Lila Soto: Hey — rough week, good week? You seemed annoyed at something when you walked in.
Iris Holm: I've been staring at a three-page letter addressed to Congress since Thursday and I can't decide if it's brave or just — really well-timed corporate self-interest.
Lila Soto: The open-weights letter.
Iris Holm: 'Open Weights and American AI Leadership.' Yes. And the hook — the thing that keeps snagging me — is that Jensen Huang posted it on July 24th as his first X post. Ever. The CEO of the world's dominant AI chip company has never posted publicly on X, and he breaks that with a congressional lobbying letter.
Lila Soto: Mm, and it's not just Nvidia on this thing — Meta, Microsoft, Hugging Face, Andreessen Horowitz — the list hits 77 companies.
Iris Holm: Right. And look — here's the thread I want to pull. Every company on that list runs on open compute ecosystems. Open-weight models need hardware. Nvidia makes the hardware. The letter is framed as a defense of American innovation and open collaboration. But the actual mechanism it protects is Nvidia's margin.
Lila Soto: So the question is — is there a real argument inside the letter, or is the letter itself the product?
Iris Holm: That's exactly where I want to start.
Lila Soto: But that framing — Nvidia scheming in a vacuum — it kind of flattens what's actually happening. Because the closed-model labs aren't just staying quiet. They're running the opposite campaign, separately, with real money. Think of it like two kinds of stores. One sells shelves and storage — they want everyone buying goods, any goods, the more the better. The other sells exclusive branded products — they want rules that make knockoffs harder to get. Nvidia sells the shelves. OpenAI and Anthropic sell the exclusive products.
Iris Holm: That's a clean model. What's the number behind it?
Lila Soto: OpenAI and Anthropic combined spent $3.17 million on federal lobbying in Q2 2026. Up 23% from Q1. In one quarter.
Iris Holm: Wait — $3.17 million in a single quarter? That's not cautious institutional positioning. That's aggressive.
Lila Soto: And it doesn't match the story Dario Amodei tells publicly. He's said Anthropic never advocated for a ban on open models. But simultaneously — and I mean the same period — they're pushing mandatory safety testing and export controls. Those aren't neutral asks. Those are structural advantages for labs that already cleared the safety bar.
Iris Holm: So the absence from the letter isn't principled neutrality. It's parallel pressure from a different direction.
Lila Soto: Yeah — it's not that they're against openness exactly, it's more like... mm, they want openness to mean something specific. Openness with a compliance cost attached. Which only the biggest labs can absorb.
Iris Holm: The Silicon Valley split isn't values. It's business model.
Lila Soto: And the business model split explains something I kept tripping over — the letter didn't just grow from 25 to 77 signatories. The 77-name version lives on Nvidia's own servers. Nvidia hosted it.
Iris Holm: That's the tell.
Lila Soto: It's not co-signing anymore. It's — mm, that's organizing. Palantir, CrowdStrike, SpaceX, Cohere, Palo Alto Networks, GitHub, Cloudflare, Replit — all on infrastructure Nvidia controls.
Iris Holm: A Bankwatch analyst called it explicitly — self-serving lobbying document. And the timing isn't coincidental. Chinese open-weight models are gaining. Kimi specifically is cited in the letter as eating ground against American offerings. So the argument becomes: restrict open weights, and you hand Kimi the field even faster.
Lila Soto: Which is — I mean, that's a rhetorically airtight move. Because now you can't oppose the letter without looking like you're siding with the Chinese model winning.
Iris Holm: Exactly the trap. Picture a policy staffer at Commerce — it's not Tuesday morning, it's Thursday after lunch, she's behind on three other files — she opens this, sees 77 names, Nvidia, Meta, Microsoft, SpaceX. She reads 'Kimi is gaining.' She drafts permissive language because it looks like consensus. Meanwhile an Anthropic lobbyist is having a separate conversation with her boss about distillation crackdowns. By Friday she doesn't know she's been worked from both sides of the same hallway.
Lila Soto: And neither side is lying to her exactly.
Iris Holm: No. They're each telling a true thing. The Bankwatch framing — self-serving — doesn't mean the China argument is wrong. Kimi is real. The competitive pressure is real. But Nvidia's capex exposure is also real, and those things are doing the same work in the letter simultaneously. That's — frankly, that's the move.
Lila Soto: And the part that makes this worse — what it cost Nvidia to break that silence — we'll get into that in a minute.
Iris Holm: What it cost — that's the part that actually changes the read. Because the neutral infrastructure framing wasn't accidental. Nvidia spent years cultivating it. Sell to the closed labs, sell to the open projects, sell to the government, never publicly choose. That's not passivity. That's a deliberate margin strategy.
Lila Soto: And July 24th ends it. One post.
Iris Holm: Permanently. You can't un-sign a letter you hosted.
Lila Soto: So the calibrated take — not 'it's a cynical sales pitch,' because that's too clean — it's more like... Nvidia decided staying quiet had become the riskier move. If Congress restricts open-weight models, the market Nvidia sells into just got smaller. You can't stay neutral about your own addressable market shrinking.
Iris Holm: And if restrictions fail — if the letter works — then Huang just wrote the playbook. Infrastructure companies, chip makers, cloud providers — they now have a template for inserting themselves into AI governance without looking like a closed-model lobby.
Lila Soto: Which is — yeah, that's the part that doesn't resolve cleanly. Because both outcomes expose Nvidia. Either they spent credibility on a losing fight, or they won and every policymaker now knows exactly how to read their next move.
Iris Holm: The neutral infrastructure framing was always a fiction. The post just made it visible. That's the defensible claim.
Lila Soto: Not villainous, not principled. Just — a dominant market player who concluded that silence cost more than exposure. That's it. That's the whole thing.
Iris Holm: I walked in Thursday annoyed at a three-page letter. I'm leaving — I mean, the letter hasn't changed. But the first X post ever feels different now than it did at the start of this.
Lila Soto: Yeah. The real tell isn't what Jensen posted. It's that he posted at all. When the company selling the shelves decides it can't stay out of the aisle anymore — that's not a statement about open-source philosophy. That's the infrastructure layer telling you exactly how scared it is of the policy layer catching up.
Iris Holm: Extremely convenient that the policy layer, in this case, buys a lot of H100s. Frankly, that's the whole letter.
