× This article follows I Tried Kimi K3 Inside Claude Code . That test asked whether Kimi K3 could work inside a familiar coding setup. Here, I ask what changes when a Swiss provider serves a model of similar scale. This is K3-class inference. Banks already treat data sovereignty as an operating issue. In How DORA Made Sovereignty a Bank Problem , I argued that banks must plan for concentration,…
× Kimi K3 is Moonshot’s new 2.8-trillion-parameter mixture-of-experts model . Its parameters are values learned during training. Because K3 uses several specialised components, only part of the model handles each token. It also accepts images and up to one million tokens of context in a request. Together, those learned parameters form the model weights. Open weights are available for others…
× On May 21, 2026, Paul Krugman concluded a long argument that Europe was not really in economic decline. He made one exception. Europe, he wrote, “can’t be sure that it will always have access to new technologies developed and produced in the other superpowers.” He added that “the risk of being cut off from strategically important technologies, once minimal, is now very…
× I dropped the YouTube app as part of degoogling my life. I had used it for one screen: new videos from the 10 to 15 channels I follow. No comments, no recommendations, no Shorts. I rebuilt that screen and nothing else. Just the feed, nothing else The result is one chronological list. It merges videos from every channel I follow and sorts them newest first. I can add a channel by URL or @handle,…
× Last December, I rebalanced around five theses for 2026 . Five months later, moving money from US large companies into cheaper markets had mostly worked. Europe was the exception. As always, this is a review of my portfolio, not a recommendation. Portfolio performance so far Through May 29, the portfolio gained 5.2% in Swiss francs on a time-weighted basis. A time-weighted return removes the…
× I’ve heard variants of the same line from probably a dozen people in the last six months. A consultant I know has done more US-to-EU migrations this calendar year than in his entire career before. A founder on LinkedIn wrote the other day: Trump effectively has a kill-switch to our highly digitalised society, and the thought of that is frightening. None of those people would have used the…
AI-generated, one-shot from the blog title via Gemini (Nano Banana). × AI-generated, one-shot from the blog title via Gemini (Nano Banana). There’s a version of this story where I move everything to Europe in one go and write a triumphant post about digital sovereignty. The actual version is less clean. As of this week, the site you’re reading runs from a small Hetzner box in Nuremberg…
× In June 2024, Leopold Aschenbrenner published a 165-page essay called Situational Awareness . He committed himself to a chain of dated forecasts. Models would get more processing while producing answers, a method called test-time compute. Power, not chips, would constrain the US AI buildout. The Marcellus shale would power data centres despite climate pledges by the largest cloud companies. By…
× This article accompanies the full research report, Reconciling Enterprise AI Revenue: A Methodological Crosswalk and Vendor-Level Census, 2025 . The PDF contains the 68-vendor census, the six-tier framework, each sourced deduction, and the netting of structural double-counts. Three widely cited enterprise artificial intelligence (AI) revenue figures published in 2025 differ by 40x. A fourth, the…
× Anthropic published a method for translating Claude’s internal activations into readable English. It works well enough to ship. The press release skips three structural problems that the paper itself acknowledges. On the SWE-bench Verified coding benchmark, a new Anthropic interpretability method called natural language autoencoders shows the model registering “this is an…
× Anthropic was founded as the safety lab that would pull rivals upward. Five years later, it was one of the most aggressive frontier scalers, with a $380 billion valuation. Its founding thesis treats capability at that scale as the reason safeguards are necessary. A personal note first. This post is an outtake from a 14,000-word profile of Dario Amodei that I published on May 9, 2026. I…
I spent two months running a Polymarket order-book collector. An order book lists resting buy and sell orders by price. My collector runs on a small virtual machine (VM) and subscribes to Polymarket’s WebSocket feed. It writes one file in the columnar Parquet format per Coordinated Universal Time (UTC) hour. By 2026-04-15, the archive contained 1,262 hourly files and 30,287,264,368 events.…
× Andrej Karpathy is one of the few people who has both built modern AI and explained it for the rest of us. He co-founded OpenAI, ran computer vision at Tesla (where he got Autopilot working), and his courses on neural networks are some of the most-watched lectures on the internet. He also has a habit of naming the era we’re already in. “Vibe coding” was his. “Software…
I built a tennis broadcast pipeline this spring and ended up running F3ED, the NeurIPS 2024 shot detector, on a couple of ATP Challenger matches. F3ED is a good model. It also kept labeling clear aces as “unforced errors”, which is what this post is about. Code: github.com/philippdubach/tennis-vision . F3ED ( NeurIPS 2024 ) detects shots well. The catch is the outcome head, which has 4…
× This month, Revolut Research and NVIDIA published PRAGMA : an encoder-only transformer trained on 26 million user histories spanning 24 billion events and 207 billion tokens across 111 countries. To my knowledge it is the largest encoder backbone for consumer banking event data anyone has put on arXiv. Nine months earlier, Nubank had published nuFormer , a similar premise with the opposite…
× Investing at the Edge of Knowledge, Part 5 · Start with Part 1 “If in an unknowable world none of your bridges fall down, you are building them too strong.” Zeckhauser’s version of this line refers to investments, not bridges, but the structural point is the same. A philosophy of investing that expects some failures is also a philosophy that accepts some losses will be borne by…
× Investing at the Edge of Knowledge, Part 4 · Start with Part 1 “He who acts in N plays to make his mean log of wealth as big as it can be made will, with odds that go to one as N soars, beat me who acts to meet my own tastes for risk.” That’s Paul Samuelson, writing in one-syllable words. The title of his 1979 paper : “Why We Should Not Make Mean Log of Wealth Big Though…
× If I’d had my way, we would have left it in the lab for longer and done more things like AlphaFold, maybe cured cancer or something like that. That’s Demis Hassabis (I cannot recomend watching The Thinking Game enough and or read The Infinity Machine ), the CEO of Google DeepMind and a Nobel Prize winner, describing the future he didn’t get. He wanted a CERN for artificial…
× Investing at the Edge of Knowledge, Part 3 · Start with Part 1 “One of these days in your travels, a guy is going to show you a brand-new deck of cards on which the seal is not yet broken. Then this guy is going to offer to bet you that he can make the jack of spades jump out of this brand-new deck of cards and squirt cider in your ear. But, son, you do not accept this bet, because as sure…
× Novo Nordisk spent decades and $1.8 billion learning how to get a peptide past the gut. Eli Lilly looked at the same problem and decided to skip it entirely. Your gastrointestinal tract is a 30-foot disassembly line for proteins. Acid denatures them, pepsin cleaves them, trypsin finishes the job, and the mucus layer blocks whatever survives. Sean Geiger’s excellent history of oral peptides…
× Investing at the Edge of Knowledge, Part 2 · Start with Part 1 Ellsberg’s urn experiment is one of the cleanest results in decision theory. Daniel Ellsberg (1961) put two urns in front of subjects. Urn A: 50 red balls, 50 black. Urn B: 100 balls, red and black, ratio unknown. Pay $100 if you draw the right color. Most people chose Urn A, the known 50/50 bet. Fine so far. But here’s…
× Investing at the Edge of Knowledge, Part 1 David Ricardo made a fortune buying British government bonds four days before the Battle of Waterloo. He was not a military analyst. He had no basis to compute the odds of Napoleon’s defeat, or victory, or any of the ambiguous outcomes in between. But he understood something that most of his contemporaries did not: the nature of his own ignorance…
× The iPhone 17 runs a 3 billion parameter language model on-device at 30 tokens per second. Obviously, the average consumer has no idea what that sentence means, and Apple hasn’t figured out how to make them care. I believe that’s about to change. Apple now has complete access to Google’s Gemini model in its own data centers, with the ability to distill it into smaller models…
No AI-discovered drug has ever received FDA approval. That sentence should sit uncomfortably next to every headline about Alphabet’s drug discovery spinoff. On February 10, Isomorphic Labs , the Google DeepMind spinoff focused on computational drug design, released IsoDDE: its Drug Design Engine. This isn’t a model or an AlphaFold upgrade. IsoDDE is a unified in silico drug discovery…
I bet there is another new architecture to find that is gonna be as big of a gain as transformers were over LSTMs. Sam Altman, the CEO of the company most invested in the transformer is telling a room of students it isn’t the final form. So what comes after the transformer? He’s probably right that something will, and the evidence is no longer anecdotal. Several recent papers have…
On March 26, 2025, Sam Altman posted the following three sentences people love MCP and we are excited to add support across our products. MCP is Anthropic’s Model Context Protocol. OpenAI is Anthropic’s most direct competitor. Altman was endorsing a rival’s standard. That post may be the most significant event in enterprise AI infrastructure this year. When your main competitor…
Qwen 3.5-35B-A3B , a model released by Alibaba in February 2026, runs on a single consumer GPU with 24 gigabytes of VRAM. A secondhand RTX 4090, available for around $2,000, generates 60 to 100 tokens per second with it. On select benchmarks per Alibaba’s own evaluations, it matches or beats Claude Sonnet 4.5. The Qwen 3.5 Flash tier costs $0.10 per million input tokens through…
Alphabet’s free cash flow is projected to fall roughly 90% in 2026. Not because the business is in trouble. Because the company has committed to spending $83–93 billion more on capital expenditure than it did last year. That is what $660–690 billion in AI capex looks like up close. Amazon guided to $200 billion alone . Meta’s long-term debt more than doubled to $58.7 billion to help…
A study published in July 2025 gave AI coding tools their most credible test yet. Sixteen experienced open-source developers, 246 real tasks, randomized controlled design. The researchers expected to measure how much faster AI made them. What they found: developers using AI took 19% longer to complete tasks than those working without it. The developers themselves thought they were 20% faster. That…
On December 11, Jimmy Carr sat on the TRIGGERnometry podcast and delivered a riff that sounded like Peter Thiel’s stagnation thesis filtered through a comedian’s timing: Minus the screens from any room, we’re living in the 1970s. Nothing’s happened in physics since ‘72. String theory has not got us anywhere. But if you take the compute power of AI and point it at…
× I spent the last week reading 12 bank AI research reports from nine of the world’s largest financial institutions: Goldman Sachs, JPMorgan, Morgan Stanley (three separate reports), UBS, Barclays, Bank of America, HSBC, Citi, Deutsche Bank, and Santander. I wanted to understand how institutions that collectively manage trillions of dollars and employ thousands of analysts actually see this…
Lockheed started by building Amelia Earhart’s favorite plane . Then came a government loan guarantee in 1971 (the L-1011 TriStar nearly killed the company), a Cold War, decades of consolidation, and now a business that earns 92.5% of its revenue from government contracts , with the F-35 alone accounting for 26% of its $71 billion in annual sales. The process took about 50 years. AI labs…
Economics doesn’t take into account what’s best for society. The goal of economics in a capitalist system is to make the most amount of money for your shareholders. That’s Jon Stewart, during his February 4 conversation with a Nobel laureate about behavioral economics. Stewart hosted Richard Thaler on “The Weekly Show” for 92 minutes. Thaler, the Chicago Booth…
Novo Nordisk was Europe’s most valuable company 20 months ago. Today its market capitalization falls behind ASML, LVMH, Hermès, L’Oréal, SAP, Prosus, Siemens, Inditex, Deutsche Telekom, and Santander. The stock has lost roughly 75% since its June 2024 peak of $142.44, falling from a $640 billion market cap to under $160 billion . Shares dropped another 16% this morning after Novo…
A wallet that observers have suggested is operated by, or close to, someone with access to non-public Google information deposited $3 million into Polymarket on December 3, 2025, placed bets on 23 separate “Google Year in Search” outcomes, got 22 right , and walked away with $1.15 million in profit in under 24 hours, according to the reporting. One of those bets: that d4vd would be the…
A 30-second Super Bowl ad costs $8 million . That’s $267,000 per second, roughly the median U.S. home price for every tick of the clock. Super Bowl LX drew 124.9 million average viewers with a peak of 137.8 million , the highest peak audience in American television history. The NFL accounted for 84 of the top 100 most-watched U.S. telecasts in 2025. The Oscars, by comparison, managed 19.7…
In the latest issue of The AI Lab Newsletter , I featured a ByteDance Seedance 2.0 clip: two men playing tennis at what looked like an ATP tournament. Photorealistic. I probably wouldn’t be able to tell it wasn’t real footage if I didn’t know. A co-worker who played junior pro-am tennis watched the same clip and said: “That backhand doesn’t exist. Nobody plays it like…
On February 2, 2026, the European Payments Initiative (EPI) signed a Memorandum of Understanding with Alliance EuroPA. The alliance links Spain’s Bizum, Italy’s Bancomat, Portugal’s SIBS, and the Nordic Vipps MobilePay system. The agreement connects 130 million users across 13 countries through one interoperable payment network. Headlines described the agreement as Europe’s…
Tail hedging earns its keep through what it lets the rest of the portfolio do. In The Variance Tax , I wrote about the ½σ² formula. Compound returns are approximately equal to arithmetic returns minus half the variance, where σ² is return variance. The penalty is quadratic, so large drawdowns destroy wealth in ways that are hard to recover from. A portfolio that falls 50% needs 100% just to break…
The market is simultaneously pricing AI capex failure and AI destroying all software. Both cannot be true. × Anthropic released 11 open-source plugins for Claude Cowork on January 30. Apache-2.0 licensed, file-based, running in a macOS-only research preview. Within a week, the IGV software ETF had fallen 32% from its September peak to a 52-week low of $79.65, roughly $2 trillion in market cap had…
The defensible asset in enterprise AI is not the model. It’s the organizational world model. Every major compute era decomposes into specialized layers with different winners at each level. Cloud split into IaaS, PaaS, and SaaS. The modern data stack split into ingestion, warehousing, transformation, and BI. Each time, specialists beat the generalists because the layers have fundamentally…
Sensor Tower’s State of Mobile 2026 report confirms what had been building for years: the mobile app economy has permanently shifted. For the first decade of mobile, games made more money than everything else combined. Clash of Clans and Candy Crush built empires on freemium. King went public. Supercell sold for $10 billion. That changed in 2025. Apps Overtake Games in Mobile Revenue ×…
Let’s say your portfolio returned +60% in 2024, then fell 40% in 2025. That’s an annualized average return of +10%. Actual return after two years: minus 4% (i.e $100 * 1.6 * 0.6 = $96). That 14-point gap is what we call the variance tax aka variance drain or volatility drag and it’s one of the least intuitive forces in investing. Take any series of returns with arithmetic mean μ…
Anthropic just released Claude Opus 4.6, the latest frontier AI model in the Claude family. It’s a big upgrade over Opus 4.5 and probably the most agentic-focused LLM release from any lab this year. Key upgrades: better agentic AI coding capabilities (plans more carefully, sustains longer tasks, catches its own mistakes), a 1M token context window (a first for Opus-class models), and 128K…
I’ve been reading the January 2026 state of markets reports from Andreessen Horowitz and AQR , and their conclusions on the AI bubble question in 2026 are almost impossible to reconcile. The a16z view is straightforward: AI fundamentals are real, and current prices reflect that reality. Their evidence is compelling. The top 50 private AI companies now generate $40.6 billion in annual revenue…
Hyperscalers spent over $350 billion on AI infrastructure in 2025 alone, with projections exceeding $500 billion in 2026. The trillion-dollar question is not whether machines can reason, but whether anyone can afford to let them. Hybrid recommender systems sit at the center of this tension. Large Language Models promised to transform how Netflix suggests your next show or how Spotify curates your…
The pitch used to be simple: accept illiquidity, get rewarded. Lock up your capital for seven years, tolerate capital calls and J-curves, and in exchange you’d earn returns that public markets couldn’t touch. It was the defining bargain of institutional investing for two decades. AQR’s latest capital market assumptions make for uncomfortable reading if you’re an allocator…
The UK is the country with no bloc. At Davos, Britain refused to join Trump’s Board of Peace , citing commitment to international law and rejection of the “pay-to-play” model. France, Germany, Sweden, Norway made the same choice. The difference is that those countries have somewhere else to go. Britain doesn’t. The SAFE instrument , the EU’s €150 billion fund for…
At Davos 2026, Canadian Prime Minister Mark Carney delivered a speech that received something rare at these gatherings: a standing ovation. Carney told the assembled elites what they already knew but hadn’t said aloud: the world is not in a “transition” but a “rupture.” The speech drew on Václav Havel’s 1978 essay The Power of the Powerless , specifically the…
Sara Hooker’s paper arrived with impeccable timing. On the slow death of scaling dropped just as hyperscalers are committing another $500 billion to GPU infrastructure, bringing total industry deployment into the scaling thesis somewhere north of a trillion dollars. I’ve been tracking these capital flows for my own portfolio. Either Hooker is early to a generational insight or…