Compute prices were already moving higher before this. AWS has raised prices for some of its GPU capacity twice this year, H100 contract pricing has strengthened and IREN recently said its contracted pricing continues to improve.

When the cost of building new capacity rises, the price of compute eventually needs to follow. However, not every neocloud benefits equally.

A company that has not ordered its GPUs will pay more for the equipment and then charge more for the compute. Its margin may not improve at all.

A company that bought its GPUs cheaply (before price hike) but already contracted nearly all of the capacity cannot fully benefit from higher market prices either.

The best position is owning hardware purchased at the old cost while still having capacity available to contract after compute prices move higher.

That is where IREN becomes interesting.

IREN secured purchase agreements taking its planned fleet to 150,000 GPUs, including more than 50,000 additional B300s scheduled for phased deployment throughout H2 2026.

Management bought early because time-to-compute is becoming more valuable. Customers do not just want GPUs. They want them delivered on time, with certainty.

These orders were signed months before Nvidia’s early-2027 price increases. Much of IREN’s 2026 fleet should therefore arrive below the replacement cost of comparable capacity delivered next year, while part of the revenue side remains open.

IREN bought the hardware before prices increased but will contract some of that capacity after the increase. That is what the investors should pay attention to.

There is another part of this thesis that I think is being overlooked.

Higher replacement costs also weaken the GPU depreciation bear case.

The bear argument assumes GPUs rapidly lose their value because Nvidia releases a new architecture almost every year. Blackwell replaces Hopper, Rubin replaces Blackwell and so on.

Current evidence is not supporting that argument yet.

CRWV recently rebooked H100 capacity at around 95% of its previous contract price. It has also contracted older A100 systems into 2029, despite that architecture first being released in 2020.

Important to note that an older GPU does not need to compete directly with Rubin to remain valuable. It only needs to handle workloads that do not require the newest and most expensive hardware.

If replacing that GPU becomes considerably more expensive, customers have even more reason to continue renting it.

That extends the useful life of the existing fleet, supports its residual value and makes aggressive depreciation assumptions harder to defend.

This also very important when it comes to financing.

Neoclouds use customer contracts and GPU-backed loans to finance enormous equipment purchases. Lenders care about what those GPUs will still be worth if a contract ends early or the borrower runs into trouble.

If older systems remain productive for longer while new systems become more expensive, the collateral becomes easier to underwrite. Stronger residual values should support better financing terms.

The benefit still differs across the three main neoclouds.

CRWV receives the largest residual-value benefit. Older GPUs remaining productive directly challenges the idea that its fleet becomes worthless after only a few years. However, around 96% of its revenue is already covered by long-term take-or-pay agreements. Most of its repricing opportunity therefore sits in renewals and future contracts rather than its existing book.

NBIS sits in the middle. It has deliberately kept some capacity available for shorter agreements, giving it more exposure to rising compute prices. Customer prepayments and its cash position also give it more room to hold that capacity while waiting for better pricing.

IREN has the cleanest exposure to the gap between 2026 equipment costs and 2027 replacement costs. It bought early, has GPUs arriving throughout 2026 and still has capacity available to contract after market prices have moved higher.

So in short, expensive memory/GPUs is not automatically bullish for every neocloud. It rewards the operators that ordered early, preserved some pricing flexibility and can convert higher replacement costs into stronger contracts.

For this setup: IREN > NBIS > CRWV

I have spent the past few weeks going through the latest Lumentum and Applied Optoelectronics earnings because optical stocks have moved so quickly that I wanted to understand how much of the move was backed by fundamentals.

My conclusion is that the demand is unreal (same as Memory sector), the laser bottleneck is getting worse and the companies with “qualified manufacturing capacity” capacity should continue to benefit. However, the market has already priced a lot of this into the sector, so choosing between $LITE, $AAOI and $SIVE comes down to how much execution risk you are willing to take.

What is qualified manufacturing capacity?

“Qualified manufacturing capacity” means production that customers have tested and approved.

A company cannot simply build a factory and immediately supply lasers to Microsoft, Meta or NVIDIA. The customer must confirm that the lasers:

• Meet performance and reliability requirements

• Work consistently across thousands of units

• Can survive heat and long-term use

• Are produced with acceptable yields

• Can be delivered at the required volume

New equipment or factories may also need to be approved again. That is why adding laser supply takes time.

Sivers may have the right technology, but it still needs customers to approve its products and prove it can manufacture them consistently at scale.

Why lasers?

The AI buildout is moving from one bottleneck to the next.

It started with GPUs. Then it moved to power, data centres and cooling. The next problem is moving data between an increasing number of GPUs without consuming too much power or slowing the entire cluster down.

Today, most data-centre optical connections use pluggable transceivers. These sit at the edge of a network switch and convert electrical data into optical data. The industry is now moving from 800G towards 1.6T, which means twice as much data needs to pass through each connection.

As speeds increase, power consumption, heat and physical space all become bigger problems. This is why the industry is also developing co-packaged optics, where the optical components are moved much closer to the switch chip.

Whether the industry uses pluggable transceivers, co-packaged optics or optical I/O, it still needs a light source. That is where high-power indium phosphide lasers come in.

These lasers are difficult to manufacture. Customers need the correct power, wavelength, reliability and efficiency, but they also need suppliers capable of producing them at high yields and in very large volumes. AAOI said the equipment required to expand laser production can take 21 to 24 months to order, install and qualify.

This is not a market where supply can appear overnight simply because prices are rising. The same pattern we are witnessing in the memory sector.

What AAOI’s earnings told us!

AAOI is useful because it manufactures both the lasers and the completed optical transceivers. It can see demand across the entire product.

The company reported Q2 revenue of $191.9 million, up 86% year over year and 27% sequentially. Its 800G volumes more than doubled from the previous quarter, and management guided for Q3 revenue of between $255 million and $290 million. At the midpoint, that would be another 42% of sequential growth.

However, the most important information came from the earnings call.

AAOI currently has the capacity to produce close to 200,000 units of 800G and 1.6T products per month. It expects this to rise to around 650,000 by the end of 2026 and more than 930,000 by the end of 2027.

Even after that expansion, management said customer demand was still 20% to 40% higher than the capacity being built. Customers are regularly asking AAOI to accelerate deliveries.

AAOI also described lasers as the biggest bottleneck in the transceiver business.

The CPO comments were even more important.

AAOI said several major CPO customers had already qualified its high-power lasers. The problem is that AAOI cannot currently produce enough lasers to supply those customers while also supporting its own transceiver business.

The company has therefore chosen to use most of its existing laser capacity internally. Management expects high-volume CPO laser production to become more meaningful around late Q3 2027 after the expansion of its Texas manufacturing facilities.

This is important for the wider sector. AAOI has qualified products and customers ready to buy them, but it still cannot participate properly in the first generation of CPO deployments because it does not have enough manufacturing capacity.

The largest AI companies are trying to secure laser supply while the manufacturers are saying demand is already ahead of production.

AAOI is still not without risk. Data-centre revenue was $107.7 million in Q2, while CATV contributed $80.6 million. Its top ten customers accounted for 99% of revenue, and the company is spending heavily to expand production.

At a market cap of roughly $12.3 billion against management’s expectation of around $1.1 billion in 2026 revenue, AAOI is trading at approximately 11 times forward sales. The company now has to deliver the production ramp that investors are paying for.

What Lumentum’s earnings told us!

Lumentum showed what this demand can do to the financials of an established supplier.

The company reported fiscal Q4 revenue of $1.01 billion, up 109% year over year and 24.5% sequentially. Non-GAAP gross margin reached 50.4%, while non-GAAP operating margin increased to 36.6%.

Lumentum then guided for fiscal Q1 revenue of between $1.225 billion and $1.275 billion, with an operating margin of roughly 40%.

This is operating leverage. Revenue is increasing quickly, but profits are growing even faster because Lumentum already has the products, manufacturing base and customer relationships.

The company is seeing record demand across high-speed lasers, optical transceivers and optical circuit switches. It has also started shipping 1.6T transceivers and received an initial order for external laser source modules.

NVIDIA’s agreement with Lumentum is another important part of the thesis.

NVIDIA invested $2 billion in the company and entered into a multibillion-dollar purchase agreement that included rights to future laser capacity. The money is helping Lumentum expand manufacturing in the United States.

NVIDIA is effectively helping finance more supply while securing part of that supply for itself. Companies do not make agreements like this for components they believe will be easy to source.

Lumentum is therefore the most established way to invest in this theme. The business risk is lower because the company is already producing at scale and generating strong margins.

The valuation risk is much higher than it was a year ago. Lumentum is now worth roughly $89 billion against $3 billion of fiscal 2026 revenue. The market is already expecting several years of exceptional growth.

Note that lower execution risk does not automatically mean a cheap stock.

Why I own Sivers!

Sivers is the highest-risk company of the three, but it also has the greatest potential upside if it executes.

The company’s Photonics division develops high-power indium phosphide DFB lasers and multi-wavelength laser arrays. Its 70 mW and 100 mW lasers are designed for 800G and 1.6T pluggable transceivers, while its multi-wavelength arrays are aimed at co-packaged optics and optical I/O.

The future optical systems will need multiple wavelengths of light in a very small space. Integrating several lasers into one array can reduce the number of alignment and packaging steps, which should lower cost and make large-scale production easier.

Sivers also uses wafer-level coating and testing through its InP100 manufacturing platform. In theory, this should improve yields because the company can test the lasers before the wafer is cut into individual chips.

The technology is not new or unproven in a laboratory.

Sivers has worked with Ayar Labs for several years, and its laser arrays have powered Ayar’s SuperNova external light source. It is also working with Jabil on a 1.6T linear-receive optical transceiver that uses Sivers DFB lasers.

More recently, Sivers announced a $3.4 million programme with SemiNex to develop high-power external laser sources, multi-wavelength DFB arrays and optical amplifiers for CPO. Customer samples and early production are expected in the second half of 2027.

This is the part of the Sivers thesis that attracts me.

AAOI has already told us it cannot make enough qualified lasers for current CPO customers. Lumentum is receiving billions of dollars from NVIDIA to expand its own capacity. Sivers owns an indium phosphide platform, has existing products and is already working with companies developing the next generation of optical systems.

Sivers does not need to replace Lumentum or AAOI. It needs to qualify its products with several large customers and become an additional source of supply.

However, there is a large difference between having the correct technology and generating meaningful revenue from it.

Sivers reported Q1 group revenue of SEK 61.9 million. The Photonics division generated only SEK 17.8 million, down 32% year over year, and reported an EBITDA loss of SEK 7.7 million.

The company’s $799 million opportunity pipeline also covers both Wireless and Photonics. Note - A pipeline is not backlog, and backlog is not revenue until the company manufactures and delivers the product.

For every Sivers programme, I am watching three stages:

1. Does the product complete customer qualification?

2. Does that qualification become a commercial order?

3. Can Sivers manufacture the product at the required volume, yield and margin?

The market often prices the third stage as soon as the first stage is announced. That is where the risk comes from.

Sivers has reduced some of the financing risk. In July, the company raised SEK 700 million at SEK 57 per share through a placement that was multiple times oversubscribed. The raise resulted in only 3.3% dilution and will be used to expand indium phosphide laser and optical-amplifier capacity.

This was a smart use of the higher share price. Sivers raised more than twice its entire 2025 revenue while giving up only a small percentage of the company. It now has the capital to order long-lead equipment before the expected 2027 production ramps.

That removes the immediate question of how the expansion will be funded.

The valuation problem

Sivers is currently valued at approximately SEK 13.4 billion. The entire company generated SEK 306.6 million of revenue in 2025, which means it trades at more than 40 times trailing sales.

The Photonics division generated only SEK 93.4 million of that revenue.

Sivers is a real InP manufacturer with legitimate technology and credible partners. The problem is that the valuation already assumes those partnerships will become large production orders. Lumentum is selling at scale today, while some of Sivers’ most important new optical programs are still targeting customer sampling and early production in the second half of 2027. At roughly 44 times trailing group revenue, Sivers has almost no room for qualification delays, manufacturing problems or customer losses.

How I rank the three:

Lumentum is the lower-risk option. It has already proven the demand, manufacturing and margin story. The main risk is that the valuation already reflects a lot of future growth.

AAOI is the medium-risk option. It has good revenue, qualified products and more visible production ramps, but it still has to execute an enormous manufacturing expansion while managing customer concentration and heavy capital spending.

Sivers is the high-risk option. It has the smallest existing revenue base and the longest distance between today’s programmes and large-scale production. It also has the most upside if it becomes a qualified volume supplier to the AI optical market.

The way I see it, Lumentum is already monetising the bottleneck. AAOI is building as quickly as possible to monetise more of it. Sivers is trying to prove that it deserves a place in the supply chain.

I own Sivers because I am willing to accept more execution risk for more potential upside. Due to the high risk factor, I started with a very small position.

I am looking forward to the next Sivers earnings report. I want to see how the new capital is being deployed, whether Photonics revenue is beginning to improve and whether management can provide clearer timelines for qualifications, orders and volume production.

Sivers has to prove it can turn that shortage into revenue.

July was probably the toughest month I have experienced since I started investing. The S&P 500 finished almost flat, but anyone holding AI infrastructure, memory, optical, semiconductor or data-centre stocks had a completely different month. Some of the most popular AI stocks fell by 20%, 30% and more than 40% within a few weeks.

July is normally a strong month for the stock market. The S&P 500 had finished higher in each of the previous 11 Julys and had not recorded a negative July since 2014.

At one point, my own portfolio was down around 20% for the year. By the end of July, I had recovered and moved back into the green. I was able to stay invested throughout the whole sell-off, while many investors using leverage had their positions closed before the recovery even started.

How the selling started

The first major shock came on July 1, when Bloomberg revealed that Meta was developing a cloud business to sell some of its AI computing capacity. Investors immediately worried that Meta had built too much infrastructure and could become a competitor to the neocloud companies it had previously used.

$CRWV fell 13.9% that day and $NBIS dropped 17%. The concern then spread across the wider data-centre trade, including $IREN and $APLD.

I thought the market took the Meta story too far. Meta was still expanding its infrastructure and later said it planned to increase its computing power to 14 gigawatts. Selling spare capacity allows Meta to earn money from infrastructure during periods when it is not being fully used. The market focused on the words “excess capacity” and treated them as proof that AI demand was weakening.

As July continued, the selling spread into memory, optics, networking and semiconductors. Concerns around cheaper Chinese AI models returned, investors started questioning the return on hundreds of billions of dollars of spending, and funds began reducing exposure to some of the most crowded trades in the market.

The important point is that these stocks were being traded as baskets. When investors became worried about AI spending, they sold neoclouds together. When concerns moved towards memory, they sold Sandisk, Micron and SK Hynix. When funds reduced their exposure to data-centre networking, they sold optical and interconnect companies together.

The AI stocks that were hit

These are the actual returns from the closing price on June 30 to the closing price on July 31.

Memory

$SNDK: -46.6%
$MU: -28.7%

Sandisk went from $2,273.73 to $1,214.83, while Micron fell from $1,154.29 to $823.03. Sandisk was the worst-performing company in the S&P 500 during July, despite still being one of the best performers of the year.

Optical and interconnects

$AAOI: -36.3%
$COHR: -33.4%
$ALAB: -35.6%
$CRDO: -23.9%
$LITE: -16.8%

These companies supply the equipment used to move data between GPUs, servers and data centres. The long-term demand may still be strong, but the trade had become extremely crowded and the stocks had already delivered huge gains. Once investors started reducing risk, the entire basket was hit.

$SIVE Semiconductors was also caught in the sell-off. The company announced a SEK700 million share issue at SEK57 at the start of July, and the stock ended the month at SEK30.74. That added dilution concerns at the same time the whole optical and photonics trade was weakening.

Neoclouds and AI data centres

$NBIS: -16%
$CRWV: -15%
$APLD: -22%
$IREN: -16%

Those returns include the huge rebound during the final two sessions. At their lowest points, the losses were much worse. $IREN fell to $28.93 on July 29 before jumping more than 30% the following day. $NBIS fell to $148.22 before gaining 27.1% on July 30. CoreWeave dropped to $60.82 before rebounding 21.5%.

What happened to SK Hynix and South Korea

South Korea became the centre of the crisis during the final week of July. The market had attracted huge amounts of retail money, much of it borrowed through margin accounts and leveraged single-stock products.

On July 28, the KOSPI fell 10.84%. SK Hynix dropped 14.7% and Samsung fell 14.4%. The following day, the KOSPI fell another 6% after SK Hynix reported a sixfold increase in profit that still fell short of the expectations built into its share price. SK Hynix closed down another 9.6%.

The KOSPI had lost almost 40% from its June peak, and as much as $2.18 trillion had been removed from the value of South Korean shares. Brokers were forcibly closing leveraged positions, which added more selling and caused further liquidations.

Then SK Hynix jumped by almost 30% on the Korean exchange on July 31, reaching the daily limit. That came too late for investors whose positions had already been liquidated.

I have left the $SKHY ADR out of the monthly percentage list because it only began trading in the United States around the start of July and traded at a large premium to the shares in Seoul. Its daily moves were also very different from the Korean-listed shares. The important part is the volatility. Investors experienced enormous gains and losses within days.

Leopold owned the same basket

Leopold’s Situational Awareness fund was heavily exposed to the same stocks that were falling across memory, energy and AI infrastructure.

Its latest public filing covered the portfolio at the end of March, so it does not show every position held during the July liquidation. It does show the structure of the trade. The filing contained 42 positions worth approximately $13.68 billion.

The main disclosed long positions included:

$BE: Bloom Energy
$SNDK: Sandisk
$CRWV: CoreWeave
$IREN: IREN
$CORZ: Core Scientific
$APLD: Applied Digital
$RIOT: Riot Platforms
$CLSK: CleanSpark
$BTDR: Bitdeer
$BITF: Bitfarms
$HIVE: HIVE Digital
$SEI: Solaris Energy Infrastructure
$TE: T1 Energy
$WYFI: WhiteFiber
$SHAZ: SharonAI

The fund also owned call options in Bloom Energy, Sandisk, CoreWeave, Micron and Taiwan Semiconductor. At the same time, it held large put positions against the semiconductor ETF, Nvidia, Broadcom, AMD, Oracle, ASML, Intel, Micron and TSMC. The portfolio was therefore more complicated than simply buying every AI stock, but the largest long positions were heavily concentrated around physical AI infrastructure.

Several of those disclosed long positions suffered large July losses at some point:

$SNDK: -46.6%
$SEI: -36.1%
$BTDR: -33.7%
$BE: -32.0%
$CRWV: -27.9%
$APLD: -26.6%
$RIOT: -26.3%
$HIVE: -22.3%
$IREN: -19.5%
$CORZ: -19.0%
$CLSK: -5.4%

The losses were happening across almost every part of the long portfolio at the same time. Diversification provides little protection when nearly every position is connected to the same investment theme.

Leopold had identified one of the biggest investment opportunities of our generation, but he also used a huge amount of leverage to increase his exposure. When prices moved against him, lenders demanded more collateral and the fund started selling positions.

Those sales placed even more pressure on the same stocks, leading to further losses and more selling. Leopold compared the situation to a bank run because weakness in the portfolio created even more weakness.

Situational Awareness lost 67% during July. It remained around 80% higher for the year because its gains before July had been so large, but the fund had to remove all leverage and sell most of its public stock portfolio to Citadel.

The $45 billion figure represented the fund’s value at an earlier peak. The fund retained private investments, including its Anthropic stake.

The clear part is that tens of billions in value disappeared and Leopold lost control of most of the public portfolio. He may still be right about AI infrastructure over the next decade, but leverage gave him far less time than his investment thesis required.

The Citadel part

During the final week of July, Citadel Securities argued that the Federal Reserve could deliver a surprise interest-rate increase. Higher rates place pressure on growth stocks because investors become less willing to pay high valuations for earnings expected many years into the future.

AI stocks were already falling heavily, the possibility of higher rates added to the pressure, Situational Awareness reached the point where it had to sell and the Citadel hedge fund then bought most of its public stock portfolio.

Whether those events are connected is something readers can decide for themselves. The timing certainly raises questions.

Citadel Securities and the Citadel hedge fund are separate businesses, although Ken Griffin founded both.

My own view is that Citadel saw a fund under enormous pressure and took advantage of the opportunity. Situational Awareness needed to complete a deal quickly, while Citadel had enough cash to walk away. That gave Citadel far more power during the negotiation.

Citadel was also one of the few firms capable of looking through a portfolio worth billions of dollars and making a decision within hours. It bought exposure to the AI infrastructure trade after a huge sell-off and prevented the whole portfolio from being dumped directly into the open market.

The final days changed the mood

Microsoft reported Azure growth of 43% and said demand continued to exceed the capacity it had available. It also maintained plans for around $175 billion of capital expenditure during 2026.

Meta raised the lower end of its capital-expenditure guidance and now expects to spend between $130 billion and $145 billion this year.

Amazon then reported AWS growth of 37% and increased its 2026 investment plans to around $220 billion. Amazon said it still expected to face capacity constraints, with customers already reserving infrastructure well into 2027 and 2028.

Those results gave investors the evidence they had been waiting for. Cloud growth remained strong, hyperscalers were continuing to spend and demand for AI infrastructure was still increasing.

The same stocks that had been crushed then recovered sharply. Investors who had avoided leverage were able to participate in that recovery, while those who had been liquidated had already lost their positions.

What I learned from July

July showed me why survival matters so much in investing. At one stage, I was down around 20% for the year. A few days later, I was back in positive territory.

I already accept a lot of risk through concentration, with one stock representing more than half of my portfolio. Leverage would give a broker the power to sell that position during the exact moment when I may want to buy more. I would rather experience a painful drawdown while keeping control of my shares.

Every generation finds a new investing genius. Cathie Wood became the face of growth investing, Sam Bankman-Fried was treated as the future of finance, and Leopold became one of the most celebrated young investors in the world after producing extraordinary returns.

Their situations are very different. The similarity comes from how quickly people crown someone after a short period of success.

Warren Buffett has survived for more than 60 years. He has invested through wars, inflation, recessions, financial crises and market crashes. His real advantage has been consistency and a structure that allowed Berkshire Hathaway to survive when markets moved against it.

That is the lesson I am taking from July. Huge returns look incredible, but protecting those gains and remaining in the market for decades is the real achievement.

If you finished July in the green, congratulations. If you broke even or finished down while keeping your portfolio intact, you still have the chance to recover.

A lot of investors lost that opportunity because leverage made the decision for them.

HBM is a type of DRAM built by stacking memory chips vertically and placing them beside the processor. This gives the memory a much wider connection to the processor, allowing far more data to move at once while using less energy.

That is important for AI accelerators. A powerful GPU cannot work efficiently if it is constantly waiting for data. HBM keeps the processor supplied, allowing AI models to train and run faster.

Some AI chips use other forms of memory. However, HBM has become the main choice for the most advanced data-centre accelerators. SK hynix, Samsung and Micron are the three major suppliers.

The case for SK Hynix comes down to five areas: its HBM leadership, its position across the wider memory market, the financial impact of the current shortage, its valuation and the risks of buying during a powerful memory cycle.

1. SK Hynix has a clear lead in HBM.

This is not currently an equal three-company market.

According to IDC figures included in SK hynix’s US prospectus, the company controlled 56.4% of global HBM revenue in Q1 2026. Every other supplier combined held 43.6%. SK hynix therefore had a larger share than the rest of the market combined.

Its lead has been built across several product generations.

SK hynix was the first company to mass-produce HBM3. It then began supplying eight-layer HBM3E in March 2024 and started volume production of twelve-layer HBM3E later that year.

The company completed its HBM4 production preparations in September 2025, and large-scale production was underway by the beginning of 2026. It has since moved ahead again by sending customers samples of its next-generation HBM4E.

This gives SK hynix something its competitors cannot create overnight: years of manufacturing experience across several generations of HBM.

Its packaging method also helps. SK hynix injects a protective material between the stacked memory layers, helping control heat and keeping the layers securely connected. Exact manufacturing yields are not publicly available, so it would be wrong to claim a permanent yield advantage. However, the company’s ability to reach large-scale production first shows that its manufacturing process is working.

Plus, SK hynix already has the largest HBM position and has reached HBM4 production ahead of its main competitors.

HBM is also becoming more closely designed around the processor using it. SK hynix is working with TSMC to manufacture the logic component at the bottom of its HBM4 stacks. This allows the memory to be adapted more closely to the requirements of different AI processors.

Every new HBM product must still pass lengthy customer testing covering speed, power consumption, heat and reliability. Nvidia will continue using several suppliers because it does not want to depend completely on one company. However, these qualification requirements make an established supplier more difficult to replace.

SK Hynix versus Micro

Micron is the closest stock comparison because it gives investors direct exposure to the memory industry through a US company.

The strongest argument for SK hynix is its HBM position.

Micron reported $41.46 billion of revenue for its fiscal Q3 2026 and an 81.2% non-GAAP operating margin. SK hynix reported 52.6 trillion won of revenue in Q1 2026 and a 72% operating margin. These quarters cover different periods, but the results show that Micron is also performing extremely well.

The difference is HBM leadership. SK hynix controls 56.4% of the market and has already started large-scale HBM4 production. Micron has a much smaller HBM position, although it is growing quickly and remains a serious competitor.

Micron also remains the simpler US investment. It is an American company with ordinary shares listed on Nasdaq. SKHY is an American depositary share representing part of a Korean-listed company. Investors still have Korean currency, governance and geopolitical exposure.

The Nasdaq listing has reduced Micron’s convenience advantage. It has not completely removed it.

SK Hynix versus Samsung

Samsung presents a different comparison.

Samsung Electronics is a far more diversified company. Alongside memory, it owns a large consumer electronics business and operates a semiconductor foundry.

That diversification can protect the business when memory prices fall. It also means investors receive less direct exposure to HBM and AI memory.

Samsung lost valuable time during the HBM3E generation because it was slower to receive important customer approvals. SK hynix used that period to increase production, strengthen its customer relationships and take more of the market.

Samsung is now recovering. It is competing for HBM4 orders and can use its own foundry to manufacture customised logic base dies internally. That could become a greater advantage as HBM is designed more closely around each customer’s processor.

Samsung should not be dismissed. It remains the largest supplier of conventional DRAM and NAND, and it has the financial resources to compete aggressively.

The difference comes down to exposure. Samsung offers memory alongside foundry operations, smartphones and consumer electronics. SK hynix gives investors a more concentrated position in HBM, DRAM and NAND.

2. Strong across the whole memory sector.

SK hynix is much more than an HBM company.

In Q1 2026, it controlled 29.1% of total DRAM revenue, making it the second-largest supplier behind Samsung. It also controlled 18.5% of NAND revenue, again placing it second.

This matters because AI infrastructure needs more than HBM.

Servers also require large amounts of conventional DRAM to run applications and enterprise SSDs to store data. Through SK hynix and Solidigm, the company can supply memory and storage across several parts of the data centre.

This gives SK hynix broader exposure than SanDisk. SanDisk is mainly a NAND and storage investment. It does not manufacture DRAM or HBM, so it cannot directly participate in the highest-value part of the AI memory market.

SK hynix offers exposure across the memory stack. It leads HBM and ranks second in both conventional DRAM and NAND.

Its strongest position is also in the category experiencing the fastest growth and highest margins.

3. The memory shortage is creating enormous operating leverage.

Memory prices have risen sharply because demand is greater than the available supply.

TrendForce estimated that conventional DRAM contract prices increased approximately 93% to 98% during Q1 2026. SK hynix reported that its own average DRAM selling price increased by the mid-60% range, while its NAND selling price increased by the mid-70% range.

The figures measure slightly different things, but they point in the same direction. Memory prices have increased substantially.

This has an enormous effect on profit because memory factories have high fixed costs. The factories and equipment cost billions to build whether the chips are sold at high or low prices. Once production costs are covered, a large part of each price increase can flow into operating profit.

SK hynix’s Q1 results show this clearly.

Revenue increased 198% to 52.6 trillion won. Operating profit increased 405% to 37.6 trillion won, producing a 72% operating margin. The company also ended the quarter with 54.3 trillion won in cash and equivalents, against 19.3 trillion won of interest-bearing debt.

Management believes the shortage could continue for several years. CEO Kwak Noh-Jung expects the supply deficit to become most severe in 2027 and believes demand could continue exceeding supply beyond 2030. This is management’s forecast, so it should not be treated as guaranteed. However, it shows the strength of the demand currently being discussed with customers.

SK hynix is responding by expanding production. The chairman of its parent company has said it aims to double total wafer capacity over the next five years. New production is being added through M15X in Cheongju, the Yongin semiconductor cluster and advanced packaging facilities in South Korea and the United States.

The company is also building a $3.87 billion advanced packaging and research facility in Indiana. Production is expected to begin during the second half of 2028, supported by up to $458 million of direct CHIPS Act funding.

There is a much larger national expansion underway as well. Samsung and SK hynix plan to invest around 800 trillion won across four new fabrication sites in South Korea. The government wants to double the country’s memory-production capacity within five years.

That investment will take years to become productive. It is also spread across different types of semiconductor production, so it should not be described as $518 billion of immediate HBM capacity.

The bullish argument is straightforward. AI is increasing the amount and speed of memory required in each server. Current supply cannot meet that demand, prices are rising and SK hynix holds the strongest position in the most valuable part of the market.

4. The valuation discount needs to be understood properly.

This section requires an important distinction between the Korean shares and SKHY.

Before the US offering, SK hynix’s Korean-listed shares traded at approximately 5.5 to 5.8 times forward earnings. Micron traded closer to 6.7 to 7 times. On that comparison, the Korean shares looked cheaper despite SK hynix having a much larger HBM position.

Part of that difference comes from the Korea discount.

Korean companies have historically received lower valuations because of concerns involving corporate governance, chaebol ownership structures, capital allocation, shareholder returns, currency movements and geopolitical risk.

Limited international access also contributed to the discount, although it was never the only cause.

The Nasdaq offering improves access. SK hynix issued 177.9 million ADSs at $149 each, raising approximately $26.5 billion before costs. Ten ADSs represent one Korean common share, so the offering created the equivalent of 17.79 million new common shares. That diluted existing shareholders by approximately 2.5%.

US trading began on July 10, with the shares trading under the ticker SKHY.

The listing gives US investors dollar-denominated access and could broaden SK hynix’s institutional shareholder base. It could also support inclusion in additional funds or semiconductor benchmarks.

There is one major detail investors need to understand.

SKHY can trade at a different valuation from the Korean shares. Ten ADSs represent one Korean share economically, but the supply of US shares is limited. The prospectus also warns that Korean shares received after surrendering ADSs may not be eligible to be deposited back into the ADS program.

That makes it difficult for traders to remove the price difference between the two listings. Within days of the offering, SKHY was trading at a large premium to the underlying Korean shares.

The potential upside still comes from two areas. Earnings can continue rising if the memory shortage lasts, and the Korean valuation could move closer to Micron’s as more international investors gain access.

5. The downside is just as important.

Memory remains one of the most cyclical parts of the semiconductor industry.

A low P/E ratio can be misleading near the top of a memory cycle. Earnings rise quickly when supply is tight and prices are increasing. If new capacity eventually creates excess supply, memory prices and profits can fall just as quickly.

That is the main risk. SK hynix may look inexpensive because the market does not believe its current 72% operating margin can last.

There are already signs that price increases are becoming less extreme. TrendForce expects conventional DRAM prices to increase another 13% to 18% in Q3 2026 and NAND prices to increase 10% to 15%. Supply remains tight, but customers in PCs and smartphones are reaching the limit of what they can afford.

Customer concentration creates another risk. SK hynix’s largest customer accounted for 23.9% of its 2025 revenue. Losing a major order or receiving lower allocations could have a meaningful effect on results.

Competition is also improving.

Samsung has the resources, manufacturing scale and internal foundry capacity to recover HBM market share. Micron has reported record margins and is signing longer-term customer agreements designed to make its earnings more predictable.

The size of the planned capacity expansion creates a longer-term risk. SK hynix wants to double its capacity, Samsung is expanding and Micron is building facilities in the United States and Japan.

Those factories do not solve the immediate shortage. However, today’s high prices are paying for tomorrow’s supply. If demand growth slows before that capacity arrives, the market could move from shortage to oversupply.

SK hynix also has less protection than Samsung if AI infrastructure spending weakens. Samsung can rely on other semiconductor and consumer businesses. SK hynix remains heavily dependent on memory prices and data-centre investment.

The ADS valuation adds another layer of risk. SKHY can fall because the Korean shares decline, because the US premium contracts or because both happen together. Investors paying a large premium over the Korean shares are accepting a risk that does not exist to the same extent in the underlying business.

There are also wider risks involving Korean currency movements, corporate governance, geopolitical tension and export restrictions. SK hynix operates major facilities in China, which makes future US semiconductor controls particularly important.

Finally, the shares have already experienced a huge rally. Investors are buying a market leader with high expectations already attached to it.

▲ The bull case

• SK hynix controlled 56.4% of HBM revenue in Q1 2026, more than every other supplier combined.

• It has led across several HBM generations, has HBM4 in large-scale production and has already supplied HBM4E samples.

• Q1 revenue increased 198%, operating profit increased 405% and the operating margin reached 72%.

• Demand continues to exceed current supply, while management expects the shortage to last several years.

• It provides exposure across HBM, conventional DRAM, NAND and enterprise storage.

• The Nasdaq listing gives more international investors access to the company and could support a higher long-term valuation.

• Its strong cash position gives it the ability to invest heavily while continuing to return capital to shareholders.

▼ The bear case

• Memory is highly cyclical and current margins may be close to peak levels.

• Samsung and Micron are credible competitors with the resources to take HBM market share.

• Large capacity investments could eventually turn the shortage into oversupply.

• The new offering diluted existing shareholders by approximately 2.5%.

• SKHY can trade at a large premium to the Korean shares, creating an additional valuation risk.

• The largest customer represented 23.9% of 2025 revenue.

• Korean currency exposure, Chinese manufacturing operations, export controls and geopolitical risk remain.

SK hynix has the strongest position in HBM today.

Not Financial Advice. I own SK hynix shares.

START: £5,000.00

CURRENT VALUE: £8,571.55

RETURNS: 71.43%

Read more

But the deeper you go, the more you realise the whole system has to work together. A data center needs GPUs, memory, storage, networking, advanced packaging, substrates, power, cooling, software and companies that can actually build and operate the full stack.

This is why memory has become such a big part of the AI trade.

A GPU can be extremely powerful but if it cannot get data fast enough, the system slows down. That is why HBM has become so important. It is stacked memory that sits very close to the GPU and moves huge amounts of data at high speed.

That is why companies like Micron, SK Hynix, and Samsung matter so much. They are supplying one of the most important parts of the AI infrastructure stack. On its June 2026 earnings call, Micron said DRAM and NAND industry demand continues to significantly exceed industry supply, and it expects tight conditions to persist beyond calendar 2027. The CEO even said they currently do not have line of sight on when supply will catch up with demand, and only expect industry supply to improve gradually in 2028.

Note that HBM is different from normal memory because it is harder to scale. It needs more advanced manufacturing, more packaging complexity, and much closer qualification with customers. This is the part that has changed the structure of the industry. Micron said in December 2025 that it had completed price and volume agreements for its entire calendar 2026 HBM supply, including HBM4. By June 2026, it went further and announced 16 signed strategic customer agreements representing over $100 billion in contract value, with most agreements running roughly five years, from calendar 2026 through the end of calendar 2030. Customers are no longer negotiating quarter by quarter. They are locking in supply years in advance.

That is the first part of the trade.

The second part is understanding where the next bottleneck could be.

Part 1: The Memory Stocks

The easiest way to invest in this theme is through the companies that make memory directly. There are only three companies in the world that make HBM at scale, so let’s go through them properly.

$MU (Micron) is the cleanest U.S.-listed memory name. It gives investors exposure to DRAM, NAND, SSDs, and HBM. If someone wants direct memory exposure without buying overseas stocks, Micron is usually the first company to research.

The numbers show why. In its fiscal Q3 2026 quarter (reported June 24, 2026), Micron delivered roughly $41 billion in revenue, around four times higher than the same quarter a year earlier, with record gross margins in the 80s. Its HBM4 ramp is tracking about twice as fast as HBM3E did, and it has already shipped over $1 billion of HBM4 revenue. Micron forecasts the total HBM market growing from around $35 billion in 2025 to around $100 billion in 2028, which is roughly a 40% annual growth rate. Micron currently sits second in HBM market share behind SK Hynix, having overtaken Samsung during 2025.

One more detail investors should know. Micron announced in late 2025 that it is exiting the consumer market, winding down its Crucial-branded SSDs and RAM. Every wafer is being pushed towards AI and data center customers. That tells you where the money is.

SK Hynix is the most important HBM company in the world, full stop. If your thesis is mainly about HBM, SK Hynix has to be studied first.

Counterpoint Research had SK Hynix at 62% of HBM shipments as of mid-2025, and reports suggest Nvidia has allocated close to 70% of the HBM4 for its Vera Rubin platform to SK Hynix. Nvidia is by far its biggest customer. SK Hynix said back in October 2025 that its entire DRAM, NAND, and HBM supply for 2026 was already sold out. In Q1 2026 it posted a record operating profit with an operating margin around 72%, which is almost unheard of for a memory company. Its market cap crossed $1 trillion, and it overtook Samsung to become the biggest company on the Korean stock market.

SK Hynix has also said HBM3E is expected to remain the flagship HBM product in 2026, making up roughly two-thirds of shipments, with HBM4 gradually increasing its share through the year. So the current cash cow and the next generation both run through SK Hynix.

Until now, buying SK Hynix meant buying on the Korean exchange or through thin OTC shares. That changes on July 10, 2026, when SK Hynix lists ADRs on the Nasdaq under the ticker SKHY, raising around $29 billion in what is the largest-ever U.S. share sale by a foreign company, bigger than Alibaba and Saudi Aramco. One Korean share is split into ten ADRs. The proceeds are going into new fabs at the Yongin cluster, a new advanced packaging fab, and EUV equipment. In other words, the HBM leader is about to become as easy to buy as Micron.

Samsung is also part of the big memory group, but it is not as pure as Micron or SK Hynix. Samsung has memory, smartphones, displays, foundry, and consumer electronics. That makes it a broader business, which is both the appeal and the problem. When memory rips, Samsung moves less than the pure plays.

But Samsung is the recovery story of this cycle. It fell behind on HBM, dropping to around 17% share in 2025 after quality issues, letting Micron overtake it. Since then it has passed Nvidia and AMD qualification for HBM4, began shipments in early 2026, and reportedly negotiated HBM4 pricing at parity with SK Hynix after previously having to discount. Analysts expect Samsung’s HBM share to climb back above 30% in 2026. If you believe the laggard closes the gap, Samsung is the value angle.

Then you have names like $SNDK, Kioxia, Seagate, Western Digital, Nanya, and Winbond. These are not all the same. This is where beginners need to be careful.

A memory stock is not automatically an HBM stock.

$SNDK is about NAND flash and storage. It became a separate public company again after completing its separation from Western Digital in February 2025 and now trades under the ticker $SNDK. It makes its NAND through the Flash Ventures joint venture with Kioxia in Japan. Interesting detail: Sandisk and SK Hynix have signed an agreement to develop High Bandwidth Flash (HBF), which is basically the HBM concept applied to NAND. Sandisk’s 2026 supply is reportedly effectively sold out.

Kioxia is also a NAND name, spun out of Toshiba, listed in Tokyo, and it is preparing its own U.S. listing after posting record results. Kioxia has said its 2026 NAND output is essentially fully committed, driven by AI inference storage demand.

And NAND contract prices rose 33-38% in Q1 2026 and were projected to rise another 70-75% in Q2, actually outpacing DRAM for the first time this cycle, because AI inference needs enormous amounts of fast enterprise SSD storage.

Seagate and Western Digital are more connected to storage and hard drives. They benefit because data centers still need cheap nearline storage, and the HDD shortage is so bad it is pushing customers into high-capacity SSDs, which then feeds back into NAND demand. Nanya and Winbond are smaller Taiwanese memory names, more exposed to conventional and specialty DRAM, not HBM.

So the question is always: what part of memory is this company exposed to?

If the thesis is HBM, the main companies are SK Hynix, Micron, and Samsung. In that order.

If the thesis is broader storage demand from AI data centers, then names like Sandisk, Kioxia, Western Digital, and Seagate become worth researching too.

But the memory trade does not stop there.

Part 2: Where $PENG Fits In

$PENG has been my 2nd to 3rd largest position since May 2026.

Firstly, Penguin Solutions should not be placed in the same bucket as the big memory producers.

The reason $PENG is interesting is because it sits in the AI infrastructure and memory systems layer. It helps customers design, build, deploy, and operate AI factories. In June 2026, Penguin became an NVIDIA AI Factory Specialized Partner, an invitation-only designation for partners that can build and operate full-stack NVIDIA-based AI factory infrastructure for enterprises, sovereign AI, neoclouds, and hyperscale customers. It was also named Dell’s Americas AI Partner of the Year.

Important to note that Penguin has 25 years of high-performance computing history, and it designed, built, and now manages one of Korea’s largest GPU clusters, with over 1,000 NVIDIA Blackwell GPUs.

There is also a strategic anchor most people miss. In December 2024, SK Telecom made a $200 million strategic investment in Penguin. So a company in the SK family, the same group as SK Hynix, is backing the AI infrastructure layer.

That makes PENG a different kind of memory play.

The issue in AI is that the full system needs to use memory properly. Inference is especially important here. Training is about building the model. Inference is when people actually use the model. Every prompt, every answer, every agentic AI workflow, and every long-context task creates more pressure on memory. Penguin’s own framing is that inference workloads are roughly 70% memory-driven and only 30% compute-driven. GPUs end up sitting idle waiting for memory.

This is where Penguin’s MemoryAI KV Cache Server comes in. Announced at NVIDIA GTC in March 2026, Penguin says it is the industry’s first production-ready KV cache server using CXL memory technology, delivering up to 11 TB of memory (3 TB of DDR5 plus up to eight 1 TB CXL add-in cards) to help solve the “memory wall” in AI inference. The company says it is around 10x faster than NVMe-based approaches, is compatible with NVIDIA Dynamo (NVIDIA’s own KV cache offloading software), and is designed to improve latency, throughput, GPU efficiency, and time-to-first-token. It has already won a Tier-1 financial institution as a customer and received a large CXL card order from a generative AI company.

In normal language, PENG is trying to help companies get more useful work out of their AI systems.

If GPUs are expensive and memory is limited, customers need better ways to manage memory. That is why CXL, KV cache, memory pooling, and inference infrastructure are becoming more important. PENG is trying to solve the system-level memory problem around AI deployment.

That is why I would put PENG in its own layer. It is an AI infrastructure stock with a memory angle.

But be honest about the risk profile too. PENG is a much smaller company than Micron or SK Hynix, the stock is already up around 250% this year, and the thesis depends on execution and on CXL-based inference infrastructure actually being adopted at scale. It is a satellite position, not a core one.

So for investors, I would think about it like this.

Micron, SK Hynix, and Samsung make the memory.

PENG helps build and optimise the AI systems that need memory.

Those are different exposures, but they are connected to the same theme.

Part 3: The Substrate Layer

This is the part I think more investors need to study.

Substrates are extremely important. A substrate is the layer inside the chip package that helps connect the processor, memory, and the rest of the system. It helps move power and signals between everything.

The GPU does the work. HBM feeds the GPU. The substrate helps connect the package so everything can function properly.

As AI chips become more advanced, the substrate becomes more difficult to make. You need more layers, more connections, better materials, and tighter manufacturing. AI chip packages are moving from 3+3 layer substrate designs towards 11+11 and eventually 13+13. Ibiden describes IC package substrates as a key part that works with IC chips, with their value increasing as semiconductors become more advanced.

If AI needs more GPUs, and GPUs need more HBM, then advanced packaging becomes more important. If advanced packaging becomes more important, high-end substrates and substrate materials become more important too.

One of the key materials here is ABF, which stands for Ajinomoto Build-up Film. Yes, the MSG food company. Ajinomoto says ABF became the material of choice for almost all high-performance CPUs after it was first adopted by a major semiconductor manufacturer in 1999, and today it controls an estimated 90%+ of the ABF film market.

And this layer is now doing exactly what memory did a year earlier. ABF substrates went back into shortage in the first half of 2026, with the supply-demand gap expected to last into 2027. In May 2026, Ajinomoto notified substrate makers of a roughly 30% price increase on ABF film, effective Q3 2026. Resonac and Mitsubishi Gas Chemical raised prices on copper-clad laminate materials by a similar amount. Substrate contract prices are rising 5-10% in the second half of 2026, with spot prices up over 30%, and the big Taiwanese substrate makers are running at full capacity with lead times stretching out. Ajinomoto is even buying land for a third ABF plant, but it will not produce until the 2030s. That tells you how structural this is.

The substrate names I would research include Ibiden, Unimicron, Nan Ya PCB, Kinsus, Samsung Electro-Mechanics, AT&S, Shinko Electric, Kyocera, Toppan, and Ajinomoto.

They are not all pure plays. Some are more directly exposed to substrates. Some are broader component or materials businesses. That does not make them bad, but it means you need to understand the exact exposure before buying anything.

Some companies may benefit from ABF substrates. Some may benefit from glass core substrates. Some may benefit from materials. Some may benefit from packaging complexity. Some may be too diversified for the stock to move only because of this theme. Ajinomoto is the perfect example: ABF is a small slice of a food company, even though it dominates the material.

So the work is understanding where each company sits in the chain.

One honest note here: there is currently no clean pure-play substrate ETF. Broad semiconductor ETFs like $SMH and $SOXX barely touch this layer, because most substrate names trade in Japan, Taiwan, and Korea. So this layer either requires buying individual foreign-listed names (some, like Ibiden and AT&S, have accessible listings or ADRs) or accepting indirect exposure through Asia-focused semiconductor funds. That difficulty is also partly why the layer may still be underpriced.

Part 4: The Risk Nobody Prices In: Locked Contracts Meet Rising Costs

Now put Part 1 and Part 3 together, because this is the main risk I want investors to understand.

The bull case for memory stocks is that long-term contracts have removed the old cyclicality. But contracts cut both ways.

Micron’s largest strategic agreements generally have a ceiling price set at current market prices, and a floor price through the term of the agreement. The floor protects the downside. But the ceiling means Micron cannot fully reprice upwards if its own costs explode.

Now look at what is happening on the cost side. Ajinomoto is raising ABF film prices 30%. Resonac and Mitsubishi Gas Chemical are raising laminate materials around 30%. Substrate prices are rising every quarter. And Micron itself has told investors that newer generations of HBM come with rising cost per bit, and that blended DRAM cost per bit is projected to rise from current levels. HBM4 even requires a logic base die made at an external foundry like TSMC, which is another cost line memory makers never used to have.

So here is the squeeze in simple terms. If your selling price has a contractual ceiling, and your input costs (substrates, packaging materials, base dies, equipment) keep rising, the difference comes out of your margin. Memory makers are printing 70-80% gross margins right now. Those margins are the ceiling minus the costs. The ceiling is fixed. The costs are not.

To be fair, the biggest ABF cost sits in the GPU package itself, so the first and hardest hit from substrate inflation lands on chip designers, packaging houses, and ultimately Nvidia’s customers. But cost inflation flows through the whole chain, and memory makers are not exempt.

There is a second version of this risk that is even simpler. A finished AI system cannot ship without a substrate. If substrates become the binding bottleneck, GPU shipments slow, and if GPU shipments slow, HBM demand slows with them. No matter how many contracts are signed. The whole chain is only as strong as its tightest link.

And then there are the classic risks that never go away:

Cyclicality. Micron is up several hundred percent in a year. SK Hynix is up over 300%. These stocks are priced for the supercycle to continue. Any pause in AI capex, the drawdowns will be violent. Memory has done this to investors many times before.

Part 5: Packaging, Equipment, and ETFs

After memory and substrates, the next layer is packaging and equipment.

This includes names like TSMC, ASE Technology, Amkor, Hanmi Semiconductor, Disco, Tokyo Electron, and Advantest.

These companies are not all memory stocks. They are not all substrate stocks either. But they sit around the same AI hardware buildout. As AI chips become more complex, advanced manufacturing, packaging, testing, and equipment all become more important.

For ETFs:

$DRAM (Roundhill Memory ETF) is the cleanest memory ETF to research. It launched on April 2, 2026 at $28, and pulled in billions within weeks, reportedly the fastest ETF launch in history. It holds companies getting the majority of revenue from memory: roughly 25% Samsung, 24% SK Hynix, and 24% Micron, so about 73% sits in just three names, with the rest spread across Kioxia, Sandisk, Western Digital, Seagate, Nanya, and Winbond in small weights. Understand what that means: this is a concentrated three-stock basket in one ticker. Its real value is giving easy access to Samsung and SK Hynix, which used to be hard for regular investors to buy. There is also a 2x leveraged version. If you are a beginner, stay away from the leveraged one.

$SMH and $SOXX give broader semiconductor exposure, dominated by Nvidia, TSMC, Broadcom and the big designers. Good for the overall theme, weak for the specific memory thesis.

Asia semiconductor ETFs give more exposure to Taiwan, Japan, and Korea, where many of the substrate and packaging companies are based, but they still need to be checked carefully because many are dominated by the largest chip names rather than the layers underneath.

The Main Takeaway

The memory trade is bigger than just buying Micron, SK Hynix, or Samsung.

Those companies are still the most obvious names because they make the memory. They benefit directly from HBM demand, stronger pricing, and long-term customer agreements.

But the supply chain does not stop there.

AI needs more compute > More compute needs more memory > More memory needs better packaging > Better packaging needs better substrates > Better AI infrastructure also needs companies that can build and optimise the full system.

The simple chain is this.

> Nvidia gives the compute.

> SK Hynix, Micron, and Samsung provide the key memory.

> Sandisk, Kioxia, Western Digital, and Seagate give broader storage exposure.

> PENG helps build and optimise AI infrastructure where memory is becoming a real bottleneck.

> Ibiden, Unimicron, Ajinomoto, AT&S, and other substrate names sit underneath the advanced packaging layer.

How I Would Approach It

This is how I personally think about building exposure. This is education, not financial advice. I don’t know your situation, and every one of these stocks has already had a massive run, so entry price and position size matter more than usual.

If you only want one simple, liquid, U.S.-listed name: $MU. It is the purest large-cap memory exposure on a U.S. exchange, number two in HBM, with over $100 billion in contracted revenue through 2030 and record margins. The trade-off: it has already gone up several hundred percent, and it captures the second-biggest slice of HBM, not the biggest.

If you want the actual HBM leader: SK Hynix. The reasons are concrete. It has roughly 60% of the HBM market, close to 70% of Nvidia’s HBM4 allocation for Vera Rubin, exclusive or first-supplier positions on chips like Microsoft’s Maia 200 and Google’s newest TPUs, sold-out supply, and it still trades at a lower forward earnings multiple than Micron. And the access problem is solved: from July 10, 2026, its ADRs trade on the Nasdaq under SKHY, and possible future index inclusion could bring forced passive buying. The trade-offs: the listing issues new shares (dilution), the stock is up over 300% this year, and Korean names have historically traded at a discount for a reason.

If you want the value/recovery angle: Samsung. It is the cheapest of the three on multiples, its HBM share is recovering towards 30%+, and it now prices HBM4 at parity with SK Hynix. The trade-off: it is a conglomerate, so memory upside gets diluted by phones, displays, and a struggling foundry business. You are betting on the laggard closing the gap.

If you don’t want to pick: $DRAM. One ticker, all three HBM makers plus the storage names. Just go in knowing 73% of it is three stocks that all move together, so it will fall as hard as it rises. It is a concentrated sector bet.

If your thesis is storage and inference data: $SNDK first, then Kioxia when its U.S. listing arrives. NAND pricing is now rising faster than DRAM, both companies are effectively sold out for 2026, and Sandisk’s HBF project with SK Hynix gives it a long-shot HBM-style story of its own.

If you want the system/inference layer: $PENG. Different exposure from the memory makers, backed by SK Telecom, elevated by NVIDIA, and attacking the memory wall in inference with a real shipping product. But it is smaller, already up 286% this year, and needs to keep executing. This is the highest risk/reward name on this list, which is exactly why it should not be your biggest position.

For substrates: individual names, not ETFs, because a clean substrate ETF does not exist. Ibiden and AT&S are the most researched substrate plays with accessible listings; Unimicron, Kinsus, and Nan Ya PCB require Taiwan access; Ajinomoto gives you the material monopoly wrapped inside a food company. This layer requires the most homework, and that is exactly why it may still be mispriced.

Nothing here is financial advice. Do your own research and size accordingly. I will be buying a memory stock soon. Separate write-up coming soon.

Technical analysis is not magic. It is just a way of studying price behaviour. The goal is to understand where buyers and sellers have shown up before.

1. What technical analysis actually is?

Technical analysis means using price, volume and chart structure to understand what the market is doing. It does not tell you exactly what will happen next. It helps you build a plan.

A beginner should focus on the basics first: trend, support, resistance, volume and risk. Most people make technical analysis too complicated too quickly.

2. How to read candlesticks?

Each candlestick shows price action for a specific period of time. A daily candle shows one day. A weekly candle shows one week. The body shows the open and close. The wicks show the high and low price reached during that period.

· Open is where the candle starts.

· Close is where the candle ends.

· High is the highest price during the period.

· Low is the lowest price during the period.

· Long bodies show stronger momentum. Long wicks show rejection or indecision.

3. Start with the right timeframe!

4. Support, resistance, and trendlines.

Support is an area where price has bounced before. Resistance is an area where price has struggled before. A trendline connects important highs or lows to show the direction of price.

· In an uptrend, price usually makes higher highs and higher lows.

· In a downtrend, price usually makes lower highs and lower lows.

· In a sideways market, price moves between support and resistance.

· Support and resistance are areas, not exact prices.

Do not overthink it: The more times price reacts near a level, the more attention that level deserves. But no level is guaranteed to hold forever.

5. Volume: the strength behind the move.

Volume shows how much trading is happening. A breakout on strong volume is usually more meaningful than a breakout on weak volume.

6. Moving averages.

A moving average smooths out price and helps you see the general direction more clearly. Many beginners watch the 20-day and 50-day moving averages.

· Price above a rising moving average often shows strength.

· Price below a falling moving average often shows weakness.

· Moving averages are guides, not guarantees.

· Use them with support, resistance, and price action.

Keep it simple: Volume and moving averages are most useful when they support what price is already showing you.

7. A simple chart reading routine.

1. Check the higher timeframe trend.

2. Mark support and resistance.

3. Look at volume.

4. Decide if price is trending, ranging, or channeling.

5. Plan your entry, stop, and target before taking a trade.

8. What a channel is?

A channel forms when price moves between two boundaries. One line acts like support. The other line acts like resistance. Channels help you see whether price is trending up, trending down, or moving sideways.

9. How to identify channels step by step?

1. Start by finding two clear swing lows in an uptrend, or two clear swing highs in a downtrend.

2. Draw the first trendline along those points.

3. Copy that line and move it to the opposite side to frame price.

4. Look for repeated touches or near touches on both sides.

5. The cleaner and more obvious the structure, the better the channel.

Channel rule: A useful channel usually has at least two touches on one side and at least one or two touches on the other side. More touches are better, but only if the lines are natural.

10. Good channels vs forced channels.

Example:

11. Channel breakout and retest.

A breakout happens when price moves outside the channel. A retest happens when price comes back to the breakout area and then reacts again.

12. Common chart patterns beginners should know.

Final Takeaway

Technical analysis is useful because it gives structure. It helps you stop guessing and start planning. For beginners, the goal is to understand the trend and identify key levels.

If you only remember one thing about channels, remember this: a valid channel should look natural. Price should respect both sides enough that the structure is clear. If you have to force the lines, skip it.

I think peptides could become a major growth engine for both companies. This guide explains what peptides are, why the market is growing, how athletes use them, what the July 2026 FDA meeting means, and why both Hims and Enhanced could benefit.

None of this is financial or medical advice.

What a peptide is?

A peptide is a short chain of amino acids. Amino acids are the building blocks of proteins. When you put a small number of amino acids together, you get a peptide. When the chain gets much longer, you get a protein.

The easiest way to understand peptides is this: peptides are signals.

A specific peptide can attach to a specific receptor in the body and tell cells to do something. Insulin is a peptide. Oxytocin is a peptide. GLP-1 drugs like semaglutide and tirzepatide are peptide based drugs. Collagen supplements are also made from peptides. Your body already produces and uses peptides every day.

That is why the category is so interesting. Peptides are not one product. They are a broad class of biological signals that can be used across weight loss, diabetes, recovery, skin, sleep, inflammation, hormone support, sexual function, and longevity. Some are already approved medicines. Others are still experimental, poorly studied, or sitting in a complicated legal area.

The peptides market

The peptide drug market is already large. One estimate from Grand View Research puts the global peptide therapeutics market at about $140.9 billion in 2025 and forecasts it could reach about $294.6 billion by 2033. Other research firms use lower numbers, so I would not rely on one exact figure, but the direction is clear. This is a large and growing healthcare category.

The biggest driver right now is metabolic disease. GLP-1 drugs changed the whole conversation. Ozempic, Wegovy, Mounjaro, and Zepbound showed that peptide based drugs can become some of the most important products in the world. That success pulled more money, more research, and more consumer attention into the wider peptide category.

This is where the investment angle starts. GLP-1s proved the demand. Now investors are asking what comes next.

Simple peptide table

How athletes use peptides

Athletes use peptides for recovery, muscle growth, and performance.

BPC-157 and TB-500 are used to speed up healing of muscle, tendon, and joint injuries. Sermorelin, Ipamorelin, and CJC-1295 are used to raise the body’s own growth hormone, which supports muscle and recovery. The appeal is faster recovery and gains with what users believe are fewer side effects than traditional steroids.

Two facts matter here. First, the clinical evidence for many of these is thin. It rests more on user reports than on large human trials. Second, most are banned in professional sport under global doping rules, which puts them off limits in mainstream competition. That ban is exactly what created the opening for the Enhanced Games.

The July 2026 peptides meeting

Everything for both companies runs through one regulatory event.

In the US, some drugs are sold through compounding, where a licensed pharmacy makes a customized version of a medicine. There is an approved list of ingredients pharmacies can use for this, called the 503A Bulk List. Telehealth companies grew fast by selling compounded drugs, including compounded GLP-1s during the shortages.

In September 2023, the FDA removed a group of peptides from the compounding list over safety concerns. That decision pushed most peptide demand into the black market of “research only” vials.

On July 23 and 24, 2026, an FDA advisory committee meets to decide whether to recommend adding several peptides back to the approved list. If the answer is yes and the FDA agrees, compounding pharmacies and the telehealth platforms attached to them get a legal path to sell these peptides at scale. This is the catalyst for both stocks. It is not guaranteed. The committee only recommends, and the FDA makes the final call.

Hims & Hers

Hims & Hers is the largest direct to consumer telehealth company in the US. It closed 2025 with about $2.35 billion in revenue, up 59%, and more than 2.5 million subscribers, largely on the back of compounded GLP-1s. That GLP-1 growth is now slowing. A deal with Novo Nordisk shifted the company toward selling lower margin branded product, and 2026 guidance calls for slower growth. Hims needs a new high margin growth engine.

Peptides are that engine, and the company has already built for it. In February 2025 it bought a peptide manufacturing facility in California. It owns the production capacity before the category is legal at scale. If the July meeting opens the door, Hims can move faster than rivals who would have to start from nothing. Management has said peptides are central to a new longevity business.

How it benefits: a legal peptide market lets Hims turn its existing subscriber base and its own factory into a fresh, high margin product line that replaces the slowing GLP-1 growth.

Enhanced Group

Enhanced Group runs the Enhanced Games, a Las Vegas event where athletes compete using supervised performance enhancing drugs. The Games are not the business. They are the advertisement. The real company is a direct to consumer telehealth platform that sells peptides, testosterone therapy, GLP-1s, and longevity protocols.

The model is simple. Elite athletes use supervised protocols and set records. That generates attention and lends the products credibility. Ordinary customers then come to the platform to buy legal versions of similar protocols.

How it benefits: Enhanced is a pure bet on peptides going mainstream. It has built a brand for a category most rivals can only market with before and after photos. If the July meeting goes its way, it has a ready funnel of attention pointed straight at a newly legal product. It is also small and burning cash, which makes it the higher risk of the two.

Bottom line

Peptides are short chains of amino acids that act as the body’s signals. The drug market is already worth around $140 billion and growing, athletes already use peptides widely, and a large consumer market is sitting in a legal gray zone waiting on the FDA. Hims & Hers is the safer, diversified way to own that shift, with a factory already in place. Enhanced is the concentrated, higher risk way to own it. Both depend heavily on the July 2026 meeting going in their favor.

I hold positions in HIMS and ENHA. This is my opinion, not financial or medical advice. Peptides carry real health risks and should only be used under medical supervision.

I think I have a different take on this one than most people.

When I buy a stock for my long term portfolio, I am asking what it looks like in five years, ten years, twenty years. If I cannot picture the company being far bigger and far more important two decades from now, it does not go in. That is the whole brutal filter, and most companies fail it.

Hims and Hers passed it for one simple reason. I looked around and realised there is a platform for almost everything in our lives now, except healthcare. There is a platform for music. A platform for video. A platform for shopping, for food, for travel, for your money. Healthcare is the one enormous part of life that never got its platform. It is still slow and a pain to deal with. That gap is the opportunity, and it is huge.

It is worth asking what it means that Andrew Dudum, the founder and CEO of Hims, reposted my Netflix of healthcare post to his own followers. Founders guard their name. They do not amplify a take that misreads the company they are building. So it tells me that I am in the right. That the thesis matches the direction the person in charge is taking it. A long term position is really a bet on the founder’s vision, and here the vision I am backing and the one he is building are the same.

2. The volatility, and why it does not scare me.

Let me deal with the obvious thing first, because it is what everyone fixates on. This is one of the most volatile names you will find anywhere. It ran up to around $70 in early 2025 and then gave most of it back. Over the last year alone the range runs from roughly $14 to $70. It has swung 30 or 40% on single headlines, including a jump of about 36% in one day this March when the Novo news landed. As I write this it sits in the mid 30s, still only about half of its old high. If you cannot stomach that kind of volatility, then this is not the stock for you.

But here is how I think about it. A share price and the business underneath it are two completely different things, and they can stay disconnected for a long time. Jeff Bezos has talked about exactly this. During the dot com crash, Amazon fell from over $100 to single digits while the business itself kept getting better. Customer growth, revenue and the long term strategy were all moving in the right direction. The stock was reacting to fear, not to fundamentals.

Hims is in that same spot today. The price moves at every headline (even with unrelated news), but the business keeps executing and expanding. Over time that gap tends to close. So the volatility does not scare me. If anything, for a patient investor, it is the opportunity.

3. This was never a weight loss company.

The bear case more or less comes down to this. People think the growth is all tied to weight loss, and that the moment the GLP-1 hype cools, the story is over. That view does not reflect how this business is actually built. Hims has never been a one category company. Hair loss, dermatology, sexual health and mental health were the core long before GLP-1s ever mattered. Weight loss added another layer of growth, but it never defined the platform.

Over the past year they have pushed even further into personalized solutions and brought more of their compounding in house. That gives them real control over product development, speed and margins, and it cuts their reliance on third parties. So the thing the market panics about most is the thing that matters least to where this company is heading.

4. The friction Hims is deleting.

Let me show you the real opportunity using how the system works where I live.

In the UK, if something is wrong with me, the journey looks like this. First I have to book an appointment to see a doctor. Most of the time I do not even get the slot I want. I cannot ring up and say give me tomorrow at 3pm. They tell me when they are free and I take it or I wait. Then I finally sit in front of the doctor and they say I need a blood test. So I book another appointment for the blood test. I go in, give blood, and wait a few days for the result. Then they call me back in to discuss it, which is yet another appointment. By the time I actually know what is going on, more than a week has gone by for something that should have taken few days.

Now look at what Hims does to that entire chain. You see the doctor straight from your phone. No waiting room, no ringing the surgery, no being told when you are allowed to see a doctor. You put your information in, you can request lab work yourself, and they post a test kit to your door. You do the test, send it back, and the results land inside the app. The doctor tells you what it means right there. The week long obstacle course becomes a few taps and once someone feels how much easier that is, they do not go back to the old way.

Important Note: Of course, there are certain situations where in-person visits will be required.

5. The real moat: your data and an AI that knows you.

The friction is only the entry point though. The part that really got me excited is the data, and what they are now doing with it.

Think about how we have actually behaved over the years. Before AI, when something went wrong with our body, we would Google our symptoms. These days people do the same thing with ChatGPT. They paste in their symptoms and ask the bot what is wrong. The bots are useful, but they have one massive blind spot. They know nothing about you. They are guessing from the open internet based on a sentence you typed. It is medical advice for a complete stranger.

Now here is what Hims has built. Late in 2025 they launched Labs, where you book a blood draw, they run over 130 biomarkers across areas like heart health, hormones, metabolism, inflammation and stress, and the results shows up in your app. Then in 2026 they launched Labs AI, and this is the piece that matters the most. Labs AI is an AI that reads your actual bloodwork. It does not pull answers from the open internet. It reasons over your real biomarker history and your personal profile, and it talks to you about you. It can look across your cholesterol and so on, spot a pattern, tell you what it means for your health specifically, tell you what to change, and then hand you to a real clinician.

Do you see what that is? It is the exact thing everyone has been trying to do with ChatGPT, but done properly. ChatGPT is advice for a stranger. Labs AI is advice for you, because it holds your data and it gets smarter every single time you test. The company has said this is only the first agent, with smarter dosing and care companions coming next. They are building the intelligence into the foundation.

Underneath all of this sits MedMatch, their AI system trained on millions of clinical interactions, so the personalization gets sharper as the platform grows. They also bought Trybe Labs to bring at home testing right into the membership, so you can track things like cholesterol, hormones, thyroid function and cardiac risk from a simple at home kit. Put it together and the structure is an unstoppable moat.

That is the infrastructure for a real healthcare membership on the same level as Costco and Amazon Prime.

6. Why nobody leaves.

This is also exactly why I am not worried about people churning out. Once Hims holds years of your biomarker history and a record of how your body responded to a given treatment, walking away means starting again from zero somewhere else. Why would you? Your whole health story lives in the app that you can access 24/7.

The numbers already back this up. Around 82% of subscribers are still active past the three month mark. As people move from one service to another, lifetime value rises and churn falls, and personalized medicine is already growing faster than the rest of the platform. It works a lot like a Costco membership. The repeat interaction is the whole point. It drives lifetime value up and it makes the platform very hard to leave. The data is the moat, and the longer you are on it, the deeper that moat gets.

7. The Netflix of healthcare.

My original line on all of this was simple. Hims is the Netflix of healthcare. Let me explain what I mean.

Netflix did not invent anything when it started. DVDs existed, renting movies existed, they just repackaged it in a more convenient way and posted discs to your door. Blockbuster laughed at the idea. Then streaming came and they ran the same play. Then they became a content machine and started making their own films and shows. Today Netflix has over 325 million paid memberships, an audience getting close to a billion people, and around 45 billion dollars a year in revenue. Studios line up to be on it. Put your film on Netflix and you reach a gigantic global audience instantly, with basically no marketing cost. Netflix is the front page of the entertainment world, and that front page is the power.

Hims is becoming that front page for healthcare, and we are already watching it happen. Look at the Novo Nordisk story. The two of them have history. They partnered in 2025, fell out, and earlier this year Novo actually took Hims to court over compounded copies of Wegovy. And then, within weeks, the lawsuit turned into a partnership. Novo dropped the case, and Hims now sells branded Wegovy and Ozempic, the injections and the pills, straight through its platform. Novo is doing the same kind of deal with Ro, WeightWatchers and LifeMD.

Read that again. One of the biggest drug makers on earth went from suing Hims to using it as a sales channel in the space of a month, because it needs the distribution. The telehealth platforms that were supposed to be a threat are being pulled into the supply chain instead.

That is the whole thesis playing out in real time. A company makes a treatment, it wants that treatment in front of as many of the right patients as possible without burning a fortune on advertising, and the obvious move is to put it on Hims and reach millions of engaged members in one shot. Right now the balance of power still mostly sits with the drug makers. But as the subscriber base climbs into the tens of millions and the data deepens, that channel only becomes more valuable, and the leverage keeps shifting toward Hims. Same energy as Blockbuster laughing, right up until it stopped being funny.

This is the move I am really betting on. Hims going from a Tier C subscription that is a nice to have, up to a Tier A membership on the level of Amazon Prime and Costco, the kind of thing people consider essential and never cancel.

8. What they have actually been building.

Step back from the share price for a second and just look at what they got done in a single year. They pulled more of their compounding in house and leaned into personalized medicine. They launched new categories like menopause care and testosterone treatment, and kept scaling the ones they already had. They added Labs as a recurring layer and bought Trybe Labs for at home testing. They pushed abroad into the UK, Europe and Canada, and lined up a major move into Australia. They put more than 200 million dollars into a new facility in New Albany, creating around 400 jobs. They poured money into MedMatch and they backed GRAIL, which closed a 325 million dollar raise in late 2025, putting Hims alongside a company working on early cancer screening for people who have no symptoms yet. That last one is worth sitting with, because that is preventative healthcare, the kind that catches things years before a normal system ever would. Put the whole list together and one thing is clear. This is not the same company it was a year ago.

9. Where this goes: five, ten, twenty years.

So picture it out a few years. Hims becomes to healthcare what Spotify is to audio. One place. Your doctor, your prescriptions, your labs, an AI that actually knows your history, your ongoing care, all sitting in one app on your phone. A drug company launches the next big thing and the obvious move is to put it on Hims, the same way an artist drops a track on Spotify. That is how simple healthcare becomes.

Eventually I do not think they stay a distributor at all. Netflix went from licensing other people’s films to making its own award winners. With enough scale, enough cash and enough data, Hims has the exact ingredients to fund its own research and build its own products, designed around real outcomes from millions of real patients.

Here is the kicker that makes this bigger than any comparison I have used. The market for this is not the size of streaming, and it is not the size of music. It is healthcare. It is one of the largest pools of spending on the entire planet, many times bigger than video and audio combined. A platform that even partly owns that is playing on a scale Netflix and Spotify never could.

People did not see what Netflix would become when it was still mailing out discs. I think a lot of people are making that exact same mistake with Hims right now.

This is why I am so bullish on the company.

None of this is financial advice. I am an investor in Hims & Hers.

Before I start, I want to be honest and say that most of the names on this list have had a great run. These are not safe stocks. They are young companies and most of them not profitable yet, and the share prices swing around like crazy. A few of them are actually up a lot, which sounds great until you remember that a stock that runs up fast can come down just as fast.

This is written for two kinds of people. Beginners who want to understand why anyone would own names like this in the first place, and long term investors who think in decades and are comfortable sitting through the noise. If that is you, read on.

1. IREN. Every AI tool you have ever touched runs in a giant building full of chips, and IREN builds those buildings. It started life as a Bitcoin miner and is now pivoting hard into renting out AI computing power, with some genuinely huge contracts already signed with Microsoft and NVIDIA (NVIDIA even took a stake). The thesis here is about as simple as it gets. The world keeps wanting more AI, and someone has to supply the raw compute to run it. IREN is sprinting to be one of those suppliers. The catch, and there is always a catch, is that this kind of buildout costs a fortune, so expect heavy spending and some not-so-great quarters while they make the switch. Other notable stocks in this sector: NBIS, CRWV, CIFR, WYFI, WULF & KEEL.

2. OUST. Robots and self driving machines need to see the world in 3D, and Ouster makes the laser vision (called lidar) that lets them do it. Its sensors are showing up in warehouse robots, smart traffic systems, mining equipment, and a bunch of other places. The bet is straightforward. If physical robots scale the way a lot of us expect, they all need eyes, and Ouster wants to be the company selling them.

3. ONDS. Ondas makes drones that guard sites on their own around the clock, plus systems that can knock hostile drones out of the sky. It is also a US approved alternative to Chinese drones, which matters more and more as defense budgets climb around the world. Think automated security for airports, borders, stadiums and military bases. It is still small and still not profitable, but the contracts have been picking up.

4. KRKNF. The ocean is one of the last frontiers we have barely touched, and Kraken makes the sonar, underwater robots, and deep sea batteries that navies and energy companies need to operate down there. Here is the part I like most. Kraken sells its parts into a lot of the competing underwater drone makers (including Anduril), so it does not really have to pick the winner of that race. It can win whoever wins. That is exactly the kind of business model I look for.

5. ASTS. AST is building satellites that work like cell towers in space, so a normal phone can get a signal even where there is zero coverage, no special hardware required. It already has the big carriers and governments lined up behind it. This one is early and speculative, but if it actually works at scale, the market is measured in billions of phones, and that is hard to ignore.

6. HIMS. This is the most established name on the list, and the only one that already has a giant base of paying customers (millions of people getting prescriptions for skin, hair, mental health, and weight loss through an app, delivered to their door). It is now pushing into hormones, lab testing, and longevity. The bet is simply that more and more of consumer healthcare moves online, and Hims is one of the brands leading that shift.

7. IBRX. Instead of attacking a tumor directly, ImmunityBio tries to wake up your own immune system and let it fight the cancer for you. Its first drug is already FDA approved for a type of bladder cancer and is expanding from there. Cancer is one of the biggest prizes in all of medicine, so if this keeps working, the upside is enormous. It is also the riskiest type of bet on this list, because biotech really does live and die on trial results.

They are basically one big bet that the future (ai, robotics, space, defense, longevity) actually plays out. I started investing in them very early, and because I run a very concentrated long-term portfolio, I’m currently rotating the same amount of money across all 7.

This is just how I think about them, not financial advice. Do your own work, and only put in money you can afford to hold for years.

CHALLENGE 1: The starting point. Turn £5,000 into £10,000. a 100% return. This was my first a…

Read more

These are companies where the majority of the money comes from one single source, and that’s what makes them so easy to understand. You don’t have to untangle a dozen different things to figure out how they make money, because it all comes down to people paying every month.

Netflix is the perfect example.

Almost everything they earn comes from people paying a monthly fee to watch shows on their platform. Those same people pay again the next month, and the month after that, so the company already knows roughly what’s coming in before the month even starts. That steadiness is the whole reason these businesses are so attractive, and once you understand it, a lot of investing starts to make more sense.

The most important thing to look at is how long they manage to keep their customers, because that’s where the real money is. If Netflix can hold onto you for five years instead of one, that’s five times as much money from the exact same person, without spending anything extra to get you. So when a company gets even slightly better at keeping people around, it makes a massive difference to how much they earn over the long run.

These companies also have room to raise their prices very gradually. You’ve probably watched your own monthly bill go up by a dollar or two over the years and barely thought twice about it, so you kept paying. Spread that small increase across millions of customers and it turns into a serious amount of money, and it works even better once the product has become something you use without thinking.

Another thing that makes them powerful is how much they can afford to spend to win you over in the first place. Since the company already expects you to stick around and pay for a long time, it makes sense for them to spend more upfront to get you signed up, knowing they’ll earn it back over the months and years you stay. That gives the strongest companies a real edge, because they can keep pulling in new customers while still making money on each one.

There’s also the habit side of it, which is easy to overlook. When you watch Netflix every night or put Spotify on every time you get in the car, the product quietly becomes part of your daily routine. Once it’s woven into your day like that, most people just keep paying, because cancelling feels like more effort than it’s worth.

On top of all that, these businesses give you a really clear read on how they’re actually doing. Because the payments come in steadily every month, you can easily tell how many people are using the product and whether that number is growing or shrinking, which makes your job as an investor a whole lot easier.

And once one of these companies gets big enough, it can start adding even more ways to make money. With millions of people already paying every month, it can introduce things like ads, cheaper plans, premium plans, or bundles, and it doesn’t have to go find a brand new audience to do it, because everyone’s already there.

So if you’re just getting started, this is a great place to start. Pick a subscription business you already use yourself, pay attention to how well it keeps its customers, and watch how it grows over time. Once this clicks for you, you’ll start spotting it everywhere you look.

My top subscription-based stocks:

1. Hims & Hers

Hims,, because the subscription is the business. People sign up for ongoing treatments like hair loss, skincare, or weight loss, and those treatments aren’t a one-time thing, they’re something you keep needing month after month. That means the customer naturally comes back because the product only works if you stay on it. On top of that, the whole experience is personal to you, so once you’ve found something that works, you’re not likely to go hunting for an alternative. That combination of recurring need and personal fit is exactly what makes a subscription business strong, and Hims has it built right into the core of what it does.

2. Netflix

Netflix is the one almost everyone understands straight away, which is why it’s such a good place to start. Nearly all of its money comes from people paying a monthly fee to watch shows, and those same people keep paying month after month without really thinking about it. The genius of it is how normal it’s become, cancelling Netflix feels like a bigger decision than just letting it roll over for another month. And now that they’ve added cheaper ad-supported plans and cracked down on password sharing, they’ve found ways to earn even more from the millions of people already on the platform, without having to go and find a brand new audience.

3. Amazon

Most people think of Amazon as a shopping website, but the part I want to focus on is Amazon Prime, because that’s the subscription side and it is absolutely massive. Prime is the perfect example of how a subscription quietly locks people in. Once you’re paying for that fast, free delivery, you naturally start buying more from Amazon and less from anywhere else, because it just feels easier. Then you get the video, the music, and all the other perks bundled in, so the membership starts to feel like incredible value even if you only signed up for the delivery. And because people pay for Prime every single year, Amazon knows that money is coming in, and it knows those members are far more loyal than a normal shopper. That’s the power of the subscription, it turns a casual buyer into someone who barely shops anywhere else.

4. Spotify

Spotify is one of those products that slips into your daily routine without you even noticing. You put it on in the car, while you’re working, while you’re at the gym, and before long it’s just part of your day. That habit is what makes it such a strong subscription, because once the music is woven into your routine like that, paying a small monthly fee feels completely natural. Most people never think about cancelling, because the moment they want to play something, it’s right there. That steady, everyday use is exactly the kind of thing that makes a subscription business so reliable.

5. Costco

Costco is a brilliant one to finish on, because the membership is the subscription and it’s honestly one of the best examples out there. You pay once a year just for the right to shop there, and the overwhelming majority of people renew it every single time. Think about how unusual that is, people happily paying a yearly fee just to be allowed through the door. They do it because they trust that Costco is genuinely giving them a good deal, so that membership fee feels completely worth it. And because almost everyone renews, Costco can count on that money coming in year after year, from customers who are some of the most loyal in all of retail. That membership is the quiet engine behind the whole thing, and it’s a perfect example of how powerful a simple subscription can be.

First, the Enhanced Games, a Las Vegas sporting event where athletes compete using performance enhancing drugs under medical supervision, with no anti-doping testing.

Second, a direct-to-consumer telehealth platform selling peptides, GLP-1 weight loss drugs, testosterone therapy, and supplements.

The idea is really simple.

The Games generate massive amounts of attention for near-zero customer acquisition cost, and that attention gets funneled into the telehealth business, which is where the actual recurring revenue is going to come from.

The numbers speak for itself.

Enhanced Games captured the world’s attention.

• 1B+ people engaged worldwide over 7 days

• 4M+ live views across owned + partner platforms

• 1M+ hours of live competition watched globally

• 1.9M live viewers driven by top Twitch streamers

Most people look at Enhanced and see a controversial sporting event. But I see a massive opportunity in a new longevity sector that is just getting started.

The main bull case

1. The Games genuinely did light up the internet.

Whether you agree with the concept or not, the company achieved something that most startups never do.

Enhanced secured more than $32 million in contracted sponsorship value with seven months still left in the year. The event generated roughly 4,000 earned media articles across publications with billions of combined monthly visitors. For a first-time sports event, getting distribution across Roku, YouTube and Rumble while pulling millions of live views is a MASSIVE achievement.

Most new sports leagues spend years burning capital trying to get noticed. Many never do.

Enhanced became a global story almost overnight.

The important part is that attention is usually the hardest thing to manufacture. Every healthcare company in the world is fighting for customer attention. Enhanced already has it.

If management can consistently turn these headlines into customers, the Games become one of the cheapest marketing engines in healthcare.

That’s my biggest bull case.

2. The telehealth thesis sits on top of multiple healthcare megatrends.

The telehealth business is probably what matters the most to use as a investor.

GLP-1 drugs like Ozempic and Mounjaro have become one of the biggest healthcare products in decades. Demand continues to explode and only a small percentage of eligible patients currently have access.

At the same time, consumers are becoming increasingly comfortable getting treatment online instead of going through traditional healthcare channels.

Hims & Hers proved the model works.

Enhanced is attempting to build a similar business, but with a heavier focus on performance, longevity, peptides, testosterone optimization and metabolic health.

The peptide opportunity is what I am most bullish on.

Today a large portion of the peptide market exists in a grey area. Consumers often buy products from questionable websites with inconsistent quality control and little medical oversight.

If Enhanced can become a trusted, physician-supervised platform operating inside regulatory guidelines, the opportunity becomes much larger than simply selling GLP-1s.

You’re talking about recurring subscriptions, high gross margins, repeat customers and potentially very attractive lifetime value economics.

That’s exactly the type of business investors tend to reward with premium multiples.

3. There is a near-term regulatory catalyst.

The FDA’s peptide meeting on July 23rd and 24th, 2026 could end up being an important event for the company.

If regulators expand the list of peptides available for compounding, Enhanced immediately gains access to a larger menu of products and services.

That matters because the company is still in the early stages of building out its platform.

More approved products means more revenue opportunities, more customer use cases and potentially higher average revenue per user.

Nobody knows exactly how the meeting will play out.

But if the outcome is favorable, it could accelerate the growth story much faster than investors currently expect.

4. Wall Street is beginning to pay attention.

Analyst coverage alone doesn’t make a stock go up. But it does matter, especially to a $500m company.

Three analysts currently rate the stock an average Strong Buy with a twelve-month target around $10.50, implying significant upside from recent levels. Individual targets range from roughly $6 to $15, while firms such as Lucid Capital and Canaccord have initiated coverage with Buy ratings.

The bigger point is that multiple firms are spending time covering the company at all.

Most small caps never receive meaningful institutional attention.

Coverage increases awareness, expands the potential shareholder base and helps put the company in front of investors who otherwise would never hear about it.

For a company this early in its lifecycle, that is an incredible achievement.

5. The market may be valuing the wrong business.

The Games don’t necessarily need to become the next UFC, Olympics, WWE etc. They simply need to remain culturally relevant enough to keep generating attention for the telehealth business.

Every world record, every athlete story, every controversy and every viral headline becomes marketing for Enhanced.

Traditional telehealth companies spend enormous amounts of money acquiring customers through Google, Facebook and television advertising (HIMS spent $16m on 2026 Superbowl ads).

Enhanced is attempting to acquire customers through global media coverage.

If it works, customer acquisition costs could end up being dramatically lower than competitors.

The risk - The entire thesis comes down to one thing.

> Can the company convert attention into recurring healthcare revenue?

Now they need to prove that viewers become customers and customers stick around.

If Enhanced gains traction, the stock could eventually be valued more than any other healthcare platform.

That’s what makes the next few quarters so important.

My plan:

> First buy complete.

> Aim to get it to 10% of my portfolio over the next few months

NFA.

> Overall my portfolio is up +48.60% YTD.
> In monetary terms, I made £77,902.04 this month.
> Which is almost 2x my yearly salary.

Started Investing on this Account: April 2023
Total Portfolio Size: £241,278.05
Net Deposits: £33,000.00

Allocation:
> Long-term Portfolio: 72.54%
> Short-ter…

Read more

Read more

Google CEO Sundar Pichai told analysts on the call “we are compute constrained in the near term” and that “our cloud revenue would have been higher if we were able to meet the demand.”

That is a company turning away revenue because it physically cannot serve it. Microsoft said nearly the same thing word for word: “even with these additional investments and continued efforts to bring GPU, CPU and storage capacity online faster, we expect to remain constrained at least through 2026.” Meta’s CFO said the company is seeing “unprecedented internal and external demand for AI compute resources.”

The spending numbers that came out of these calls are difficult to process at normal human scale. Google, Amazon, Microsoft, and Meta collectively plan to spend +$700 billion on infrastructure in 2026, up over 70% from last year’s record $410 billion. Google alone is spending over $170 billion this year, and its CFO confirmed that 2027 spending will significantly increase beyond that. To give that a reference point, the entire US defence budget is approximately $850 billion. These four technology companies are spending roughly 85% of that on data centres, chips, and cables in a single year.

The revenue that is justifying this spending is extraordinary. Google Cloud revenue grew 63% year over year, more than doubling its growth rate, and its cloud contract backlog nearly doubled in a single quarter to $462 billion. That backlog number is the most important signal in all of these results. It means customers have already signed contracts and committed money for computing power they want Google to provide, and Google cannot deliver it yet because it does not have the infrastructure built. The demand is locked in and signed. The supply is the problem.

To understand why these companies cannot simply spend their way out of this constraint immediately, you need to understand what an AI data centre actually is. It is one of the most complex industrial objects humans have ever built, requiring the simultaneous availability of extremely scarce components, each with its own supply chain that takes years to scale.

The most critical piece of hardware is the GPU, primarily made by Nvidia. These chips are manufactured almost exclusively by TSMC in Taiwan using the most advanced semiconductor processes on earth and TSMC can only increase its output gradually regardless of how much money is thrown at it. Every major technology company and every national government is investing in AI capability, and every research institution in the world is trying to buy these chips at the same time.

But the GPU shortage, as serious as it is, is actually not the binding constraint right now. The binding constraint is electricity. A single hyperscalers AI data centre requires between 100 and 300 MW of continuous power, the equivalent of a mid-sized city. The US electrical grid has been essentially flat for two decades because efficiency improvements in appliances and manufacturing kept overall consumption stable. Nobody needed to build serious new generation or transmission capacity. The infrastructure planners who run the American grid were not preparing for the sudden appearance of thousands of city-sized power draws in the span of a few years.

Utilities simply do not have either the grid capacity or the generating capacity to build it fast enough to accommodate these new large energy demand centres. Building new power generation and transmission infrastructure takes a decade or more in the United States. A new high-voltage transmission line can take seven to ten years to permit before a single cable is laid.

The specific bottleneck right now is not even the power plants. The core electrical components needed both inside data centres and for the grid upgrades required to supply hundreds of megawatts to a single campus, are in severe shortage. Lead times for high-voltage transformers have stretched from 12 to 18 months to as long as 36 to 48 months in some cases. A transformer is not sophisticated technology. It is a large metal box of copper coils and oil. But you cannot connect any facility to the electrical grid without one, and the factories that manufacture them have not been running at the scale now required because nobody needed them to until very recently. Scaling those factories takes years because the machines that make transformers are themselves specialised equipment with long lead times.

Nearly half of all US data centres planned for 2026, representing approximately 7 gigawatts out of 12 gigawatts of announced capacity, have been cancelled or delayed. They are cancelled because there is nowhere to plug them in. OpenAI’s $500 billion Stargate project in Texas has shown no significant physical progress as of April 2026, despite massive financial backing from SoftBank, facing the same power infrastructure and component shortage constraints affecting the entire industry. The company arguably most central to the entire AI moment, with essentially unlimited funding, cannot move faster because the physical world will not cooperate. Morgan Stanley forecasts a 49 gigawatt shortfall in US data centre power capacity through 2028, and Gartner predicts that power shortages will restrict 40% of AI data centres by 2027.

The memory market is a separate but equally serious problem. The GPUs used for AI require a specialised type of memory called high-bandwidth memory, or HBM, which is stacked directly onto the chip package and allows the processor to move data at the speeds AI workloads require. Manufacturing HBM is itself extremely capital and energy intensive, and only a handful of companies in the world can make it, primarily SK Hynix and Samsung in South Korea. HBM is sold out through 2026 and prices are spiking. Meta was explicit about this on its earnings call, telling investors that its CapEx guidance is going up partly because the price of memory has risen sharply. Microsoft’s CFO attributed $25 billion of its record CapEx budget specifically to rising memory chip prices.

This memory squeeze flows directly into consumer products in a way most people have not connected. Data centres and smartphone manufacturers buy from the same global pool of DRAM and NAND flash memory. AI data centre construction has driven memory costs up five-fold and storage costs up three-fold since early 2025, and smartphone manufacturers, PC makers, and gaming console producers are all competing for the same supply that AI data centres consume in massive quantities. The technology companies building AI infrastructure have deeper pockets than consumer electronics manufacturers and will simply outbid them. This is a direct reason why phones, laptops, and consumer storage have become more expensive over the past year and will likely remain so.

The reason none of this spending is slowing down despite all these constraints comes back to the revenue numbers. When Google Cloud grows 63% and accumulates a $462 billion backlog in signed contracts, the rational response for Google’s board is to spend more, not less. The investment is visibly working. Every dollar spent on infrastructure in 2023 and 2024 is now generating measurable returns in 2026, which makes the case for the next wave of spending stronger, not weaker.

There is also a competitive logic that removes the option of restraint entirely. If any one of these companies pulls back, its rivals gain cloud customers, AI capability, and cost advantages that compound over years. The company that builds the most compute today will have the largest and cheapest infrastructure in 2028, which means it can offer the best AI products at the lowest price, which means it wins more customers, which funds even more infrastructure. Every CEO understands this dynamic, which means the decision to keep spending is not really a decision at all. It is a compulsion created by the competitive structure of the industry.

Plus, the US and China are each treating AI compute capacity as a strategic national asset. Normal market logic, where companies build until returns equalise with costs, does not fully apply when national security is the frame. Neither side can afford to be seen as slowing down, regardless of short-term economics.

What resolves this over time is not any single breakthrough but a gradual unlocking of each constraint layer at different speeds. Nvidia will slowly increase GPU production. TSMC will expand its most advanced manufacturing nodes. HBM manufacturers are building new capacity that will come online over the next two years. Transformer manufacturers are now running at full output and will eventually close the gap. New power generation, including gas plants built specifically to serve data centres, small modular nuclear reactors, and large-scale renewables projects, is being planned and built. But each of these timelines is measured in years, and they all need to align simultaneously for a new data centre to become operational.

Why $IREN?

IREN builds and operates its own data centres using in-house design, engineering, and construction teams, which removes third-party colocation costs and counterparty risk. It is vertically integrated across power, land, data centres, and GPU infrastructure, which means it controls every layer of the stack that everyone else is currently fighting over. That vertical integration provides faster customer onboarding and pricing leverage precisely in the environment where supply is constrained and customers have signed-and-waiting demand.

While the hyperscalers are trying to find land, power, and grid connections they should have secured years ago, IREN built its position from the ground up with power access as the foundation. IREN has 4.5GW of grid-connected power secured across North America.

Remember, IREN has spent years building this portfolio, which means it now holds something that the largest companies on earth are desperately trying to acquire.

To summarise, the companies that signed land and power agreements early, that built relationships with utilities years ago, and that have the balance sheets to absorb cost inflation in memory and components will compound their infrastructure advantage into something that becomes very difficult to close. The companies that waited will find themselves at the back of every queue at once.

2. The path to +4 million subscribers.

The company is acquiring Eucalyptus that reported over 775,000 customers. This is an Australian telehealth company operating across Australia, Europe, and Japan doing north of $450 million in annualised revenue with triple digit growth. Add ZAVA in Europe and the Canadian expansion, and the subscriber base is on a clear path toward 4 million. Management believes international revenues alone can hit $1 billion within three years. Crucially, the 2026 guidance doesn't include a single dollar from Eucalyptus.

3. $83 per month per subscriber.

Monthly revenue per subscriber hit $83, up 28% year over year. This number continues to move higher as the platform expands into testosterone support, menopause, and lab testing, each of which layers additional revenue on top of existing subscribers. Hims is upselling health to people who already trust it. Every new specialty is another revenue stream from the same customer base they have already paid to acquire.

4. 59% revenue growth, recurring and compounding.

Full year 2025 revenue grew 59% to $2.35 billion. The Hers brand alone grew over 100%. Adjusted EBITDA grew 80% to $318 million. This is a recurring subscription business with improving margins, shortening payback periods, and management guiding $2.7 to $2.9 billion for 2026. The long term target is $6.5 billion in revenue and $1.3 billion in EBITDA by 2030, roughly 3x from where the business sits today.

5. Novo Nordisk: From Lawsuit to partnership.

In February 2026, Novo Nordisk sued Hims for selling what it called illegal knock-off versions of Wegovy and Ozempic. They then dropped the lawsuit and signed a full distribution deal. Hims is now an authorised distributor of Ozempic and Wegovy across all doses, starting at $149 per month. Novo basically had to partner with Hims because they realised Hims has 2.5 million subscribers and a distribution machine no pharma company can replicate.

6. The Peptide factory and the July catalyst.

Recently, RFK Jr. announced the FDA will convene an advisory committee in July 2026 to evaluate legalising a range of peptide substances for compounding pharmacies. The stock jumped 25% in two days.

In early 2025, Hims acquired a peptide manufacturing facility in California. Hims is the only company that already has the factory built, the subscribers in place, and the distribution infrastructure ready to go. The moment that regulatory door opens, they don't need to build anything. They just turn the factory on. No competitor is anywhere close to being as well positioned. Analysts project the peptide market could exceed $100 billion by 2030.

Bottom line:
- The subscriber base is compounding.
- Novo lawsuit ended up in a partnership.
- International expansion is accelerating.
- Peptide factory is ready and waiting for a green light.

I've never been more bullish.

The stock has been beaten down so badly due to the lawsuit situation that, even though it is up 49% in the past week, it is still down 13% year-to-date and more than 50% from its all-time highs.

NFA. I have made HIMS my second-largest position, currently 9% of my portfolio.

Read more

YTD: -8.20%

LONG-TERM [83.49%]:
$IREN: 68.26%
$KRKNF: 6.78%
$ONDS: 5.56%
$HIMS: 2.89%

MEDIUM-TERM HOLDINGS [13.26%]:
$VELO: 4.71%
$OSS: 3.13%
$OUST: 3.09%
$IBRX: 2.02% -> New Stock
$TE: 0.31%

SHORT-TERM HOLDINGS [3.26%]:
$NUAI: 2.94%
$ALMU: 0.32%

Week 11 & 12 Trades:
REDUCED: $NUAI, $TE
BOUGHT: $IBRX
ADDED: $IREN, $VELO & $OSS
SOLD: $UAMY

-------------------------------

This update will focus on the macro and what I am planning to do over the next month (PART 2).

Something is happening in the Middle East right now that could reshape the global economy for the rest of the year. Not just for traders or energy companies but for everyone.

Prices, jobs, interest rates, the stock market. All of it ties back to two narrow strips of water on the other side of the world.

1. The first is the Strait of Hormuz.
On February 28, the United States and Israel launched military strikes on Iran. Within 48 hours, a senior IRGC commander declared the Strait of Hormuz “closed,” warning that any vessel attempting to transit would be set ablaze.

It is the main exit point for the Gulf’s oil. Since the U.S. and Israel struck Iran at the end of February, traffic through the strait has fallen sharply. On a normal day, around 17 to 20 million barrels of oil pass through it. That’s roughly 20% of global consumption and about 30% of all seaborne oil trade.

The Strait of Hormuz normally sees around 138 commercial vessels transit each day. Since the war began, traffic has collapsed, at times falling to just a handful of ships per day and even single digit transits over multi-day periods. Overall, traffic through the strait has dropped by roughly 90-95%.

The price of Brent crude, the world's most important oil benchmark, topped $106 per barrel on March 15. A year ago it was trading at $70.

2. The countries that have no way out.
The reason this is so serious is that most Gulf states have no backup plan. Iraq, Kuwait, Qatar, and Bahrain rely almost entirely on the Strait of Hormuz for their oil exports, with no alternative pipeline routes.

Saudi Arabia and the UAE have pipelines that bypass the strait. Saudi Arabia's East-West pipeline can carry up to around 7 million barrels per day, while the UAE's Abu Dhabi pipeline moves roughly 1.5 million barrels per day. Combined, that is under 9 million barrels per day, far short of the strait’s normal 20 million barrels per day.

The biggest buyers of this oil are China, India, Japan, and South Korea, with roughly 84% of all crude passing through Hormuz heading to Asia.

3. Now add the second strait.
In the first half of 2025, roughly 4.2 million barrels per day of crude oil and petroleum products flowed through the Bab el‑Mandeb Strait, the narrow waterway connecting the Red Sea to the Gulf of Aden.

That volume was part of global seaborne oil trade, which historically has seen around 10–12% of seaborne oil pass through this chokepoint. LNG traffic via Bab el‑Mandeb has been minimal in recent years as many LNG routes avoid the strait due to primarily security concerns.

Yemen’s Iran‑aligned Houthi movement, which controls much of the territory along the southern Red Sea has repeatedly attacked commercial vessels since late 2023 (Due to situation in Gaza), forcing many carriers to reroute thousands of miles around Africa. Since the attacks began, there have been well over 100 separate strikes on commercial shipping in the southern Red Sea, Bab el‑Mandeb, and Gulf of Aden.

On March 28, 2026, the Houthis launched ballistic missiles at Israel and basically joined the US/America/Iran conflict, saying they were prepared to escalate further. They have also threatened that closing the Bab el‑Mandeb Strait is an option.

If Bab el‑Mandeb were effectively closed, tankers from the Persian Gulf could not reach the Suez Canal or European markets without routing around the southern tip of Africa. This is a journey that adds weeks to transit times and substantially increases shipping costs across virtually every industry.

Right now, the world faces two major maritime chokepoints under simultaneous threat, and combined they account for nearly 25 million barrels per day of oil flows that are critical to the global energy system.

4. Can the World fall into recession because of Oil?
Yes. It has happened before and the conditions are now worse than any previous oil shock on record.

The 1973 oil embargo triggered a global recession. The 1979 Iranian Revolution did the same. In both cases, the actual supply disruption was relatively small. Roughly 4–5 million barrels per day, or about 7% of global supply at the time. What is happening now is larger in scale.

If the Strait of Hormuz stays closed and Bab el‑Mandeb goes with it, the world loses access to over 20 million barrels per day from Hormuz alone, plus several million more that normally move through the Red Sea. That’s why the International Energy Agency has called the current situation the biggest oil disruption in history, far exceeding the scale of the 1970s shocks.

There is no alternative route that can absorb that volume. The pipelines that bypass Hormuz can only handle a fraction of normal flows.

Reminder that Oil is not just a fuel that you use to run your cars. It sits at the base of the entire economic system. It affects manufacturing because factories rely on oil. It affects agriculture because fertilizers, tractors, and transport all rely on oil. It affects logistics because every truck, ship, and plane depends on it. Plastics, packaging, chemicals, heating —> all of it traces back to crude.

When oil prices rise, the increase spreads slowly through every layer of the economy. That’s why even a small supply shock can trigger inflation.

5. Could Oil hit $150+?
Oil is trading around $115, and getting to $150 would require both straits to remain closed or under sustained attack for months. That is an extreme case. But it is no longer unthinkable.

The base case is that oil stays elevated between $100 and $130 while the conflict continues. The extreme case, where both straits are disrupted for an extended period, is where you see $150 to $180.

Note, even the base case is enough to reshape central bank policy, consumer spending, and the broader market narrative for the rest of the year.

6. What this does to Inflation and Interest Rates.
Every central bank in the Western world was already thinking about pausing on rate cuts before any of this started.

Inflation had not come down far enough. Tariffs were already adding to price pressures.

If oil sustains at current level for several months, then inflation goes back up. It's as simple as that. Central banks do not cut rates into rising inflation. They hold. If inflation gets out of control, they raise. The rate cut narrative that markets had priced in for 2026 is essentially gone. Depending on how bad it gets, rate hikes become a real conversation again.

7. Two Scenarios for the next 3 to 6 months:

There are really only two ways this resolves.

A) Trump has said the war could end in weeks. If a ceasefire happens, oil drops quickly, shipping normalises, and inflation pressure starts to ease. Markets would rally hard on that kind of news.

B) The second is escalation -> Ground troops, drone war (similar to Ukraine vs Russia), strikes on Gulf infrastructure.

This is already happening to a degree. The longer it drags on, the more the economic damage compounds. Even if a deal is reached after three or four months of closure, the repair to infrastructure and the rebuilding of supply takes additional months on top of that.

8. What about the Economy more broadly?
Recession is the obvious fear but it is probably not the right conclusion. Consumer spending across Western economies is still running at good levels. Big technology companies have committed to spending over a trillion dollars on AI infrastructure in the coming years. That capital committed does not stop because of a conflict in the Middle East.

Western governments also have the tools to cushion the blow. Benefits systems in the UK and across Europe provide a floor for people who lose jobs. Governments will spend to protect citizens from the worst shocks. That prevents the kind of freefall that leads to a deep recession.

What is more likely is a slow grind. High prices, slower growth, rising unemployment as AI continues to displace jobs in sectors that used to feel safe. White collar roles in law, finance, and professional services are being compressed. Robotics has not even entered homes and care settings at scale yet. When it does, the unemployment will get noticeably worse over time.

9. What this means for investors?
Defence companies benefit directly. Energy producers with output outside the Gulf benefit from higher prices. AI infrastructure and power names hold up because the capex cycle is locked in regardless of oil prices.

The areas that struggle are anything rate sensitive, consumer discretionary stocks, and companies with complex supply chains running through the Gulf region.

The market does not crash in a straight line. I believe a deal with Iran will send oil prices down 20% overnight and trigger a sharp rally. But an escalation that closes both straits for months will send oil up +30% and stocks down a lot (potentially repeat of covid crash).

The volatility is the story for the next six months.

[PART 2]

If you read Part 1, you already know why I think this is the most serious macro environment in decades. This is probably the biggest geopolitical conflict since the Iraq War. But what makes this one different is the oil. When you physically stop the flow of oil through the world's most critical chokepoints, you are putting pressure on every layer of the global economy at once.

1. Portfolios are down.

Almost every retail investor is sitting on losses right now.

The S&P 500 is down around 7% from its recent highs. The S&P 500 is the most diversified, most liquid, most institutionally owned index in the world. When it drops 7%, that is a significant drop. Note that, when the S&P drops 7%, small caps and micro caps do not drop 7%. They drop far more.

$DUOL -46%
$HIMS -44%
$EOSE -66%
$IREN -25%
$PGY -50%
$ONDS -26%

Most of the hyperscalers are already down 10% or more YTD.

$MSFT -24%
$GOOG -12%
$AMZN -11%
$TSLA -18%
$META -17%

In that environment, small cap and micro cap names getting hit 50% or harder is not surprising. A lot of them have already hit that.

I am not surprised by any of this.

2. The communication out of the White House has been a mess.

You have mixed messages coming out of Washington and Tehran simultaneously, and honestly, I trust what is coming out of Iran more than what is coming out of the White House right now.

Trump says one thing one day and something completely different the next. Iran says there are no negotiations taking place. The White House says there are. Someone is lying, and the market does not know who to believe. Neither do I.

One thing we have all learned from following markets over the last few years is that markets hate uncertainty more than they hate bad news. Bad news you can price in. Uncertainty you cannot. And right now, uncertainty is at an extreme. This market is so sensitive to Trump's next statement that a single tweet can swing it several percent in either direction. That is not a good for investors.

3. So what am I actually doing?

I have deliberately sold 10% of my IREN position to raise cash. On top of that, my stop loss triggered at $33, which added more cash to the account. I am now sitting on roughly one third cash.

For those of you who have followed me for a while, you know this is unusual for me. I am not a cash holder. I have historically deployed capital as quickly as possible because I believe time in the market beats timing the market. I still believe that. But I also believe that catching a falling knife in the middle of an unresolved geopolitical conflict with no clear endpoint would be a disaster for my portfolio.

I am not deploying this cash until I see some resolution to this conflict. And I want to be clear about what that means. I am not waiting for a full peace deal. I am not waiting for oil to go back to $70. I am waiting for enough clarity that I can make a decision based on something other than what Trump might tweet tomorrow morning.

If IREN drops to $20 in a worst case scenario and then starts recovering to $25, I will buy it at $25. I am happy to pay more for a recovering stock than to buy into a falling one. I am not trying to pick the bottom.

I am also looking at pulling some money from my savings account into my investment account. I keep six months of salary in savings. I am thinking about deploying some of that into the market once conditions stabilise. Nothing is confirmed yet but that is where my head is at.

The U.S. defense industrial base has a manufacturing problem. It was built for a Cold War‑era pace of production, not for speed or flexibility.

Producing complex metal parts for missiles, aircraft, ships, and other critical systems still depends on traditional methods like casting and forging. These processes require custom tooling, specialized suppliers, and long lead times that can stretch into many months. In many defense programs, strict approval requirements and limited supplier capacity push timelines even further.

The Department of Defense has publicly acknowledged that current supply chains are not responsive enough for modern warfare.

Additive manufacturing directly targets this constraint by eliminating the need for custom tooling and enabling parts to be produced directly, compressing production timelines from months to weeks.

But speed alone is not sufficient. For a part to actually enter production and be used in defense or aerospace systems, it must go through a formal qualification process that proves it meets performance and reliability standards. That qualification is the core barrier to adoption.

Read more