AI Sovereignty Wars, Palantir-Nvidia Deal, SCOTUS Birthright Ruling, Newsom's CA Budget Lie
Chamath Palihapitiya, David Sacks, David Friedberg, Jason Calacanis
4 Jul 20266 min read1h 45m
TL;DR
The All-In hosts argue that enterprises sharing proprietary data with frontier AI labs like Anthropic are mortgaging their future — citing Anthropic's launch of Claude Design, which blindsided Figma after Anthropic's CPO sat on Figma's board until 3 days before launch. Chamath's company 8090 tested open-source models wrapped in their harness and found them 16.4x cheaper than Anthropic Opus 4, while Sacks warns that Anthropic is following a Microsoft/Google playbook of using model dominance to capture lucrative verticals. The Palantir-Nvidia sovereign AI partnership is framed as a direct counter-move: letting governments and enterprises own their hardware, data, and model weights outright.
Key Moments
David Sacks
“according to the information, Anthropic quote unquote blindsided its then business partner with the launch of Clawude Design. So this was a new vertical app that Anthropic launched to compete in the design category. And Figma's founder said that Anthropic had not been completely honest with them. Enthropic chief product officer had actually even served on Figma's board and didn't resign until 3 days before the launch of Claude Design.”
Sacks using the Figma-Anthropic situation as a concrete example of why enterprises cannot trust frontier AI labs with their data
“when you use our harness with Claude, it was simultaneously 1.4x cheaper and 1.5x faster than just using anthropic opus 48 alone. But if you wrap the open source model with our software factory, it was 16.4x cheaper. Now it was three times slower. But you know, you're talking about a couple of extra hours to save 16.4x”
Chamath sharing real benchmark data from 8090's experiment comparing Claude alone vs their harness with open-source models on an enterprise code migration task
“there's been an effort by anthropic to go around and sign up life sciences companies to contributing to a new life sciences focused model that effort has been they're approaching these large companies with large proprietary data sets and saying hey if you share your data we will give you early access some sort of proprietary value sign this NDA and you can participate with us and I think nearly everyone I've spoken with has woken up to the fact that they are basically trying to commoditize everyone's business”
Friedberg describing Anthropic's active outreach to life sciences companies to contribute proprietary experimental data to a new specialized model
“Nobody who went to bed with Microsoft in the 80s, Facebook in the 2000s, or Sam Alman now in the 2020s did not wake up with their throats slit. This is a message to founders. If you partner with any of these people, they will slit your throat and take your business wholesale. There is nothing to discuss here. Don't trust them. Use your own models.”
Calacanis warning founders after discussing Sam Altman's offer of $2M in free tokens to Y Combinator startups, framing it as intelligence harvesting
“if you're an application, you don't want to be beholden to one model provider, right? You want to have a choice. And if you're an enterprise, you want to have a choice because you don't want have to give up all of your proprietary knowledge. And if you're a chip company, you don't want a monopsiny buyer situation where there's only one or two companies who can buy your chips and by the way, they're producing their own.”
Sacks explaining why the Palantir-Nvidia partnership makes strategic sense — both benefit from a competitive model layer rather than an Anthropic-OpenAI duopoly
All-In is a weekly panel podcast hosted by four close friends and tech investors: Jason Calacanis, Chamath Palihapitiya, David Sacks, and David Friedberg. The show covers technology, business, politics, and markets with an insider Silicon Valley perspective. Sacks served as AI Czar for the first half of the Trump administration, giving the panel direct access to policy discussions shaping the AI industry.
Takeaways
1
Anthropic's vertical integration is a direct partner betrayal Anthropic launched Claude Design while its CPO still sat on Figma's board, resigning only 3 days before launch. It has since launched Claude Code, Claude Science, Claude Legal, Claude Security, and Claude Financial — every one competing directly with companies built on its own models. The pattern is deliberate: watch where value accrues on your platform, then vertically integrate.
2
Open-source with a harness beats frontier models on cost 8090's live experiment showed that wrapping an open-source model in their software factory delivered results 16.4x cheaper than Anthropic Opus 4 alone on a real enterprise code migration task, at the cost of 3x slower speed. For non-latency-sensitive enterprise workflows, this trade-off is often worth it, and the gap will widen as open-source models improve.
3
AI safety for enterprises means owning the model weights Sacks reframes 'AI safety' away from alignment discourse: for enterprises, safety means controlling their own compute, model weights, and data so a frontier lab cannot absorb their proprietary knowledge and compete against them. Anthropic's safety rhetoric, which argues against open-source access, conveniently protects its own business model.
4
Shared intelligence infrastructure is strategically self-defeating Chimath's core point: you cannot rent intelligence from the same provider as your competitor — it collapses differentiation to the lowest common denominator. With half of large US companies already unable to generate returns exceeding the 8-11% cost of capital (per BCG), paying premium API costs to commoditize your own proprietary knowledge is doubly destructive.
5
Palantir-Nvidia sovereign AI partnership challenges the duopoly The partnership lets US government agencies own their hardware, data, and model weights outright using Nvidia's Neotron open models — a direct alternative to routing sensitive data through Anthropic or OpenAI. Sacks frames this as rational self-interest: Nvidia needs a diverse buyer ecosystem, Palantir needs a competitive model layer, and enterprises need a data safe option.
6
Nvidia's open-source model ambitions were deliberately hidden from customers Calacanis argues Nvidia suppressed public discussion of Neotron to avoid alarming its biggest customers — OpenAI, Anthropic — who had made substantial progress against it. Now that those labs are building their own chips (OpenAI's Jalapeno, Anthropic's silicon, AMD partnerships), Nvidia has lifted the veil and is openly competing at the full stack: chips plus open-source models.
7
Enterprises will shift to a 70/20/10 hybrid inference model Friedberg predicts enterprises move from fully outsourced cloud inference toward roughly 70% big cloud, 20% local on-prem, 10% experimental — driven by cost, latency, data sovereignty, and the falling price of running open-source models on owned hardware. Token prices are expected to drop 90% per year for the next three years.