Subscribe now

1. Anthropic is ahead for now, but I wouldn’t be so sure that will remain the case

2. Nvidia’s stock price looks fine, even though OpenAI now has their own chip that is exceeding expectations on performance

3. Sid Palas joins Trelis Converse. My friend Sid from MIT has joined as a founding engineer to lead infrastructure build-out. Delighted to be working with someone as capable as Sid.

Cheers, Ronan

Subscribe now

1. Poolside == Groq? I posted this as early speculation on Wednesday and the terms leaked the next day. Nvidia is paying $6bn for a non-exclusive licence to Poolside’s “Model Factory” (the system it used to build its Laguna models), hiring 109 of its staff, and investing $1bn in what remains at a $12bn pre-money valuation. The founders seem to be staying while the researchers leave for Nvidia. The $6bn is a licence fee that flows out to Poolside’s investors, and the $12bn is the valuation assigned to what’s left - a 1.2GW Texas data centre project and a new direction the founders say they’re not ready to share. It’s the third deal Nvidia has structured this way, after Groq ($20bn) and Enfabrica (roughly $900m): a licence plus a hiring round rather than a purchase, so no change of control and no regulatory merger review required!

2. Are tokens actually profitable? The tweet below is about Jevons paradox (when something gets cheaper, total usage rises so much that total spend goes up - coal in 1865, sugar in the 1800s, tokens now). Usage of GPT Luna went up 13x when the price went down 10x.

That’s fine for revenue, but what about profits? I dug into this a bit and the short version is that tokens are gross profitable at both Anthropic and OpenAI, and increasingly net profitable at Anthropic only. SemiAnalysis estimates Anthropic’s inference gross margins are now in the mid-60s, up from 38% in 2025 and negative in 2024, with the API business above 80%. A big driver is that Claude Code’s agentic workloads hit cache rates above 90%, so the true cost of serving Opus on agent tasks is around $1 per million tokens against a $5 list price. OpenAI is gross profitable on inference too (Altman: “If we didn’t pay for training, we’d be a very profitable company”) but is on track for roughly $14bn in losses this year, largely because free ChatGPT users consume compute against zero revenue.

3. Apple paid Ireland $17bn. No tweet for this one - just something i was thinking about! Apple’s first country-by-country tax disclosure shows it paid Ireland $17bn in tax last year, about 40% of its worldwide total, boosted by the EUR13bn in back taxes the EU’s top court ordered in 2024. Three companies (believed to be Apple, Eli Lilly and Microsoft) now pay almost half of all corporation tax collected in Ireland. Two things that get misreported: the 12.5% headline rate no longer applies to these companies (large multinationals pay a 15% minimum under the OECD rules since 2024), and the “double Irish” was closed to new entrants in 2015 but existing users kept it until 2020.

The irony is that Ireland fought for eight years not to receive EUR14bn (including interest), and after the loophole closed companies moved their intellectual property into Ireland rather than leaving, so corporation tax receipts went from about EUR4-5bn a decade ago to EUR33bn in 2025.

In the rear view mirror it ended up a very odd strategy that offering extremely low taxes ended up with quite a high tax take. I have a lot of mixed views on Irish corporate tax - one of which is that small countries should keep corporate taxes low as it’s one of few advantages they have vs larger companies. Another view of mine is that something would be wrong if tax receipts were not concentrated, because big company revenue follows power law type behaviour. You cannot (and don’t want to) escape the power law!

4. Coding plans versus metered tokens. This is pretty technical, but I asked whether anyone could vouch for Cursor over Claude Code or Codex.

Fixed-price coding plans are only offered by companies that own the model (Anthropic, OpenAI, xAI, Cursor, the Chinese labs), because they are subsidising their own inference. Neutral hosts like Fireworks or Together run someone else’s open weights on rented GPUs and have no margin to subsidise, so they meter. And metered is genuinely risky for agent work: a cheap model that gets stuck in a loop can burn millions of tokens on a single task (one open model averaged 4.8 million tokens per task in a recent Artificial Analysis run), and you find out about the bill afterwards. That is exactly what a plan protects you from (so long as they continue).

I use both Claude and Codex for my team at Trelis but we do exhaust credits, so we’re considering adding Cursor. We could add the chinese lab plans but then data becomes a risk. We’re not sure if Cursor (now SpaceX owned) is as good, and it’s primary form factor is not a CLI, so its less drop in as a substitute.

Cheers, Ronan

Subscribe now

1. Trelis Converse is Open (Invite-Only). This is the new voice venture I first wrote about in July, now taking in first customers - individual developers and enterprises - for our partner program.

2. Does Voice AI Hear You or Read You? I tested top voice models to see if they can repeat back words that sound different but are written the same.

3. Your AI Subscription is an Early Uber Ride?

4. Uncapped Notes.
An uncapped note (typically SAFE) is an investment where one agrees to invest at the price of the next round of investment. If there’s no future round of investment before an IPO or acquisition, you would just get your money back (with no further return). This is clearly very founder-favourable and investor-unfavourable. However, in highly valued startups where many investors want a part, it is sometimes the only way to get involved as an investor. On the surface it doesn’t make much sense, but it probably does make sense given behavioural aspects of private markets (e.g. founders want to always keep an upwards trajectory of valuation, which makes it unfavourable to explicitly price a startup too high or too low too early). Uncapped SAFE notes, to a degree, are a sign of strength.

5. Which Voice AI Models Respond Fastest? This is pretty technical, and Trelis Research has a horse in the race, but gives a feel for the quality of leading voice models.

Cheers, Ronan

Issue number 3 of this weekly roundup!

1. Why do voice agents sound so slow?

Fixing this is a big part of the new venture I’m working on. Details to come soon!

2. OpenAI is Google, Anthropic is Microsoft. The thesis here is that selling tokens may not be enough for OpenAI and Anthropic to survive, because there may be too much competition. Rather they might have to embed deeply within enterprise to make their services sticky.

3. Something I Wrawt

Wrawt a comment below! Ronan

Issue number 2 of this new weekly roundup I’m doing! As they say in the weekly wind-up writing business… Lez go!

1. The Anthropic Theory of Everything. You might remember FTX crypto exchange going bankrupt some years ago, with Sam Bankman Fried (SBF) as CEO. Well, the bankruptcy proceedings are ongoing and depositors are going to get fully repaid! This is thanks in large part to the fact that SBF had made a very early investment in Anthropic. That investment greatly increased in value during much of the bankruptcy and helped to pull the whole thing above water! [probably would have even bailed out equity holders had the shares been held for longer!]

More obscure, but an essay entitled Situational Awareness, written by twenty-something year old Leopold Aschenbrenner, went viral in 2024. He then raised $200M+ for an investment fund, backed by the Collison Brothers, Nat Friedman (former Github CTO), investor Daniel Gross and others. Since then the fund has skyrocketed to billions of dollars in value based on private investments (in Anthropic) but also public investments in AI stocks, often involving leverage (where he borrowed to juice his returns - yes, regularly slated as a strategy on this blog!).

Over the past month, when AI stocks fell in value, he was forced to sell most of his public stocks to repay borrowed money, resulting in losing over 2/3rds of the fund’s value! One of the positions he didn’t have to sell (cos it is private!) is Anthropic, and that’s a large part of what his fund has left! Apparently the fund is still up 80% for the year!

2. Fake it til you make it. Fish Audio is a speech AI company that raised $50M+ this week and had an incredible video (two deep if you click on these tweets) where they order pizza via AI. I wanted to know if the demo was real, so I took the audio and put it through their transcription service online. The transcript for the order didn’t match what the assistant repeated back, so it appears this demo was stitched. Does it matter? Not sure I’d do it, but very effective marketing.

3. Stripe buying OpenRouter - not yet a done deal. OpenRouter is a single service that gives access to many AI models (OpenAI, Anthropic, Grok etc.) for a small 5% surcharge. It’s useful because developers only need one API key to access thousands of models. OpenRouter uses Stripe to collect credit card payments. So of that 5% fee, probably 50-66% of that is going straight to Stripe! Good for Stripe. Now Stripe wants to buy OpenRouter!

Stripe sees - through their billing - how many companies use OpenAI and Anthropic and others. So - owning OpenRouter - they could presumably offer a very nice and high margin (for Stripe) service!

Reply if you believe in fakin’ til you makin’, Ronan

Holding just one stock clearly puts all eggs in one basket. Still, it’s a worthwhile thought experiment that brings long term thinking into focus.

Another educational framing is to ask which stock might have limited downside (sometimes companies that themselves are diversified) AND room for unexpected upsides. Try and think through this for yourself and then ask Claude or ChatGPT.

To be clear, I’m not advocating for nor have I ever limited myself to just one stock, BUT this kind of framing did lead me to make a few trims and changes mid-quarter - something I typically would not do.

More details follow for paid subscribers.

Read more

I’m trying something new where I’ll send out a weekly post highlighting of what I’ve been up to this past week. LMK in comments or replies if you like it.

Cheers, Ronan

1. Founding Team Nominations - I’m starting a new company. It will focus solely on conversational intelligence (voice AI) that is seamless and trusted. I believe voice will grow enormously in the coming years as our primary means of interacting with technology. I’m searching globally for a founding team across four specialties - evals & post-training, infrastructure, data, and architecture & mid-training - and I’m asking for nominations of exceptional people (you can write to ronan at trelis dot com). Full details in the post below.

2. Trelis Tara - frontier Hindi transcription. Hindi is the 3rd most spoken language in the world, although the state of Hindi AI models wouldn’t suggest it [they are not as good at hindi as english]. Trelis Research released a Hindi transcription model that outperforms ElevenLabs (the most well known voice company) on the Hindi benchmarks we tested and is competitive with Sarvam Saarvas v3 (a top Indian voice company).

3. Trelis Tiron - multi-speaker meeting transcription and attribution. This was a second release from Trelis Research this week. and something I’ve been working on for a while, since discovering a single tiny neural net can cleanly separate and transcribe overlapping speech. It specifically targets meeting with many speakers and overlapping speech.

4. BPBench - which frontier Large Language Models (e.g. ChatGPT vs Claude etc.) best compress text? This is pretty technical, but a large language model can be viewed as storing language in compressed form. Among models where full details are shared publicly (not ChatGPT or Claude unfortuantely), the Chinese labs dominate my measure of compression. Of Western models, only Nvidia’s Nemotron performs among top models.

One highlight of the quarter was the SpaceX IPO, which I didn’t get involved in, although previously I had been (an indirect) shareholder. You can read more at those two links.

Specific updates now follow for paid subs:

Read more

SpaceX goes public on Friday of this week and I have no plans to buy shares soon because I don’t see a way to recognise a good price in the short term.

TL;DR

• It’s hard to tell whether informed insiders think the IPO price is high or low.

• Initially, those who are best informed (Musk, key investors) are barred from selling. There will be no signal for a few months of whether they want to hold or sell.

• Many shareholders are rightly be expected to sell, because SpaceX has been private for so long and they will want some liquidity, somewhat irrespective of price.

• Perhaps some shareholders, especially funds without time horizons like Gigafund, Valor and Vy, will decide to hold and not sell, and that may give some positive signal (albeit only useful after lockups have expired) on price.

• Index funds are forced to buy SpaceX shares, but they buy roughly proportional to the float, not market cap, which makes the buying a much smaller number. Perhaps this upward pressure on price is overstated.

• Buying SpaceX stock early is a gamble that could go up or down, but has concrete downside because SpaceX is loss making, even if I believe in the long term future of data centers and space.

This isn’t advice on how to invest your money or someone else’s money. I’m sharing my experience so that it serves as an example to develop your own framework.

1. What’s a good price?

One can do all sorts of financial projections and discounted cashflows to get to a “fair value” for SpaceX, whatever that might be.

I don’t look at it that way. Instead, a good price for me is one at which those with deep insight into the company (management + current investors) are willing to a) buy more shares, or at least b) not sell many of their current shares.

Concretely, if Elon Musk decides to buy shares personally OR if SpaceX decides to buy back its own stock (as it did in late 2024!) then I would begin to consider buying some shares at that price!

Both are unlikely because a) I don’t know where Musk would get that cash from (although he could borrow more against his current shares) and b) this is an Initial Public Offering where SpaceX is explicitly selling - not buying back - its shares so that it has tons of cash on hand to invest in its business.

2. What can I tell looking at whether VCs sell their shares?

SpaceX has been around a long time as a company and many investors are sitting on large gains. For VCs, the IPO (and following months/year) is when investors get their money and returns back.

Many VCs funds are required or expected to distribute shares when their portfolio companies become public. So, probably a lot of VCs cash out (or at least distribute shares) to their investors over the next months, and that’s not necessarily a sign that the stock is overpriced.

Granted, SpaceX offered limited liquidity (i.e. the opportunity for existing investors to sell shares to other investors) over the last few years in private transactions. This somewhat mitigates the urgency for investors to cash in. Still, the IPO and beyond is an obvious opportunity for investors to cash in some chips - somewhat regardless of price - especially if they are already up a lot on their initial investment, which most are.

On the other hand - and I give this as a loose example - if some of the early SpaceX investors do not sell any (or many) shares, like Founders Fund, then perhaps I might consider buying.

3. If not VCs, then who should I look at for signals to buy/sell?

Perhaps:

• Gigafund + Vy Capital (+ maybe also Sequoia evergreen fund): These are not fixed term funds that have to give money back at a certain point. As such, if they sell, perhaps it’s less because of compulsion and more because the stock is at a high price.

• Antonio Gracias: A board member who may not have a structural/legal compulsion to sell his fund’s position. If he sells heavily, that’s perhaps a sign of a high price.

• Google: Unless the price is very high, my guess is Google should just hold and not realise taxes. If they sell a lot, that’s probably a sign the price is high.

Even still, it’s hard to conclude much from selling behaviour because SpaceX has been private for so long, and it probably makes sense for everybody to take some chips off the table.

4. The above signals will all be very slow.

Musk has signed up to not selling his shares for 366 days. For most others, selling is spread out over roughly a half of a year.

At the initial IPO on Friday, most investors won’t be able to sell much, so there is nothing to see in terms of signals.

And even when we get out to the 6 month and 1 year mark, when we can see who decides to sell, notice how most signals tell me the price is too high, not that the price is low enough to be a good deal. The only small indication of the price being good to buy is if I see VC funds that I would expect to cash out who are instead holding on.

5. Price is a slot machine, for now

Initial buyers will not be well informed, and I expect the stock price could go anywhere on Friday, up or down. For at least a few months, there will be few signals from insiders of what they consider a good price.

6. Will index funds prop up the stock price?

Index funds are forced buyers of any new stock that becomes public and enters an index. There is a narrative that this forced buying will prop up the SpaceX stock price. I suspect the effect is not as large as portrayed.

If index funds had to hold SpaceX in proportion to market cap, they have to shift a few percent of assets into SpaceX. Say the US equity market is $70T in size, and SpaceX is valued at $2T, theory says index funds should hold ~3% of their holdings in SpaceX.

Now, all index funds won’t hold SpaceX, but if they did, that might mean ~3% of the ~$15T worth of US equity index funds would have to flow into SpaceX - about $400B. That is a large number relative to the $75B of shares that SpaceX will sell in this IPO.

In practise, the flow from index funds is FAR smaller, for a few reasons:

1. Index funds typically include stocks proportional not to company valuation, but to the free float of shares, i.e. the amount actually traded on exchanges. Initially this will be $75B SpaceX, and index funds will tend to target more like $75B / $75T = 0.1% (or maybe a bit more, because that $75T denominator is lower when you only sum up free float). So, the back of envelope inflow from index funds is more like 0.1% x $15T = $15B at most.

2. Even this estimate above is high because many index funds either won’t include SpaceX or won’t include them for some time. S&P500 won’t include SpaceX for one year at least. Nasdaq-100 and FTSE Russell inclusion can happen around two weeks, but is float-weighted, making the buying much smaller.

And yes, as more shares unlock after IPO, the free float will get larger and so index funds will buy more SpaceX BUT because those shares have unlocked, there is more supply of shares to meet that demand.

In summary:

• SpaceX stock price could do anything over the first month(s). SpaceX is a loss making company, so it relies on its cash balance plus ability to raise capital. Insiders were transacting at $800B of valuation in December 2025, the xAI merger valued the combined entity at $1.25T (SpaceX ~$1T) in February 2026, and the IPO asks ~$1.75T - a significant further markup. So, buying SpaceX is a gamble with some concrete downside, even if I see real upside in space exploration.

• Over a six month to twelve month time period, we’ll start to get a sense for who is selling, and whether that indicates the stock price is high. There may be some signals of the price being low (if key investors hold on to their shares), but those signals will be murky.

• Ultimately, until a point where there are clear signals, like Elon Musk buying more shares, SpaceX buying back more shares, or other notable investors buying shares, it will be hard to tell what is a good price for me.

Subscribe now

Obviously humans run more slowly when covering longer distances, but there are fun nuances I was missing. Here are three running regimes:

1. Sprinting speed is fairly constant up to around 200 m. In fact, 200 m can be a little higher in speed than 100 m because the standing start is a smaller portion of the race. For example, Usain Bolt’s fastest 100 metres is the second hundred metres of his 200 metre record.

2. From about 200 m up to around a marathon, minutes per kilometre increases roughly linearly with the logarithm of distance. I don’t believe there is a clean model for exactly why this is the case, but this is what we see empirically.

3. At distances longer than a marathon, there’s an upwards knee in the curve of minutes per kilometre. This is partly because terrain is less smooth, but also because because of depleting glycogen and instead burning fat, muscle damage and limits of the gut (GI).

Subscribe now

I’ve written before about how I try to hold stocks for as long as possible to avoid rash trading and constantly realising capital gains.

I try to keep my rule for selling simple: if the founder CEO steps away, I sell. Not because the company is bad, but because the specific edge I’m betting on - a founder with outsized ownership, control, and long-term thinking - is no longer there. A few quarters ago, that rule triggered on Spotify in my growth fund (Daniel Ek moved to Executive Chairman) and led me to continue selling down my position. In my personal fund, it was the reasoning behind starting to trim Berkshire last year after Buffett stepped down. In an unexpected and sad way, that happened again last quarter when David Simon, the CEO of Simon Property Group, passed away.

I also broke slightly from my preference for founder-CEO owned stocks when I added small positions in ASML and Intel - each about 2% of the portfolio. My thesis isn’t about these companies individually. It’s a (not confident) hedge against Taiwan risk. If something disrupts TSMC - whether geopolitical or otherwise - ASML (a key part of the supply chain for machines to make advanced chips) and Intel (the only Western company building leading-ish-edge fabs) are maybe positioned differently than most tech stocks. It’s also a small way to get some AI exposure in a portfolio that is otherwise very traditional.

Specific updates now follow for paid subs.

On the personal funds side, I trimmed my Simon Property Group position slightly and redeployed into TIPS, EUR, Intel and ASML. On the growth funds side, I continued the Spotify sell-down, opened positions in two new founder-led companies, and have a reflection on a private investment that didn’t work out.

Read more

My prompt:

Let’s say that Ireland immediately had no more access to gas or oil, can you walk me through the steps that would happen? And walk me through what the price would be for electricity in the country, you think? Try to make your best estimate. Tell me what happens on the first day the first week the first month the first year and where we are at the end of each in terms of rough electricity prices, and also in terms of the sources of power in play. I also want you to factor in policy changes that might occur with respect to importing any types of goods, including those that might have tariffs on them right now.

Also, give me your best estimation of where things would be in 5 years. And also in 10 years. Actually, make it 20 years.

Claude’s Response (TL;DR below, full response here and worth a read):

Day 1: 40% of electricity generation vanishes. Rolling blackouts begin (4-10 hrs/day). Transport grinds to a halt — 91% of vehicles can’t move. Supermarket shelves emptying by evening. Price unchanged (contracts hold). National emergency declared.

Week 1: Managed blackout regime. UK interconnectors maxed out. Food distribution via Defence Forces. Government suspends all tariffs on solar panels, batteries, EVs, heat pumps. Wholesale price 5-10x normal.

Month 1: Economy in freefall, unemployment spikes to 15-20%. Emergency battery storage and biofuel generators trickling in. Retail price effectively doubles via a tiered system (~35c for a lifeline allocation, €1+ above that). Power mix: wind ~52%, imports ~32%.

Year 1: First winter is the human catastrophe — excess deaths among elderly, political crisis. Heat pumps being installed frantically but only scratching the surface. EVs triple to ~350k. GDP down 15-25%. Retail price ~55-70c/kWh. Wind ~48%, imports ~22%, solar ~10%.

Year 5: Offshore wind boom transforms the picture. 40-50% EV penetration. Heat pumps in over a million homes. Economy recovering, restructured around green energy. Retail price back near pre-crisis levels: ~30-45c/kWh. Wind ~62%.

Year 10: Post-fossil electricity system. Net exporter on windy days. 80-90% EVs. Heating mostly electrified. Retail price below pre-crisis: ~25-35c/kWh. Wind ~72%.

Year 20: Ireland is the Saudi Arabia of Atlantic wind. Major electricity and green hydrogen exporter. Energy sovereignty is now a competitive advantage. Retail ~15-25c/kWh. Energy poverty largely eliminated.

Bottom line: Catastrophic short-term pain (especially that first winter), but Ireland’s wind resource is so exceptional that by Year 5 it’s recovering and by Year 10 it’s genuinely better off than before. The crisis compresses a 30-year transition into 5-10 years at enormous human and economic cost.

I asked ChatGPT (5.4 Extended Thinking) to critique the response but it got knotted up on whether this is “no domestic access” vs “global fuel exists” [fair but pedantic] and also on whether, legally, Ireland could waive EU tariff law (which I assume would be irrelevant given the devastation).

I resorted to Gemini Pro 3.1 for a critique (TL;DR below, full response here):

Day 1 to Week 1 (The Physics Failure)

• Total Grid Collapse, Not Just Blackouts: The grid doesn’t just face a supply shortage; it loses its “synchronous inertia” (the heavy, spinning turbines of gas plants that keep the grid frequency stable at 50Hz). The system completely trips.

• Retail Nationalization: Retail suppliers don’t just “absorb the hit” for a few weeks. They go bankrupt immediately, forcing the state to instantly nationalize the energy sector.

Month 1 (The Supply Chain Illusion)

• No Battery Airlifts: You cannot magically airlift hundreds of megawatt-hours of batteries in a few weeks. They weigh thousands of tonnes and have a 1-to-2-year global backlog.

• Agricultural Collapse: Modern farming requires diesel (tractors) and gas (fertilizer). Supermarkets don’t just face logistics delays; the domestic food supply chain dies, pivoting the crisis from an economic recession to a massive famine threat.

Year 1 (The Bootstrapping Paradox & The LNG Cheat)

• You Can’t Build Green Without Black: You cannot pour concrete for wind turbines, clear land for solar farms, or move 70-metre blades without diesel trucks and excavators. The massive Year 1 renewable building boom is physically impossible.

• The LNG Loophole: The estimate cheated by assuming Ireland would build a floating Liquified Natural Gas (LNG) terminal. The prompt strictly said no gas or oil. Without LNG, the grid completely dies every time the wind stops blowing (Dunkelflaute).

• Civilizational Reset: It’s not a 25% GDP drop. It’s a wartime rationing economy heavily reliant on EU/UK military aid, resulting in mass emigration to escape starvation and freezing winters.

Year 5 (The Offshore Wind Fantasy)

• No Installation Ships: The 3-5 GW of offshore wind predicted by Year 5 cannot happen. The massive, specialized ships required to install offshore wind turbines run exclusively on marine gas oil. Without oil, they cannot operate in Irish waters.

Years 10 to 20 (The Timeline Reality)

• Delayed Utopia: Because fossil fuels are required to manufacture, transport, and install green infrastructure, the “Green Tiger” utopian recovery is delayed by decades.

• The True Aftermath: It takes 20 to 30 years to recover—not 5 to 10. The transition is only made possible by foreign humanitarian aid doing the heavy lifting, prefaced by catastrophic population loss and deep societal trauma.

My own view, closer to Gemini’s, is that:

1. Suddenly left without fossil fuels, Ireland would be much worse off in the short term than what Claude portrays. The grid would not remain stable, and industry and households would be crushed. Pharma, Manufacturing, IT would do very badly.

2. As we saw with COVID, in some ways the response would be slow (i.e. contradictory government guidance on how to act) but in some ways things would adapt faster than expected (i.e. restaurants moving to takeaway, business going online, vaccines emerging quickly). So I would expect the 1 month to 3 year response to be better and faster than what Claude writes out, with creativity around finding alternative energy (possibly solar panels, and indeed biofuels).

What does this thought experiment tell us?

We can glean from it some wise actions for the present moment, especially in the light of today’s Gulf War.

A do-everything approach seems in order:

1. Solar Panels + Electric Cars:

   1. Lobby the EU to drop Chinese tariffs (more realistic now with the Gulf War). Note, you can see already the dynamics here of Europe re-approaching China in the next decade.

2. Wind + Solar + Batteries:

   1. Ramp up subsidy programs that heavily weight availability and dispatch-ability during hours of need, not just a levelised price. These probably should be higher than seems palatable or reasonable.

3. Increase interconnection with France, UK and more where possible. Build excess interconnection, but don’t rely on it for more than a certain amount of regular power.

4. Smart meter customers defaulted to time-of-use tariffs, but can opt for flat fee if they wish - which electricity suppliers must hedge 12 months ahead to guarantee.

5. Fossil Fuels:

   1. Accelerate building Liquid Natural Gas Terminals to better diversify natural gas imports.

   2. Open up Irish oil + natural gas exploration.

6. Fission:

   1. Repeal the 1999 act blocking nuclear and set up a dedicated group within the Irish grid operator to immediately liaise with South Korea, Estonia, Finland, Canada and Poland to establish a partnership around proven nuclear fission reactors (~300MW size). Immediately start a nuclear engineering program in an Irish university at the undergraduate and postgraduate level.

Read more

The question is - did this portfolio outperform the S&P500 over the 2021-2026 period?

My public stocks portfolio provided an IRR of 8.1% after fees and taxes, compared to a synthetic 100% S&P500 price portfolio (money-weighted IRR under my specific cashflow timing), which would have returned 7.1% before dividends, taxes and fees. [A buy-and-hold S&P investor with a lump sum in Jan 2021 would have experienced a materially different return, as would my portfolio had I initiated my positions in Jan 2021. This is why I compare based on the same cashflow timings.*]

And so, over this time period:

1. My portfolio performed well, driven by my large purchases of Spotify and of Meta (which together generated 3-6x returns over the 2022-2026 period).

2. Given the 8.1% IRR included a substantial drag from cash holdings of 15-40% over the period (and taxes), the performance of the stocks I chose significantly outperformed the S&P500 (~21% IRR considering cashflows into and out of stocks alone**).

3. Even with this very good/fortuitous choice of stocks, allowing for the drag of cash, and the reality of paying taxes, it shows how hard it is to beat the S&P500 benchmark.

Good performance here is no guarantee of my skill in stock picking going forward. There are many stocks one could have chosen in the 2021-2023 time-period that - if held as a large chunk of one’s portfolio (~25% each at initiation in late 2022 for Spotify and Meta in my portfolio) - would have outperformed the market if held until today. Holding Nvidia, for example, would have generated even higher returns still. If I were to take any credit for performance, it’s that I betted big on companies that were cashflow positive, at a reasonable price (hard to gauge), with CEOs that had a substantial shareholding (incentives aligned) and substantial managerial control. Those concentrated bets delivered this high performance. Of course, concentrated bets can also deliver terrible performance!

Choosing, Timing & Sizing.

To do well in investing, there is:

• the choice question of which stocks to buy / sell,

• the timing question of when to buy / sell, and

• the size question, of how much of one’s wealth to allocate to each investment.

One needs all three, but perhaps - of the three - it’s the size question that is most under appreciated. To do well in (active) investing, I need to choose the right investments, buy or sell at the right time, and allocate the right weight in my portfolio to each of those investments. Usually this means investing more when I’m more confident and/or the price is cheap, and vice-versa. It’s a truism, but the size question probably where some good investors distinguish themselves - betting rarely but heavily.

Regarding portfolio risk and the danger of measuring performance

This is not my only pool of funds for investment. It would be risky if it were. It is the specific pool I have allocated to picking public stocks and drawing a comparison with the performance of a broader index. Even in this pool, I hold a significant amount of cash - often 15-35%, which drags down portfolio return, but also allows me to be more aggressive with concentrated bets.

Numbers mean something, but they mean a lot more when they cover performance over longer periods, much longer than the 4-5 years I am showing here. When there’s a Berkshire-Hathaway-length track record over decades, and that outperforms the S&P500, that’s when skill becomes measurable and meaningful.

At the same time, uncertainty abounds in investing, and I never rely solely on numbers. A track record is a useful measure of the past, and gives some indication of the future, but I prefer honing a strong set of principles and heuristics. To make that concrete: I see in the data that holding 100% stocks outperforms most other strategies over the last century, but I still hold about one third reserves (currencies, gold, bonds) in my personal portfolio because I think that’s more robust looking forward.

*Comparing portfolio returns to the performance of an index is complicated if cash is added to or removed from the portfolio during the period under consideration, i.e. in the form of deposits, withdrawals, taxes and fees. There are always shortcomings in how one might define an apples-to-apples baseline. I chose two means of comparison. 1. I looked at the cashflows (converted to USD) into and out of the brokerage account and calculated an IRR on that basis, as well as a synthetic IRR assuming all deposits and withdrawals were from a portfolio that is 100% the S&P500), 2. I considered only the return on stock holdings (not cash holdings) and looked at the flows of cash into (stock purchases) and out of (stock sales) those holdings to calculate IRR, also calculating a synthetic IRR for a 100% S&P500 portfolio. Another relevant - although not directly comparable - benchmark is to look at annualised returns of the S&P500 over the same period of ~2022 → 2026. This yields higher IRRs because it only considers stock market valuations at the start of each year. I made large deposits into my portfolio in late 2021 and early 2022 when stock markets were at a high, meaning that baseline compounded annual return of the S&P is low.

**This stock-only IRR is a like-for-like stock selection comparison against the S&P500, conditional on my actual timing, and is useful for isolating selection effects independent of cash drag and taxes.

Specific updates now follow for paid subs:

• On the personal funds side, I continue to manage my Berkshire position as Buffett has retired from CEO (as of January 2026). I provide some updates as well on my gold holdings.

• On the growth funds side [including the public stocks portfolio I reference above, as well as a private stocks portfolio], I continue to take profits in places, and opened up one new small private investment, as well as making a new investment as an LP in a pre-seed fund.

Read more

It’s hard to justify any case for murder, or even for drunk driving. With social media, you can learn, make friends, build communities and create. You can also bully or be bullied, be absorbed into good or bad behaviours, live meaningful or dangerous experiences, and grow or fall in self-confidence.

So, aside from obvious challenges with enforcement - such as needing online identification for everyone, and the existence of VPNs - I take specific issue with this approach of compulsion rather than helping children guide their usage through good reasoning.

The intent is to protect children, but many will - for very good reason (because there are good uses) - want to work around these laws. And, we should always question whether we want laws that are respected out of fear or laws that are respected for good reason.

I, controversially, suggest we start with treating children as people, believing that virtues and dangers can be explained - although not always followed, and leading more by example.

Rather than gumming up society with single-swoop legislation like GDPR and DSA, we need more humility and responsiveness in our laws, testing steps that are adaptive and reversible, not utopian schemes we’ll be stuck with once we prove once again we are fallible.

However, that Ireland is unable to support the connection of new data centres, nor protect undersea cables is undercutting both reality and brand.

Ireland has been unable to jointly weigh and confront decisions around security, economic strength and the environment. In one part, this is because decision making is siloed across independent bodies, with limited Dáil (parliament) authority to resolve decisions that interact. In other part, we do not often provide the arguments for how to weigh these decisions, because it is easier - in the short term - to pretend there are no conflicts at all.

In this piece in The Currency, I start with why Ireland should continue, not just have data centers, but many more data centers per capita than other nations. This is both specialisation and strategy at work:

“From Ireland’s – and the EU’s – standpoint, there are strong strategic reasons why Ireland should want more data centres. This starts with the strategic benefits of data and compute sovereignty. In a geopolitically reorganising world, countries slowing or blocking the build-out of data centres are blocking today’s primary avenue of economic growth and relative economic strength.”

I note how this challenge is the same as faced by households with high energy prices:

“The challenge faced by households and by data centres is the same, namely that the power supply is held down.”

I move on to the reasons why data centers largely cannot connect to the grid, namely, barriers to building new power generation:

“New data centres are unable to obtain a grid connection because the Government decided to transition to a renewables-powered grid, while throttling the ability to build new natural-gas plants and liquid natural gas terminals. The key motivation for the shift to renewables was to reduce carbon dioxide emissions. The under-appreciated challenge (although not by grid operators!) with renewables is that Ireland is not located near the equator.

…Ireland [has] more wind than sun, and we also have seasons. We see hourly variability, multi-day wind lulls, and strong seasonal variation because North Atlantic wind is stronger in winter and weaker in summer.

…For these reasons, and this was true prior to the latest surge of data-centre growth, a grid with increasing renewables penetration is reliant on natural gas.”

And finally, I move to my recommendations around liquid natural gas terminals, weighing grid incentives for renewables towards availability, and removing the ban on nuclear development in Ireland so we deeply evaluate all possible options for our energy future.

I’m not fixed on my recommendations or views on the above. In writing the piece, some of my views changed, including what is possible or not with a carbon tax, and whether renewables contribute or hurt energy prices and independence.

I conclude:

“Data centres are not a play for jobs, they are a play for data security and reinforcement of a reputation where other nations can entrust their data and their compute. Balancing security, economic and climate issues is not an easy task, but it is important.”

As wages rise in software, Baumol’s argument is that wages will also tend to rise in these other sectors. Otherwise, workers would shift towards software. The way employers retain doctors, nurses and teachers over time is by offering them higher wages too.

It’s called “Baumol’s cost disease” because the theory involves wages in these sectors rising without productivity rising nearly as fast. Teachers still roughly teach the same numbers of students, and nurses see similar numbers of patients. So the cost per student or per patient rises, whereas, in software, costs per unit tend to fall.

In this era of large language models (unfortunately termed “AI”), there is renewed attention on what technological progress means for jobs and what it means for mobility (the ability for those who start poor or in lower-paying jobs to rise to higher-paying ones).

A salient question seems: In addition to progressive income taxes, what is it – if anything – that helps to compress the income distribution? The Baumol effect is perhaps one such example, at least in the sense of compressing wage gaps between high-productivity and low-productivity sectors.

To get a little more technical:

• Baumol doesn’t preclude that the very top performers in a given sector will pull away in terms of earnings. Any given “sector” also has a finite life – what’s hot in technology tends to change (otherwise it wouldn’t be technology).

• What Baumol does imply is that wages in both productive sectors and these labour-intensive service sectors go up*, so the difference in earnings is smaller than it would be without the Baumol effect.

• While wages go up, costs also go up. One may earn more as a childcare worker, but the cost of childcare and healthcare also goes up. In a simple two-sector economy (a “productive” sector plus a “stagnant” service sector), workers are better off in aggregate – per the theory – because one sector now has lower costs relative to wages, while the other sector has roughly constant costs relative to wages**. So long as workers buy something from both sectors, they can be better off. (The exact proportions they buy from each sector before and after matter for how much they gain. This is relevant because lower income workers tend to have purchases weighted more towards lower productivity sectors, such as healthcare or childcare.)

I’m not at all the first to point this out. Sissons, Helland & Tabarrok, and Worstall make similar arguments, including the suggestion that Baumol’s Cost Disease should more simply be called the Baumol Effect.

But is Baumol’s Effect empirically true?

There is strong evidence that prices in labour-intensive services (healthcare, education, childcare) rise faster than goods (see Figure 1), and that these sectors absorb a growing share of employment and spending.

However, showing price (or wage) increases over time - necessary if Baumol effects are present - is very different from proving Baumol effects are the cause, or even a significant cause of those rises. In any sector, there are a wide range of effects, e.g. workforce education, regulation, taxes and tariffs that influence productivity and wages. Picking apart any one effect is difficult, and something I would need to dig a lot further on for Baumol, if it is possible at all to see strong effects. For example, in the housing sector, cost increases are likely dominated by planning and zoning rules, labour shortages, or difficulties in providing utilities, rather than by Baumol effects that drive up wages owing to IT/finance/pharma sector growth.

For now, I see the Baumol Effect as a theory supporting the importance of worker mobility and developing worker skills for high productivity sectors. Contingent on those being strong, there can be gains for workers who move and potentially for those who don’t. But, in many sectors, there are likely more important explainers for trends in wages and costs.

Given present attention on what AI means for workers, it’s well worth better understanding the forces driving income growth and distribution. Perhaps Baumol is of help, but not our saviour.

Notes:

*For wages to rise in these service sectors requires that it be possible for at least some workers to switch into the more productive sector. Only because this switching is possible is there an incentive for employers to raise wages in services. This has implications for education and employment policy: anything that increases skills relevant for the productive sector or reduces switching barriers (I won’t go down a rabbit hole here) increases worker mobility and tends to drive up service-sector wages.

Further Reading:

• Andrew Sissons — “The Baumol effect: not a disease but a cure” (Medium)
  https://acjsissons.medium.com/the-baumol-effect-not-a-disease-but-a-cure-c63d0ae0b481

• Eric Helland & Alexander Tabarrok — Why Are the Prices So Damn High? Health, Education, and the Baumol Effect (Mercatus Center / PDF)
  https://www.mercatus.org/system/files/helland-tabarrok_why-are-the-prices-so-damn-high_v1.pdf

• Tim Worstall (Adam Smith Institute blog) — “Add Solow to Baumol and you’ve really got something”
  https://www.adamsmith.org/blog/add-solow-to-baumol-and-youve-really-got-something

“While financial bubbles are dismissed as irrational excess, we often overlook their positive aspects. Bubbles serve as a coordination mechanism to funnel ideas and attract resources towards a common cause. Bubbles were involved in the build-out of semiconductor fabs, the shale boom, and a bubble is now well underway with AI and the AI data-centre build-out. Bubbles have winners and losers. Until a bubble plays out, it is not entirely clear who those winners and losers will be. Will it be those who make the hardware, like Nvidia and AMD? Or those who train the models, such as Anthropic, Google,…”

“...I have almost exclusively used Nvidia GPU chips for training AI models. While I have used many services for coding (e.g., Windsurf, Cursor, Claude Code, OpenAI Codex), I have almost always used ChatGPT for my chat queries. Perhaps the day I move from ChatGPT and Nvidia is the day for me to be really concerned about their future.”

Full article ($): https://thecurrency.news/articles/206004/would-we-be-better-off-without-this-ai-bubble/

I voted for Catherine Connolly because she was clear on her views and, I felt, gave consideration to opposing views. Her slogan was “A President for All” and I felt that was genuine. I liked that she ran as an independent, and had been independent for quite some time since leaving the Labour Party. The parties endorsing her are socialist, and many of her views are not in line with my own, but that was less important for me here, on balance.

The alternate candidate, Heather Humphreys would have been fine too. I appreciate that both of them ran for office. Not a lot of people want to run for office (and there is debate about the presidential nomination process). It is stressful and somewhat thankless to be in politics, but important.

“Ireland poised to pick hard-left president”

That is the headline in the Financial Times, and here are some reasons for that branding, all quoted from the FT piece:

1. Palestine. “The staunchly pro-Palestinian politician has also branded the US, UK and France untrustworthy over their stance on Gaza”.

2. Neutrality and rearmament. “The former barrister and clinical psychologist has blasted the EU’s military build-up in response to Russia’s war in Ukraine, saying the bloc “has lost its moral compass” and that German rearmament spending was reminiscent of the 1930s.”

3. Her trip to Syria. During the campaign Connolly faced fierce criticism over a “fact-finding” trip she made to Assad-era Syria in 2018;

4. Her decision to hire a dissident republican convicted of gun crimes to work in her parliamentary office;

5. Her work as a barrister representing banks at a time when homes were repossessed during the financial crisis.

I’ll go through each in turn.

Palestine

I lived in the US for 10 years and in Ireland for 25. There is a wide gulf between median views on the Israel-Palestine conflict in the two places, although the range of views in the US is quite a bit broader.

In Ireland, there were and are few Israelis or jews, although the current Israeli president’s father grew up in Ireland, and his grandfather spoke fluent Irish. Irish people are reasonably well informed on the second world war and the Holocaust. Speaking for myself, that was largely where my knowledge of Israel and judaism stopped. My knowledge of the history of Palestine was near zero, and I knew no Palestinians in Ireland before going to the US. In the US, I worked with many Palestinians and Israelis and have close friends who are Palestinian and who are jewish.

It’s tempting to cast Irish solidarity for Palestinians in colonial terms, i.e. Ireland was a British colony and therefore there is solidarity with Palestinians who are oppressed by Israel. While this angle likely has some influence, my experience growing up in Ireland was not colonial. There was and is the odd joke at the expense of the British, but the Irish – particularly with Sky, UTV and Channel 4 via cable TV – felt aligned and friendly with the British. Now there is Netflix, and Ireland – like much of the world – has become more culturally American.

My grandparents did not have much to say about the British rule in Ireland. It was before their time, and well before my time. They did live through “The Troubles”, the period of violence in from ~1960s-1990s that was largely confined to Northern Ireland, but affected the Republic and England too. My grandparents and parents remember the troubles well, although were not directly affected much. I was born in 1989 and grew up in the Republic, unaffected by violence. My only memory is of going to a protest with my Dad in my local town. It was a protest against IRA bombings in Omagh in the 1990s. I learned today that a friend of mine was due to be in Omagh on that day, but he broke a window while playing and the family got delayed by four hours, missing the explosions.

As I see it, the Troubles provide a more recent and influential lens for Irish people on the Israel-Palestine conflict than does the history of British Rule. There are three lenses that stand out to me as affecting Irish thinking:

1. Irish people dislike the IRA. There is no social doubt that a person genuinely shouting “Up the RA” is a nutter to be wary of. Irish people know the IRA killed and tortured civilians, British and Irish. Irish people also know that the IRA’s actions did not justify the killing of Irish people – directly, or accidentally – by the British Army. In the US, where there is compassion for Gazans, it is sometimes from the viewpoint of Palestinians being collateral in a proxy war led by Iran and/or as collateral damage trying to take out a Hamas government they themselves elected. While there is logic to those perspectives, you can see why this might not be the primary lens through which Irish people see the deaths of tens of thousands of Palestinians.

2. Some US media narratives treats Hamas as a static or traditional army that can be “defeated”, for example by taking out all of its soldiers or its leaders. By contrast, the Irish experience on the IRA is more dynamic – until there was peace and (somewhat) trusted institutions, there was an ever-renewing source of IRA leaders and “soldiers”. The principle of “never negotiating with terrorists” has good first-order logic to it, but – in some ways – negotiating with terrorists is how peace came about in Northern Ireland. I grew up in peace in the south of Ireland and can only imagine, but that must have been incredibly difficult when there were Irish and British people whose family members were tortured, killed and kidnapped by the “other side”.

3. There is a perverse lens that the Irish troubles and colonial history bring to the Middle Eastern conflict, namely that “One man’s terrorist is another man’s freedom fighter perspective”. Again, while there is truth to this angle, it is an endorsement of violence and a step backwards. While most people in Ireland condemn Hamas’ violence, there are more people than I would like that are ambivalent or justify it.

Neutrality and rearmament

I can agree – although am not entirely confident – that Ireland should remain neutral. Obviously, we are not neutral in many ways, because we rely (relied?) on the UK and US for security, and – as part of the EU and other organisations – we clearly have economic relationship and preferences.

I largely don’t share Connolly’s staunch criticism of NATO expansion and EU militarisation, but it is criticism that should be taken more seriously. Can the EU and NATO continue to expand an economic zone and military alliance without expecting some geopolitical pushback? Even if there are moral and/or political reasons to expand, can the EU and NATO expand without being adequately resourced – militarily and economically?

Connolly’s criticism of Germany’s current rearmament is partially right, but mostly wrong. It is wrong because there is a military threat to European countries from Russia, and, one cannot repel a military threat without having a solid economy and military. She is right that armament is dangerous for the world, and it increases risks for everyone. It’s a tragedy of the commons, and you can’t overcome that tragedy by giving away your land. She is also right to be careful and skeptical of growing military spending. Defense contracts easily become a grift, syphoning off taxpayer money to defence contractors whose incentives are to drive up costs and increase war.

The best scenario is for Ireland to increase defence and infrastructure spending as the Finnish do, trying to make as much spending as possible dual purpose, to serve both civilian and military purposes. I do worry this is far from a free lunch. Military spending is a high additional cost to impose on economies with already high taxes and sluggish growth.

Connolly’s trip to Syria.

I’m skeptical of critiques of person X for meeting person Y. Society has come to view meetings of X with Y as endorsements by X of Y. This is unhealthy because – at least in my business experience – I have rarely had a worse outcome by meeting people in person. Perhaps it is even more important to meet people we disagree with or are in conflict with.

That said, I don’t understand why Connolly went on that visit and why she met the people she did. Without digging deep on the visit, it seemed quite out of place. What exactly was/is Ireland’s relationship with Syria at the time of Bashar al-Assad, if any?

Connolly’s decision to hire a dissident republican

This one made me feel worst about voting for Connolly. She hired a lady who had been convicted of gun running, although had served her time and was said to have been rehabilitated. That lady came recommended by Eamon O Cuiv, a Fianna Fáil (main Irish party) TD. This strikes me as an odd fact pattern and a dubious choice.

Connolly’s work as a barrister representing banks

Heather Humphreys’ targeting of Connolly for representing banks as a barrister during the financial crisis made me want to vote more for Connolly than for Humphreys.

Barristers’ jobs are to represent their clients in matters of law. The rules around eviction of tenants is set in law, and those laws can be updated and set by the Dáil (parliament). It’s true, you don’t have to choose banks as your client if you are a barrister. Still, it strikes me as unhelpful if serving banks disposing of property after a crisis is deemed socially unacceptable. There is a structured way to handle the difficult situations when loans cannot be paid, and legal representation and court cases provide that mechanism. If the laws are bad, then change those laws. If anything, that Connolly has this experience suggested to me she has some real world experience of thorny issues on housing and banking.

Connolly on Housing

Connolly opposed ending a 2023 moratorium on evictions. Yes, eviction moratoriums reduce the suffering of those risking eviction at a given point in time, but that comes at the expense of there being less rentals and less housing overall.

A law could also be passed that food manufacturers have to continue providing food even if customers stop paying. Pretty soon, there will be less food manufacturers, and less food.

If there are 70 homes and 100 families who want homes, it doesn’t matter what the eviction policy is, there are going to be 30 families that don’t get homes. The better policy is the one that best closes the gap over time between the homes there are and the homes people need. Ireland’s population, and demand for homes, is growing fast. We need more homes for purchase and for rent, not less homes and less rentals.

I am somewhat encouraged by new apartment buildings I see going up in Dublin, but pessimistic that Ireland can sufficiently increase housing supply. While price controls reduce the supply and quality of homes, the planning process is the larger barrier to increasing supply. Sadly, perhaps the planning system in Ireland does represent popular views. Many people don’t want building in their neighbourhood, don’t want build-to-rent developments – because they block home ownership – and don’t want build-to-buy because they are too expensive for people to afford. There is also an unfairly negative view of builders and developers (not unique to Ireland), which can mean ambivalence when it comes to supporting planning application and resisting objectors.

We could revamp the planning system, but perhaps that only can start with a more positive attitude towards development, and more social pressure on the concentrated groups that do block planning permission and building.

While the President does not create or directly influence policies, I am concerned that policies I see as reducing the supply of housing will be promoted.

The council of state

The President nominates members of a council of state – in addition to statutory members, such as former presidents – to provide guidance on what bills to refer to the Supreme Court. In the debate, both candidates were asked who they would appoint and neither appeared to have given it much thought. Humphreys made the point that the council should be 50-50 men and women, and suggested someone with climate change experience should be included. Factoring in a variety of perspectives seems important, but I’m often skeptical of strict gender quotas. My experience - in education and business – is that people don’t think the same of those who are chosen under a quota system. I’m not holding this view too strongly though. Maybe it’s fine or even good to have things 50-50. That said, climate change and gender quotas wouldn’t be the first things that come to mind for choosing a panel to assess constitutionality.

A high variance pick?

As a friend of mine put succinctly, Connolly is the high variance pick. There is upside if she can listen well, moderate and represent a variety of views. There is more downside if people latch on to her views around defence spending and housing. I’ll come back to read this in a few years, both to see how the presidency and my views have unfolded.

This quarter, I have one short thought on tax. My approach around tax and investing is to:

a. Try and hold stocks for as long as possible to avoid constantly realising capital gains from trading.

b. Try not to over-optimize for tax otherwise*.

Towards the end of each financial year, you’ll hear chatter of investors selling assets to realise losses that can be offset against any gains. The maths makes sense (although can be limited by law as to how this is done), but I worry that if I start optimising for this it leads to a) things being more complicated and b) worse decision making. For example, I might delay selling a stock until the new year - to delay taxes from this year - only to find the stock price has fallen in January. Simpler is often better.

*Although, the punitive tax treatment in Ireland for holding index funds does shift me more towards owning individual stocks. Possibly this will change within a few years.

Spotify

In news this week, Spotify have announced that Daniel Ek will move from CEO into the role of executive chairman. Spotify stock has performed strongly over the past years, increasing from around $90 when I initiated my position a few years ago to just under $700 at present. I don’t fully understand where this valuation increase has come from, but that’s not my strategy. My strategy is to focus on cash-flow positive founder-led companies and wait for a few to occasionally unlock new products and directions that are not foreseen by anyone. In this case, it wasn’t a new product or direction that led to the Spotify stock price increase. Probably the stock rise was more attributable to subscription price increases and steady subscriber growth. So, I’m mostly chalking this down to my luck.

For interest - here was my note regarding Spotify last month to paid subscribers:

Spotify has a stock repurchase program but it is largely unused. The CEO has sold stock at a wide range of prices up to today’s price - accounting for 1%+ of total shares, not insignificant. It’s hard for me to tell what the signal is here. If anything, these behaviours are not indicating the stock is highly undervalued. I have sold a small portion of my position to take profits, which will be moved back to cash and allocated towards private investments.

Specific updates now follow for paid subs:

• On the personal funds side, I describe how I’m managing my Berkshire position as Buffett retires at year-end, as well as some new positions I’ve picked up or increased.

• On the growth funds side, I opened a position in another founder-led public company, took profits on a longer term holding where the CEO is changing, and made one private investment in an early stage startup.

Read more