26 April 2026

News

Tim Cook steps back

After a lot of careful prepositioning and messaging, Tim Cook announced that he will become exec chairman and hand over the CEO role to John Ternus, an Apple (almost) lifer who runs hardware. See this week’s column. LINK

Anthropic, Alphabet and Amazon 

On Monday Anthropic announced a ten-year, ‘$100bn’, 5GW infra deal with Amazon, including Amazon’s Trainium chips. Then on Friday, it announced another deal, this time with Google, which will invest $10bn now and up to $30bn more based on performance targets, expanding an investment & computer deal that Anthropic announced with Google and Broadcom earlier this month. Effectively, Anthropic is buying compute with equity, in the context that agentic models, and especially agentic coding, have increased demand for raw capacity by orders of magnitude and the company can’t keep up. GOOGLE, AMAZON

The Elon AI shuffle

Back in February, Elon Musk arranged for SpaceX to buy his AI lab xAI, which had launched a pretty good frontier model last year, but has since lost all of its founders, with Musk saying it needs to be rebuilt from scratch. Part of that turnaround was apparently exploring a merger with the French lab Mistral, and now there’s a deal to ‘partner’ with Cursor for $10bn, with an option to buy the company later in the year for $60bn. 

Cursor has also been struggling - it was the super-hot AI coding story this time last year, but Anthropic is now far ahead, and Cursor is ultimately dependent on OpenAI and Anthropic models (it announced its own model, which was really an open-source Chinese model), while agentic coding needs a huge amount of infrastructure that Cursor doesn’t have. The Information reported that Cursor had struggled to raise from VCs, not least because it had over 20% negative gross margins last quarter, though it turned positive more recently. 

Meanwhile, SpaceX itself (now including xAI) gets closer to IPO, apparently targeting a $75bn raise at a $1.75tr valuation, which is roughly the same as the latest private market prices for both Anthropic ($1tr) and OpenAI ($900bn) combined, where SpaceX itself valued xAI at $250bn in February, although all of these numbers are obviously very impressionistic. Publications that’ve seen the S1 say the Starlink internet business produced $4.4bn profits in 2025 and the company overall had $19bn of revenue (xAI is obviously pre-revenue and expensive). That’s a close-to 100x trailing revenue multiple, but what’s the future?  

Well… in the dotcom bubble, we had a term for companies that said ‘we’re this now, but eventually we’ll be something else and you should value us on that” - the term was ‘multi-stage rockets’, which is a good comparison for an actual rocket company, but that’s Elon Musk’s contribution to capital markets. “Don’t value Tesla on what it is, value it on things that don’t exist yet” - ten years of autonomy promises, plus solar roofs, batteries, insurance, and sports cars, none of which happened, and now robots. On most people’s SOTP valuations, the actual cars are at most 10% of Tesla’s market cap. The same here: SpaceX has a very cool rocket business with a very unclear TAM, but look! Orbital data centres! Mining on the moon! The problem, as I said a while ago, is that Elon Musk is a bullshitter who delivers… something: he says all this stuff, but there are the car company and the self-landing rockets! He raised money promising the moon, and delivers LEO, and that’s still pretty good. If only he didn’t spend half his time on Twitter promoting white supremacy and Holocaust deniers. CURSOR, BACKGROUND, MISTRAL, SPACEX

Intel spikes

Continuing the hardware surge and capacity shortage story, Intel’s stock went up 24% on Friday after it beat estimates ($13.6bn revenue versus $12.4bn consensus) and raised guidance for next quarter as companies building datacentres snap up pretty much any chips they can get their hands on. That said, the foundry business remains unproven (for more… ask a semis analyst). LINK

The week in AI

As rumoured a few weeks ago, Meta will cut 10% of staff - 8k people - in an efficiency drive to fund AI. As I wrote when this was first suggested, Meta ended 2019 with 45k employees, hired 27k across 2021 and 2022, did an 11k layoff in 2023 but still ended the year up by net 15k, cut another 20k in 2023, and added 12k back by the end of 2025 bringing it to 79k. Cutting 8k now would mean that since 2019 Meta has hired 54k people and laid off 39k, just going by the headline numbers. LINK

Anthropic's security around its 'superhuman hacking' model Mythos turned out to have holes that let a bunch of people on Reddit play with it. LINK

Meta will record employee mouse and keyboard use as AI training data. LINK

Microsoft finally managed to ship a version of Copilot that can interact with your documents in Office. Claude launched this in January after a preview last October, and ChatGPT launched the same in March. Yes, that’s only a few months behind, but isn’t this Microsoft’s entire thing? LINK

Anthropic has started requiring some users to provide a photo ID, which means a lot of Chinese tech people who’ve been quietly using Claude over a VPN are being cut off. LINK

In other news

(Note - every now and then someone complains that there’s too much about AI in this newsletter, which is like someone in 1998 complaining that a tech newsletter talks about the internet too much. Still, there is some other stuff…)

If you can make an EV, then they’re a commodity (and Tesla has no differentiation). But EVs are integrated products that are very difficult for the disaggregated model of the legacy auto OEMs. Detroit went in hard but couldn’t make it work and pulled back - now Ford is discussing a partnership with Geely, one of the Chinese leaders, about a tech partnership. LINK

The EU has a privacy law waiver that lets tech companies scan user content for CSAM, but that waiver expired this week amid legislative logjam. They will continue scanning anyway. LINK

The French weather agency thinks someone tampered with a sensor to win a bet on Polymarket. LINK

Tech media

A few weeks ago OpenAI bought tech talkshow TBPN (me neither) for (apparently) $100m, and now A16Z has invested in MTS, another entrant in the radically under-served market for tech people talking to each other. 

I have some sympathy for the idea that the media is too often default-negative on tech, and clearly AI has a major comms challenge, with poll after poll showing very clearly that people are really sceptical about whether this is a good thing, which isn’t helped by the CEOs of OpenAI and Anthropic spending so much time saying that this is dangerous and scary. But if you want to persuade people otherwise, and want to achieve that by building new media voices, those voices can't just be talking to you. Both TBPN and now MTS are based on Twitter, when the only people still there are the people who already agree with you, and the first interview on MTS was a MAGA partner and culture warrior at Sequoia. Good luck with that. LINK

Ideas

Flipbook, the latest viral AI concept: a dynamic UX that’s generated live and streams from the model. LINK

The GPS modules in at least some cars store a complete location record, unencrypted. LINK

Dylan Patel’s SemiAnalysis has 85 people, had $20m revenue last year, and expects $100m this year (mostly from selling proprietary research). LINK

While Dwarkesh Patel (no relation) remains indie. LINK

China is testing a deep-water submarine fibre cutter. Because, of course, it is. LINK

Mozilla used Anthropic Mythos to find hundreds of bugs. LINK

Outside interests

How to read your airline confirmation code. LINK

A few weeks ago I linked to a wonderful room-sized suite of mirrors made by Claude Lalanne for Yves Saint Laurent and Pierre Bergé, on view at Sotheby’s New York as part of a beautiful collection. In the estate sale back in 2009, these were estimated at €700k-1m and went for €1.9m; this week, they sold again for $33.5m.LINK

Data

A comprehensive IEA report on the energy consumption of data centres (a field especially prone right now to people making things up). LINK

Meta paid Broadcom $2.3bn for chip design services in 2025 (reported under related party transactions). LINK

Adobe on AI traffic to retailers. LINK

Adobe on consumer use of AI. LINK

What’s the market size for in-flight Wi-Fi? LINK

Ramp reckons that 74% of AI SaaS spend is based on token consumption, versus the standard seat-based model. LINK

The 2025 IAB report on US online advertising. LINK

Column

Tim Cook, Maps and AI

Tim Cook stepping down is the obvious subject for a column. Indeed, it’s so obvious, and the things that people say are so obvious, that you could probably get ChatGPT to write it for you. The ops guy who took over from Steve and made everything work, but maybe didn’t make anything new and exciting. Mention outsourcing the supply chain to China, Apple Silicon, drop in a few missteps (Maps, the Car, AI), perhaps some product grumbles (the iPad has three different windowing systems, Liquid Glass is embarrassingly bad). Turn to John Ternus (Who? Did I spell that right? But everyone seems to like him, and now there’ll be a product guy running things?). Maybe Craig Federighi will leave? Oh, yes, and the stock price is up A LOT since he took over. Great! Hand that to ChatGPT to pad out into two thousand words….

Can we say anything else, though? The striking (and characteristic) thing is now little drama there is in this move, and also, how much Apple sits outside all of the hot debates in tech, which (as noted above) are now almost entirely about AI. Who cares? 

Well, there are two things that seem worth digging into that are perhaps a little less obvious. The first, I think, is to go back to the slightly dismissive way that people talk about Cook as sorting out the operations - useful, yes, and very competent, yes, but not really ’founder-mode’ material? That’s a blind spot. For over 20 years Apple has shipped new products, at massive scale, with metronomic efficiency. Every year a new iPod and then iPhone, and every year a new set of operating systems, and no bricked phones and hasty fixes, no recalls, no exploding phones and only two slips I can remember that are even worth mentioning - Maps, at the beginning, and AI, at the end, of which more in a moment. This really hard. This is a huge achievement. 

You can dismiss this further, and people do, by saying that Cook isn’t a ‘product guy’ and in particular he isn’t Steve Jobs, but then, neither is anyone else. And Apple hasn’t done a product that changed the world as much as the iPhone, but then, as Joseph Heller said, neither has anyone else. Meanwhile, Tim Cook might not be a ‘product guy’ but the Apple products created while he’s been CEO dominate every category and price point they choose to compete in. We know that the choices Cook took led here, and they look like the obvious choices because they always do, but there were plenty of other choices that looked just as good and led to disaster, and he didn’t take any of them. 

The other topic is AI. There’s a narrative now that Apple are somehow geniuses for ‘sitting out’ the AI boom, with other big tech companies spending $6-700bn on AI infrastructure this year alone, while Apple waits to pick a winner and integrate it (hence the Gemini deal). This is absurd. Three years ago Apple spent an hour of WWDC laying out a comprehensive (and very coherent) vision for integrating generative AI into its products - and since then none of that has shipped, the head of the AI group was ‘retired’ and half of the team quit. That is not a plan. 

On the other hand, I think people who spend all day with their head down in AI can miss how early this still is and how little is settled. We really don’t know what the right interaction models and products for LLMs will be, and nothing has been missed yet. Even then, the most likely bear case for Apple is that AI shifts the level of abstraction and Apple software platforms are stranded something like the way the web left Microsoft stranded - but people bought billions of Windows PCs to use the web, and you’ll still need to buy a box to put in your pocket to use the new AI things, and who makes the best box with the best battery and edge compute? The opposite case, meanwhile, is that chatbots are more technology than product and most people struggle to make them a daily tool, which is exactly the kind of problem that Apple used to be good at solving. Either way, it’s too early to have missed AI yet, just as 1997 was too early to have ‘missed’ the web. And either way, the real test of a CEO is how you manage those kinds of problems.