25 May 2025
News
OpenAI aquihires Jonny Ive for $6.5bn
OpenAI bought Jony Ive’s startup ‘io’ for $6.5bn in stock. Io has about 50 people and has been working on some kind of personal AI hardware device, planning to ship in 2027: apparently this is not smart glasses and Ive is sceptical of a new wearable. Meanwhile, Ive will ‘take responsibility’ for design. There’s a video (in which Ive also demonstrates that he’s invented a new way of walking). See this week’s column. ANNOUNCEMENT, COVERAGE
A viral video from earlier this month: Ive was interviewed by Patrick Collison at a Stripe conference, talking about the value of design in products. His observation that not everything that’s important in a product can be measured resonated with a lot of people (especially those concerned by Tim Cook’s Apple). LINK
The Google IO news blizzard
Google held its annual IO developer event, and as usual announced an overwhelming amount of stuff (much of it ‘coming soon’ or ‘in early trials’). Beyond the ‘feeds and speeds’ benchmark scores for new models (Google’s models are now frequently at the top of the leader boards), a couple of things to highlight:
‘AI mode’ in search, added to all sorts of surfaces including Chrome (Google thumbing its nose at the DoJ) and further shifting towards zero-click search. LINK
Lots of discussion of shopping and an AI-powered ‘virtual try-on’ for apparel. LINK
Google is taking on Meta (and probably Apple) in smart glasses, in partnership with Warby Parker, though the demo was a HUD rather than an AR display (i.e. it’s not trying to place things into the real world). LINK
Lots of new AI features for Android, increasing the pressure on Apple for WWDC next month (at least for tech insiders) after the fiasco of the postponed / canceled Siri relaunch that it announced last year. LINK
And a new SOTA video generation model. LINK
In general, though, two observations. First, incumbents always try to make the new thing a feature, and Google is doing a lot of that. And second, for now these models are clearly becoming commodities, where Google has caught up, and what matters most might now be distribution - for which, see the previous point.
The week in AI
As Xiaomi moves from phones to cars, it’s investing $7bn in chip design. LINK
LM Arena, an academic project to build and run crowd-sourced evaluations and benchmarks for LLMs, has raised $100m ‘seed’ funding at a $600m valuation to become a company. Evaluation is certainly a big problem, but this reminds me a bit of Clubhouse. LINK
Oracle has always been all hat and no cattle when it comes to AI data centres, but the FT reports that it’s now planning to spend $40bn on Nvidia chips for a 1.2GW datacenter in Texas as part of OpenAI’s ‘Project Stargate’. LINK
I missed this a few weeks ago: the WSJ says Baidu is thinking about expanding its robotaxi platform to Europe. LINK
News from Idiocracy
Donald Trump is threatening to apply 25% tariffs to all imports of iPhones (and Samsung phones), demanding that Apple shift manufacturing to the USA, after Apple reacted to his tariffs on imports from China by planning to supply the USA from India. None of the manufacturing ecosystem to build consumer electronics at volume exists in the USA, and it would take years to build, so Apple doesn’t have any way to comply in the short term, and (as for all of these kinds of stories) it’s very hard to plan long-term capital investment when the policy changes from week to week.
There are good geopolitical reasons to reduce dependance on China and to increase US capability, and Trump’s tactic is always to make dramatic demands that he then walks back from… but all this policy churn is incredibly wasteful. LINK
Fortnite and the App Store
Today in arguments from the past, Fortnite is finally back in the iOS app store (and went to the top of the games chart). LINK
Ideas
Demis Hassabis and Sergey Brin in conversation after Google IO. LINK
The Information says Perplexity had $34m of revenue last year, burnt $65m of cash, had $850m on hand in December… but had just 1.6m users and 260k paying users. Perplexity gets a lot of buzz in tech, but struggles to break the top 100 in the app store. When models are commodities (and Perplexity doesn’t even have its own) then everything is about distribution. LINK
A staggeringly long interview with Microsoft CTO Kevin Scott about ‘the agentic web’ - which really just means “hey, imagine if every website replaced its internal search engine with an LLM, and then that LLM could use MCP to give generative AI responses to queries from LLM search engines, and then the web would run on answers instead of links”. Well yes, maybe, but that’s like speculating about social media in 1995, and has huge incentive problems. (If it worked this would be a great blockchain use case). LINK
Bloomberg has a long profile of Anthropic’s Dario Amodei. No big revelations but lots of colour. LINK
Nvidia CO Jenson Huang’s latest 90-minute keynote, at Computex. It’s worth watching one of these even if you’re not deep into the semiconductor and datacenter worlds, just to get a sense of how broad the ambitions are and how the hardware ecosystem really works. He wants to make Nvidia into the new Sun Microsystems, not the new Intel. LINK
Your stolen iPhone probably ends up passing through one particular building in Shenzhen before being broken down for parts. Also interesting - there’s a good market for SIM-locked phones as WiFi-only media & games machines. LINK
How AI is changing dating apps (which are now over half of all new relationships in the USA). LINK
How Formula One moves data around. LINK
Henry Blodget’s new thing. LINK
Outside interests
Want. LINK
A birder spent a very long time trying to work out why ‘Charlie’s Angels’ used the ‘wrong’ bird. Nerdily interesting, and also fun to see all the slightly perplexed movie people trying to be helpful. LINK
Data
The MIT Tech Review tries to work out LLM energy use. TLDR: it’s hard, and fuzzy, but a simple ChatGPT query uses very little energy. LINK
The number of independent physical bookshops keeps growing. Worrying that Amazon would take all book sales is a bit like worrying that Walmart would take all grocery sales. LINK
Google showed a chart of ‘token generation’ showing massive growth. This is a bit like ‘hits’ in the early days of the web, or analysing YouTube by looking at bandwidth: it’s indicative but hard to interpret, since on one hand the use is growing and (agents need far more tokens) but on the other hand the models are also getting far more efficient. Still, the chart goes up. LINKk
BYD is now outselling Tesla in Europe. Elon Musk gave Bloomberg an interview in which he denied this (or just didn’t know the numbers): it’s probably a good thing he’s going to spend more time there. LINK
Column
Jony Ive and OpenAI
Sam Altman has OpenAI problems and AI problems.
He has to turn a science lab with weird governance and capital structures into a company, and manage an associated spoiler lawsuit from Elon Musk. He should probably move away from dependence on Microsoft for infrastructure, and so he flies around the world drumming up capital for ‘Project Stargate’ (last week’s good idea was trading this for oil money and getting Trump to drop the export restrictions on the Gulf). He has to worry about AGI, or at least manage the people who still do. He has to work out which politicians to lobby, how and in what direction. OpenAI still sets the agenda in AI research up to a point, but it’s no longer a clear leader and Google in particular is frequently ahead in the benchmarks. And given how much he has going on and given OpenAI’s talent retention issues, it was a good idea to hire Instacart’s CEO as COO.
Then there are the AI problems. What is this, and what will it become? Everyone in Silicon Valley is using chatbots all the time (and this is already transforming software development and marketing), but it’s not actually clear what the product is for everyone else. How can you make the chatbot into a product that more than a small fraction of people can think of ways to use? Does that need a device? A new paradigm? Better design? More talent?
And of course, frequently these questions merge. If the models themselves are commodities, does that mean distribution is the most important thing? The Microsoft partnership brings some of that (presuming it lasts), but Microsoft has no mobile, search or social. Meanwhile, incumbents always try to make the new thing a feature, and we saw a lot of that from Google this week: the elephants have all woken up now, even if Apple has stumbled badly. So how can OpenAI muscle its way through that and up into ‘platform company’ status?
That’s a lot of chess pieces and a lot of places to store up option value, and paying a few percentage points to get a ready-made rock-star hardware product team, led by the greatest product rock-star of tech, might be a good chess piece to add to the board.
Yes, but.
Apple needed Jony Ive, but Jony Ive needed Tim Cook and the chips team and the hardware engineering to turn the beautiful thing into reality at 200m units a year with 45% gross margins, and it needed the software people to light it up. (That’s not even thinking about Steve). Jony Ive made one part of the thing, not the thing itself, and though his new team has many people who did do great work in interaction, building outside the mothership can be very different (as many ex-Apple and ex-Google people have discovered).
And then, is the way to take generative AI from promise to universal reality getting all of us to buy a new device (that’s not glasses)? Or is it to make the software behave and present itself in fundamentally different ways? Is that a new device, or an app or a website on your phone (and whatever else happens, Apple will still have the best hardware), or reworking the phone software (which still means Apple or Google, despite all the recent antitrust unlocking)? Or is that not a real problem and LLMs will take over everything anyway?
I don’t know, and I doubt Sam Altman or Jony Ive know either. But if you were Sam Altman, why wouldn’t you buy the option, and if you were Jony Ive and you’d retired becasue you were bored of designing beautiful cases for glowing rectangles, what else could get you out of retirement if not trying to redefine computing again?
The trouble is, Sam Altman seems to be much better at getting brilliant people to work for him than at keeping them. A lot of people pointed out that the acquisition announcement looked a lot like a rather sweet wedding announcement. But all Sam Altman's business relationships seem to end the same way.