Three guests, three different bets on where the money actually sits in the AI stack — and for once, nobody's hiding behind the word "superintelligence." If you're just joining, here's the sovereignty story so far: in a post-AGI world, access to frontier AI becomes part of national power, not just industrial policy. Anton Leicht floated allied democracies trying a roughly $500 billion frontier-model moonshot, while warning that politics, financing, and U.S. pressure make that coalition really hard to assemble. This is Tech Podcast Podcast. Today: a fallen Intel CEO on the post-mortem tour, Mistral betting against the whole race, and a PE-backed startup swearing a few engineers can replace a consulting army. Let's see who brought receipts. You've got the details on this one. Two segments in one here: Pat Gelsinger doing the what-went-wrong tour, and Lovable's Anton Osika claiming a million new apps a week. One of those gives me a number I can push on. The Gelsinger bit I actually want is at 15:19 — why a Taiwan blockade cripples the U.S. economy. That's a former Intel CEO with real supply-chain knowledge, instead of the retrospective self-mythologizing. Right, but did Jason pin him down on a number, or did Gelsinger get to deliver the keynote version of his own failure? Taiwan making the most advanced chips isn't news — what's Intel's actual counterfactual timeline to fab parity? And the Lovable million-apps-a-week figure — that's measuring activity more than economics. The interesting section is 33:38, on bringing builder costs down. Give me a cost-per-app and then it means something. A million apps a week — how many survive Tuesday? That's the vibe-coding ceiling question all over again. Osika's making the same "a few people replace a whole team" promise, just with a bigger denominator. The Economist writes:
Most of the world’s biggest AI firms are locked in a race to build “superintelligence”. But Mistral AI, a French frontier lab, is betting against this “winner-takes-all” competition. The firm thinks that smaller, more customisable, open-weight models will prove more valuable. Mistral’s boss outlines his plan—and why he thinks that, wherever you are, AI should be built and served closer to home. How does his vision stack up?
Arthur Mensch is doing something none of the frontier-lab bosses have done on-mic this week — he's making the opposite of the superintelligence race his actual business plan. Smaller, customizable, open-weight, served closer to home. Which sounds great as a values statement. I want to know whether the Economist made him defend it as economics — open weights are cheaper to run at scale, fine, but show me a customer or a margin, not a manifesto. For me, that's the test. Every All-In guest declaring AGI is here is burning cash to beat him on raw capability. If Mensch can't put a revenue or customer number behind "smaller is more valuable," it's just positioning. And remember where we started this week — China out-downloading the U.S. on Hugging Face, open weights quietly eating the incumbents. Now there's a European lab saying it out loud as strategy. The defection story finally has a French accent. This also ties back to the sovereign-AI story we opened on — middle powers and that half-a-trillion compute figure. Mistral's the flagship saying the answer is local, open models built closer to home instead of one giant closed brain. Right, but "built closer to home" is a slogan until you know who pays for the racks. Sovereignty's expensive. If he waved past that, it's a nice keynote and not much else. Rebecca Bellan, writing in TechCrunch:
Can a handful of engineers really do the work of an army of consultants? That’s the bet behind Ode with Anthropic — the joint venture dedicated to embedding forward-deployed engineers in enterprise firms, backed by Anthropic, Blackstone, Hellman & Friedman, Goldman Sachs and others.
So Ode's whole pitch is a handful of engineers doing the work of an army of consultants. That's basically the ceiling claim we just heard from the Lovable guy on All-In — except now Blackstone and Goldman are writing checks. And it's PE-backed rather than VC-backed: Blackstone, Hellman and Friedman, Goldman, plus Anthropic. That's a different capital structure from anything else in the AI-services conversation this week. Right, and the piece leans on "hundreds of millions of dollars in business value." That sounds like a slide. I want the number: how many FDEs replaced how many consultants, at what billing rate? They do name one concrete thing: most enterprise AI pilots die before production, and their whole thesis is closing that gap with embedded engineers. If they'd said cost-per-engagement, that would be the proof the services thesis has been missing all week. And they're hiring generalists and ex-founders, rather than AI researchers. That pushes the moat down to the Anthropic model underneath. A lot of the defensibility may belong to Anthropic more than Ode. From SaaStr:
He went back through his portfolio. About 10% of pre-2022 companies were DOA the moment ChatGPT launched, solving problems that foundation models made trivially easy overnight. Another 30% are genuinely threatened and need to move fast or die. The rest are either insulated, made stronger by AI, or sitting on opportunities that didn't exist before.
Rory O'Driscoll went back through his own portfolio and put numbers on the doom. Ten percent DOA the second ChatGPT launched, another thirty percent on the clock. That's the operating detail I've been begging for all week. And it's a real counter to the "software is finished" mood we've had since Monday. His title says software got harder, not dead — I want to hear which layer got harder. Right — because the ten percent dead is clean. The scary part is the thirty percent that's alive but has to move or die, and he's saying most founders can't tell which bucket they're in. What I want from the episode is whether that thirty percent maps to a stack layer — is it the thin application wrapper that's exposed, or something deeper? If he says the rails are commoditizing too, that complicates the whole "own the infra" bet. A guy who's been investing in software since before most of these founders were born, actually auditing his book instead of vibing about AGI. Refreshing, honestly. If Tech Podcast Podcast helps you keep up, take a second to subscribe and leave a review wherever you're listening. It really helps other people find the show, and it means a lot to us.
We put links to every story we covered in the show notes, so if something stuck with you, you can dig into the source there.
That's Tech Podcast Podcast for Thursday, July 16th. This is a Lantern Podcast.