the RFQ

Before acquiring MixComm, Sivers was invited to an SATCOM beamformer RFQ.

They lost. To MixComm.

That moment exposed a clear gap in beamforming IC capability — and became the trigger for the next move.

2. The Strategic Response (2021): Acquiring MixComm

Sivers responded by acquiring MixComm.

This added:

Advanced BFIC technology
mmWave expertise
U.S. presence DoD access

3. The NRE Phase Begins

Following the acquisition, Sivers began securing NRE (Non-Recurring Engineering) engagements. Read my other post about the significance of NRE to fund a business in EU due to lack of capital. The long game is always the balance of getting NRE funds and keep the IP to make standardize products. This is not a consulting business, this is funding to become a product-based business.

These signal:

Design-ins
Early validation
Future production potential

PR example: Biggest NRE project

4. First System Validation: All.Space

One of the most important milestones:

Sivers BFICs used in All.Space ground terminals

Advanced multi-beam SATCOM terminal
High BFIC content per unit
Real-world validation

5. European Traction: Thorium Space & Others

Sivers expanded into the European SATCOM ecosystem:

Thorium Space
Other EU partners

This builds:

Regional exposure
Defense relevance

6. Institutional Backing: ESA Programs

Sivers participates in ESA-backed (European Space Agency) programs:

Technology validation
Funding support
Strategic alignment

7. Expanding NRE Pipeline

Sivers continues to announce:

Multiple active NRE engagements
Expanding customer base with well-known names

Transition path: NRE → Design-in → Qualification → Production

8. U.S. Positioning: DoD & Ecosystem

Through MixComm and U.S. presence:

Access to DoD ecosystem
Alignment with CHIPS Act
Partnerships with major players
Entering Asien market via Doosan

Looking across the timeline:

Technology gap → solved
Validation → achieved
Ecosystem → established
Pipeline → expanding
End customer Qualification and approval by US Army

What remains: Production ramp and volume production!.

10. Key Takeaway

This is not a new story. It is a multi-year build-up now approaching commercialization.

NRE → Design-in → Qualification → Ramp →Production

Sivers has already moved through the first 3 stages. Sivers management estimate $1B SAM in Q4 webinar Februari 2026. Full deal flow below:

Final Thought

The market focuses on when revenue shows up.

But in SATCOM: The timeline tells you when the ramp is coming and Product sales has increased over the last years to support customer qualification.

Wireless BU is 2x Photonics today with $23m in Net sales 2025 16% Product and 84% NRE. Where NRE shall be seen as investments in Future products sales funded by the customer. This is “To have your cake and eat it too”!

2025 sales by BU:

Sivers seems to approach the ramp phase within SATCOM and AI Photonics at the sametime.

Anders Storm

Former CEO of Sivers for 8 years.

Disclaimer: This newsletter is NOT investment advice. This is my analysis based on publicly available information; not investment advice. I have no current relationship with the company. The aim of this newsletter is to provide subscribers with basic and simplified insights about the semiconductor industry and its participants. Articles will be published semi-frequently.

Sivers Semiconductors: More Than AI Photonics — The Overlooked SATCOM Engine

Anders Storm — Fri, 17 Apr 2026 19:08:23 GMT

Before the acquisition of MixComm in 2021, SIVE was invited to an SATCOM beamformer RFQ the year prior.

Sivers lost that RFQ — to MixComm.

That loss became a catalyst.

One year later, Sivers acquired MixComm, effectively closing a critical technology gap and positioning itself directly in the SATCOM and defense RF stack.

Two Businesses. Two Different Curves.

Most investors today view Sivers through one lens: AI photonics and optical interconnects. That narrative is compelling and justified. But it is incomplete.

Sivers today represents a combination of:

Photonics (AI / optical interconnects) → high upside, non-linear growth

SATCOM / Wireless (defense, satellite, 5G/6G) → steady growth, high barriers, recurring demand

The market is pricing mostly the first today. It is largely overlooking the second somewhat, due to less knowledge around this area. Let's dig in ⤵️.

Why SATCOM Matters Now

The global SATCOM landscape is shifting:

Defense budgets are rising
Satellite networks are becoming mission-critical infrastructure
Demand for secure, resilient communication systems is accelerating

SATCOM is no longer niche, it is becoming a core layer of national security and digital infrastructure.

Where Sivers Fits Today

The MixComm acquisition was not opportunistic, it was strategic. Not just getting access to customers, but also access to best of bread IP for high-frequency beamformers and US presens which also has secured access to U.S. CHIPS and Science Act, with partners like Raytheon, Ericsson. Becoming important for national security.

It enabled Sivers to compete in:

Phased array antennas
Satellite ground terminals
Defense communication systems

These markets require:

High-frequency RF expertise
Reliability under extreme conditions
Long qualification cycles
This creates high barriers to entry and sticky customer relationships

The “Overlooked” Opportunity

While photonics defines the upside narrative, SATCOM provides:

visibility
stability
execution credibility
Access to DoD with high margins

Revenue possibilities in SATCOM

Estimated for 2027-2030 (This is my analysis based on publicly available information; not investment advice. I have no current relationship with the company.)

Credo $CRDO Acquires DustPhotonics for $750M — What This Deal Means for InP Players Like $SIVE

Anders Storm — Tue, 14 Apr 2026 19:26:19 GMT

At first glance, this looks like another step toward vertical integration in the optical stack, combining DSP, connectivity, and photonics into one platform.

However the deeper implication is more interesting. This deal does not remove the key bottleneck in AI infrastructure. Even if DustPhotonics planed to use of the shelf Lasers. Because while silicon photonics can route and process light, it still depends on something it cannot generate efficiently: The laser. And increasingly, that laser is based on Indium Phosphide (InP).

The Deal: Paying for Positioning, Not Revenue

Credo is acquiring DustPhotonics for approximately $750M (~ +$150m in shares). DustPhotonics generates an estimated $20–35M in annual revenue. That implies a multiple of roughly:

20x to 35x EV/Sales

This is not a valuation based on current sales. It is a valuation based on future positioning in the optical stack. Credo is not buying revenue. It is buying:

Silicon photonics capability
Optical engine design expertise
Time-to-market in 800G and 1.6T architectures

And importantly, the ability to participate in the next phase of AI infrastructure.

What Silicon Photonics Can — and Can not do

Silicon photonics is essential.

It allows:

Routing of optical signals
Modulation of data
Integration with CMOS processes

But it has a fundamental limitation. Silicon cannot efficiently generate light.

Hence:

Every silicon photonics system still requires an external laser source. At scale, and at the required wavelengths (1310nm / 1550nm), that source is almost always Indium Phosphide.

Why This Matters More After the Deal

This is the key point. The more the industry invests in silicon photonics:

The more optical engines are deployed
The more data is moved optically
The greater the demand for laser sources

In other words: More silicon photonics does not replace InP. It increases the dependency on it.

The Valuation Signal

This deal sends a clear message:

The market is willing to pay 20–35x sales
For companies positioned in the optical stack
Before full revenue scaling is visible

At the same time:

Established players like Lumentum are trading at 30x multiples
Optical assets are being repriced across the board

Yet smaller InP-focused players remain:

Early
Less understood
Discounted relative to the broader trend

$SIVE is trading at 1/2 this approximately 15x sales. This 3-6 months old slide from Sivers CEO, clearly shows the path for $SIVE and today’s deal under pins the multiples for AI Photonics and InP Laser companies. Pre-CPO sales ramp and volumes.

In Summary

Credo’s acquisition is not just about silicon photonics. It is about securing a position in the future of data movement. And in that future:

Optical interconnects scale
Silicon photonics expands
But the bottleneck remains the same

The laser. And increasingly: Indium Phosphide (InP)

Trying to quantify the market for DFB arrays on 4” waffers. Assuming 2500 arrays per waffer (yielded). $50-150 per array.

64 SuperNova per rack. With a large 100 rack AI cluster that is ~6,400+ SuperNova modules.

This gives the Following estimate:

Why Indium Phosphide (InP) Is Becoming the Most Important Material in AI Infrastructure

Anders Storm — Mon, 13 Apr 2026 18:37:03 GMT

Everyone has been focused on GPUs

But the real bottleneck in AI infrastructure is no longer compute. It is data movement.

As AI scales, moving data between chips, racks, and clusters is becoming harder, more expensive, and more power-hungry than the compute itself.

In June 2020, two years before the AI boom , and $SIVE (+25%) jointly founded the CW-WDM consortium to standardize WDM CW sources in the O-band for AI, LiDAR, and silicon photonics applications. The MSA aims to define specifications for CW-WDM laser sources to support emerging 8, 16, and 32 wavelength grid configurations, which are critical for high-density co-packaged optics (CPO) and optical computing. Basically, what the whole market is about today, six years later, is solving this “data movement challenge.”

The specification was agreed in 2022, and all three chose to use Indium Phosphide (InP) as the main laser technology to solve this bottleneck. They are all pushing one material to the center of the entire stack.

This is what said about InP at OFC Lumentum Investor Briefing at OFC 2026

This shift has started to be priced in..

AI clusters are scaling at an unprecedented rate:

More GPUs per rack
More racks per cluster
Exploding east-west traffic

This creates a simple constraint:

Data must move faster than ever before.

That is why the industry is rapidly transitioning:

400G to 800G to 1.6T to 3.2T
Pluggable optics to Co-Packaged Optics (CPO)

This is not a linear upgrade.

It is a reset of the optical layer of computing.

Why optics suddenly matter

Inside modern AI data centers:

Copper interconnects hit physical limits
Electrical signaling becomes inefficient
Power and heat become dominant constraints

So the system shifts toward optical interconnects.

This introduces a new requirement:

A high-performance light source and this is exactly where InP Lasers fits in.

The physics problem

Fiber optics only work efficiently at specific wavelengths:

1310 nm
1550 nm (O-band)

These are the points where signal loss is lowest and efficiency is highest.

This is exactly where InP lasers operate naturally.

Why silicon cannot solve this alone

Silicon is the foundation of computing. But it has a fundamental limitation:

It cannot efficiently generate light.

So even in silicon photonics systems:

Silicon handles routing and modulation
External lasers are required

Those lasers are increasingly based on InP.

Why InP wins

Native wavelength advantage

InP operates naturally at 1310 nm and 1550 nm, aligning perfectly with fiber optics.

Built for 1.6T and beyond

At higher speeds, signal integrity becomes critical and noise tolerance drops.

InP provides cleaner signals, higher bandwidth, and longer reach.

Designed for CPO environments

CPO places optics directly next to chips, creating high heat and density.

InP is more stable under these conditions than legacy solutions like VCSEL.

System-level efficiency

Optical interconnects now consume a meaningful share of data center power. Better lasers (InP) reduce energy per bit and overall system cost.

The end of “good enough optics”

VCSEL lasers dominated the last cycle because they were cheap, short-reach, and easy to scale.

At 800G, 1.6T, and 3.2T, they hit limits:

Operate at 850 nm, which has higher loss
Less efficient at longer distances
More sensitive to heat

The result is a clear industry shift toward InP and new players like $SIVE comes in with long history in DFB InP laser. Even so the big incumbents Lumentum, MACOM, COHERENT are aalso doing InP.

The data confirms the transition

2022: around 40 percent InP
2025: around 80 percent
2030: around 90 percent or more

This is a structural transition, not a cycle.

The hidden constraint: supply

InP is not like silicon.

Smaller wafers (2 inch, 3 inch, 4 inch, moving toward 6 inch)
Fewer fabs globally
More complex manufacturing

Supply is limited, demand is growing rapidly and yield is often clearly below the +90% we are use to in Silicon.

The winners: who controls InP

This is a concentrated market with a few key players.

Lumentum Holdings

A leading supplier of high-speed InP lasers used in 800G and 1.6T systems.

Strengths:

Proven performance at scale
Deep experience in telecom and datacom
Strong relationships with hyperscalers

Coherent Corp.

A major player in InP manufacturing and materials.

Strengths:

Large-scale manufacturing
Investment in 6-inch wafer scaling
Vertical integration

Sivers Semiconductors

Focused on InP lasers and optical front-end technologies.

Strengths:

Positioned in high-growth segments such as CPO and optical I/O
Works with foundry partners like WIN Semiconductors
Higher sensitivity to growth due to smaller scale

WIN Semiconductors

A key compound semiconductor foundry, patner with Sivers Semiconductors.

Role:

Provides manufacturing capacity for InP-based devices
Essential for scaling production across the ecosystem

The key insight

This is not just a technology story.

It is a supply-demand imbalance forming in real time:

AI demand is accelerating
InP capacity is limited
Barriers to entry are high

This creates structural aadvantages for the main suppliers of InP.

In summary

The biggest opportunities are often not where most people are looking. AI is no longer compute-limited. It is interconnect-limited.
And InP is becoming the material that enables the next phase of scaling, this journey has just started.

Anders Storm