Good morning,

We cover the triple-IPO season that is about to define 2026's tech calendar — SpaceX in June, Anthropic in October, OpenAI in 2027 — and Mistral’s Mensch testimony in Paris that put into words Europe's "vassal state" anxiety. Plus the usual top news roundup including the new Anthropic x SpaceX relationship, a softening of the EU AI act and more.

But before that: this week’s edition comes with a new format — as simple as the other, but managed from outside Zurich. As some of you might know, I have left Zurich Insurance to pursue a PhD in AI. However, I’ve decided to keep the newsletter going in my own time, as it has been a great platform to share what I’m learning and, it seems, some of you appreciate it. Please feel free to share this newsletter with others (new readers can subscribe here). Or, if you would prefer not to receive it anymore, you can unsubscribe here. In any case: thank you for your support and your time over the last couple of months!

Three private companies that did not exist as serious commercial entities six years ago are about to become the three largest tech IPOs in history — within roughly twelve months of each other.

SpaceX leads in June. Reuters reports that SpaceX has accelerated its timeline to aim for Nasdaq entry on 12 June. Aiming to raise $75bn at $1.75 trillions it would be the biggest IPO to date, surpassing Saudi Aramco's $1.7tn 2019 debut. SpaceX is now bundling launch services, Starlink (4.4bn in profit in 2025, more than doubled YoY), and what used to be xAI (rebranded "SpaceXAI" since the merger, $6.4bn loss). The company plans for perhaps the most agressive IPO in history intending to guarantee Elon Musk cannot be fired and offering a trillion-dollar pay deal linked to building a colony on Mars.

Anthropic in October. Bloomberg confirmed on 12 May the company is closing a $30bn round at a +$900bn valuation, IPO targeted for October. Annualized revenue is now reported at ~$14bn (vs. $4bn in March), with Claude Code alone running at $2.5bn ARR. Anthropic has gone from "the safety shop" to the top 5 largest private company in the world in eighteen months.

OpenAI in 2027. The company seems to be targeting an H2 2026 filing and a 2027 listing at up to $1tn. ChatGPT crossed 900 million weekly users in March, with enterprise now ~40% of revenue. Cash burn is also climbing — from $17bn in 2026 to a projected $47bn in 2028 — which makes the IPO less of a choice and more of a necessity.

On top of those 3 giants, US IPO proceeds so far this year amount to $28.4 billion. The semiconductor and supercomputer developer Cerebras was the last notable entry on May 14.

This is happening on top of a market that has already been almost entirely driven by AI. The Magnificent Seven now make up 34.8% of the S&P 500 as of 12 May, with the top ten companies accounting for around 40% of the index - higher concentration than at the dot-com peak in 2000 (when the top ten held ~27%). In 2025, 42% of total S&P 500 returns came from the Mag7 alone. And in the March–May 2026 rally, it was estimated that ten stocks — Alphabet, NVIDIA, Amazon, Broadcom, Microsoft and five others — drove 69% of all index gains. The other 490, as the analysts put it, were "largely along for the ride." Add SpaceX, Anthropic and OpenAI to this and the index might need to be restructured.

On Tuesday 12 May, Mistral CEO Arthur Mensch appeared for ninety minutes before the French National Assembly's inquiry commission on "structural dependencies and systemic vulnerabilities in the digital sector." Given Mensch's position as Europe's top AI leader, the testimony was perhaps the clearest, most quotable public articulation we have heard yet of the European AI sovereignty case.

The headline framing was deliberately blunt. Europe has two years, Mensch said, to build its own AI infrastructure before becoming a US "vassal state.""Once supply is monopolised by American players, suddenly we no longer have supply, and we can no longer transform electrons into tokens."

A couple of specific points worth noting.

On Mythos. Mensch made the clearest European intervention yet on the cybersecurity model from Anthropic. He argued the French army's code bases should not be scanned by Mythos, as the dependency would be "nearly impossible to reverse." His broader point: Mythos-class capability is not exclusive to Anthropic, and Mistral's own models or Chinese open-weights could, in principle, find the same vulnerabilities — a point Anthropic’s CEO made too a couple of weeks back. The EU will get access to OpenAI's GPT-5.5-Cyber but it has yet to find its way to Mythos. Mistral is developing a dedicated model for European banks that cannot get Mythos access.

On Mistral's own positioning. Now valued at around $14bn, Mensch noted that US investors hold less than 30% of the company, as European capital was preferred but not available in sufficient amounts. The 1,000-employee company has committed to 1 GW of AI computing capacity by 2029, and is now working with many major European groups, including LVMH, ASML, AXA, and CMA CGM.

On AI spending and workforce impact. Mensch repeatedly explained that Mistral now consumes AI internally to the tune of 10% of its payroll — or about 10,000 euros per employee. He believes that most companies should eventually expect to pay similar rates to achieve optimal productivity gain from AI use. At the same time, he admitted that we should expect significant impact on the workforce, with a mix of layoffs and role changes.

On Europe. Mistral’s CEO pushed back hard on European regulatory complexity, framing Brussels rules as a sovereignty risk in their own right: rigid compliance regimes hand the field to US incumbents who can absorb the legal cost. It is the same argument Mensch and seventeen other European CEOs already made last year in an open letter calling for "agile guardrails". Not just a critic, Mistral is trying to contribute to the change and published last month a white paper, "European AI: a playbook to own it", laying out 22 concrete measures on how to make Europe a better AI playground — a sort of Draghi report for AI.

A couple of months back, I listened to a Lex Fridman podcast with Peter Steinberger — founder of OpenClaw — and was surprised to hear him say his hands were “too precious for writing now” and that he used “bespoke prompts” to build. In other words: Steinberger doesn’t use a keyboard to interact with Claude or ChatGPT; he just talks to them (to the point where he “lost [his] voice”). I was intrigued by the idea, started experimenting myself, and a couple of months in, my conclusion is that it’s definitely worth trying.

Two moments where I found it especially useful:

(1) On a walk, when I want to brainstorm ideas or take notes. I just open Claude and “talk to it,” as if I’m on the phone with a friend. (Of course, one can argue a walk is meant to be peaceful, not a brainstorming session with an AI agent.)

(2) For long writing work. Recently, I had to write a 20-page report. Normally, that would take hours of writing and refining. Instead, I created a meeting with myself (you can use Teams or others) and started transcribing (I use NotionAI). For about 40 minutes, I explained the key points of the report, what I wanted, and how I was thinking about it. I then fed the transcript to an LLM for analysis and… boom: a clean 20-page report following my guidelines. I still had to adjust some sentences, but overall I saved an incredible amount of time. (To be clear, it did not generated content but just organized my though)

No doubt there are many other useful use cases for speaking instead of writing. I highly recommend you give it a try and see what works for you.