top of page
BB White and Orange.png

SNI: WEEK 22

  • 3 days ago
  • 6 min read
image of sun with text sector udpates: wk 22

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, advanced manufacturing and insurance. Where the money rushed in, what it's building and who's being watched.


tl;dr: What's in short supply? It's not the money.


Anthropic raised $65 billion at a $965 billion valuation, making it the world's most valuable AI startup. But it's not the only AI-thing capital has been a-courting.


China's troubled consumer funds turned tail to become tech stock-pickers, while a US chipmaker fund hit the quickest $10 billion on record. DigitalBridge bought a private equity energy firm - to feed the power to fuel the models. And Kirkland & Ellis, a law firm, committed $500 million to its own AI.


Capital also chased physical asset-makers. Dell's shares soared 40% - server demand is now reaching well beyond Nvidia. Nokia, meanwhile, rallied 140% – turning a left-for-dead name into a take-me-now offer. Which made Architect's move to turn compute into a futures market – financialising GPUs the same way as stocks, power and freight – slightly less surprising.


And while we counted mixed economic signals in the productivity data, Glean disclosed $300 million in recurring revenue by promising to cut customers' AI bills. Some might say the boom is maturing.


But – there is a but. All that capital is pointing one way. And that manufactures fragility. The ECB warned private-credit is now a threat to financial stability. And that was before we wondered if the money has priced in the Iran war's helium shortage.


The Pope, for his part, spent the week appealing to the industry's conscience. The FT, meanwhile, found room to argue that AI can be art.


But if not art, what is all that capital actually building? Anthropic shipped Opus 4.8 and confirmed a wide release of Mythos-class models within weeks.


Mythos has found vulnerabilities in every major operating system and browser – the exact capability a surveillance state runs on. Which means the same money is also helping build surveillance mechanics for many nation states.


And right on cue, Washington approved $9 billion for its spy agencies to catch up on AI. Beijing overhauled the world's biggest surveillance network. And in Brazil, US investors backed Pax – an AI system for police. Will this ultimately lead to a Minority Report situation outside of China?


Europe appears to be the one liberal beacon currently pushing back. Brussels is pressing for tech sovereignty to cut its reliance on the US, while Mistral goes vertical – designing its own chips, building data centres, reaching across the whole value chain. A countervailing bet against both US dominance and the Chinese model.


Whether any of that works is the week's real question. The Pope's moral appeal and Europe's regulatory stance are both pushing against industry momentum. Neither has bent it. Yet.


And on that note, here's everything else worth reading this week:


Biopharma:


The race for biology's foundation model accelerated, the cheque-writing carried on, and one trial failure kept the news honest.



Medtech:


Irish medtech led the week's AI news – two home-market wins ahead of a run of clinical, regulatory and device firsts.


Advanced manufacturing:


Even more physical space: AI in product formulation, the factory floor and the silicon underneath.



Insurance:


Another busy week as AI becomes business-as-usual.



But what set podcast tongues a-wagging?


In a word: Labour.


Coding hasn't fallen yet. But it's getting close.


Dan Shipper on Lenny's Podcast hit on the paradox of the week: the more we automate, the more humans we need – and the more work there is to do.


But whilst various commentators wondered how, exactly, that works – you don't have to. Hop over to YouTube and watch Chad Jones explain why radiologists are more numerous and better paid today than when Geoffrey Hinton told us to stop training them in 2016.


Jones' argument hinges on the idea of weak links – you still need humans to perform tasks, and that makes those tasks the 'weak links'. The slowest part, the limiting factor that stops rapid productivity gains. But finally remove all the weak links and full automation does make for incredible productivity.


No one seems to doubt that coding is closest to being weak-link free. Just this week: ING announced a working trading system built via vibe coding, Starbucks began weighing AI use into its tech-workers' bonuses - and finance is head-over-heals with Claude, at a price.


None of which means coding has fallen. But many think we can see the floor from here.


Economic signals are mixed. Jones can explain that one too.


Azeem Azhar's Exponential View asked the question some CFOs are also wondering: why isn't AI showing up on the bottom line? On a16z, a related puzzle: why isn't AI actually killing SaaS yet?


Lots of expert answers. Including:

  1. The productivity isn't in the numbers;

  2. The disruption is slower than the headlines; and

  3. The experts' own models disagree on which jobs go first.


Jones' take: infinite amounts of a cheap thing only lift GDP by that thing's share – software is about 2% of the economy, so automating all of it buys you roughly 2% growth. The signals look mixed because most of us haven't absorbed this idea yet.


Although some tend to disagree.


Moonshots' resident accelerationists argue AI adoption will be more rapid than Jones expects. Agentic AI, Salim Ismail claims, has killed the modern company: for ninety years firms grew because, as Ronald Coase showed, coordination was cheaper inside them than out. But that logic may now have inverted: 'building the feature is cheaper than having the meeting about the feature'.


So you rebuild the firm AI-native, around intelligence rather than hierarchy, run it at a fifth of the headcount, and let it improve itself. This is recursive self-improvement at the workflow level – the invoice process that asks how to do better on every loop. Not in decades. In a month or two.


It's the week's most provocative listen. Where it lands is unexpected.


NLW said there was no slowdown. He was right by Friday.


On Tuesday's AI Daily Brief, Nathaniel Whittemore spent the episode dismantling the annual AI-slowdown panic – the ritual, every spring, of declaring the boom is over. By Friday he'd been proved right in the most emphatic way: Anthropic's news ripped the rug right from under the argument. Perhaps this summer won't be spent debating a bubble?


Redistribution is moving from philosophy to policy.


Not that Whittemore thinks it's all rainbows and lollipops. On Thursday's AI Daily Brief, he walked through a redistribution debate that has moved, in months, from seminar rooms into real proposals: Elizabeth Warren's data-centre energy excise, Mallory McMorrow's token tax feeding a workforce reinvestment fund, Mark Cuban's sub-50-cents-a-million-tokens floor, Brookings' staged plan, and Anthropic's Dario Amodei floating a 3% AI revenue tax. The argument that Americans deserve more from the AI economy even arrived from an unlikely direction.


Underneath all of it sits Jones's hard number: the labour share of the economy has fallen about 10% in the last 25 years, while capital's share has risen. Which is the empirical spine of the whole worry. And why it dominated the many, many debates this week.


You could be forgiven for needing cheering up after all that. If so, this episode of Odd Lots is very much worth a listen.


Thanks for reading. Join us again next week for all the AI news you need to know.


 
 
BB White and Orange.png
Get in touch bubble roll.png
Get in touch bubble.png
Button overlay.jpg

Home

​

​

Further reading

​

​

Careers

​

​

Contact us

BB White and Orange.png
bottom of page