Two open-source stories collide: NVIDIA quietly bets on Ilya Sutskever's secretive Safe Superintelligence, while $8 trillion of tech market cap signs an open letter defending open-weight AI — leaving Anthropic standing alone and getting either villain or hero framing depending on where you sit.
Eight trillion dollars of Big Tech just made open-weight AI a coalition — and Anthropic the lone holdout.
As Washington edges toward banning Chinese open models, NVIDIA, Google, Microsoft, Meta, Elon Musk, Palantir and eventually OpenAI co-signed an open letter defending open weights. Read cynically it's everyone talking their book against the two labs running away with the market; read charitably it's a fight against regulatory capture. Either way, Anthropic's refusal to sign turned it into a Rorschach test — villain to some, principled and 'aura-farming' to others — and the messy employee snark on both sides shows just how heated the open-versus-closed war has become.
NVIDIA made a substantial (undisclosed) investment in Ilya Sutskever's secretive SSI and granted access to its next-gen Vera Rubin chips, after gaining 'rare access' to SSI's technology. Sources say SSI had been relying on Google's TPUs.
— Safe Superintelligence, on the NVIDIA partnership. SSI framed the NVIDIA deal as a signal of progress, saying the partnership will let it scale, and expects to 10x its compute over the next 12 months.
— Daniel Levy, SSI co-founder. SSI co-founder Daniel Levy's line captured the company's straight-shot-to-superintelligence ethos — treating short-term commercial work as a distraction from the ultimate mission.
NVIDIA fell 4% Monday as roughly $750B in new deals struck some as too circular — including a $500B+ partnership with SK Group and a $250B deal backstopping an OpenAI data center. Expect the circular-deal chart from last year to make a comeback.
After a Huawei demo showed dramatic progress, Beijing pushed hard toward domestic chips, with the AI czar reportedly warning big AI users that resisting was traitorous. China's US-chip dependence fell from 90% in 2021 to 60% in 2025, and Morgan Stanley sees 25% within five years.
— Eric Xu, Huawei Deputy Chairman. Huawei Deputy Chairman Eric Xu on why China built its own chip economy. Analysts still expect China to lag the US in the chip race through at least 2030, but Huawei alone plans to ship ~1.5M AI chips this year.
Denied ASML's cutting-edge fabs, China's supply chain makes ~28,000 advanced logic chips a month — a fraction of TSMC — but Bernstein sees that doubling yearly for three years. The Information reported a Shanghai firm has begun mass-producing DUV lithography machines, tech previously kept from China.
Apple is lobbying the White House to buy Chinese memory chips from makers like CXMT, warning of $5,000 iPhones as prices spike. Micron — the only sizable US memory maker — is lobbying to keep restrictions, warning Chinese entrants could decimate the domestic industry like steel.
NLW frames the open-weight fight as completely dominating the insider AI conversation — Jensen Huang's open letter drew 62 million views on his first-ever X post, a signal of just how big this has become.
Concern built through GLM and then Kimi K3 — at 2.8 trillion parameters, the largest open-source model ever, beating Fable on a front-end code arena. That was enough for non-technical policymakers to move from quietly concerned to a full-on freak-out and float banning Chinese open models.
— Treasury Secretary Scott Bessent. Treasury Secretary Scott Bessent said the administration supports open source but that covert industrial-scale distillation attacks crossing into IP theft would put sanctions and entity-list designations on the table.
Commerce Secretary Howard Lutnick became the lone administration official pushing competition over a ban, reportedly meeting multiple labs. His case got a boost when Kimi K3 was revealed to lag significantly behind Mythos on crucial cybersecurity benchmarks. Y Combinator's 'Little Tech Association' also pushed back, warning of a startup extinction event.
On Friday a widely supported open letter invoked the history of open-source software, arguing US AI leadership will be judged by whether it builds a strong open ecosystem — open weights being models anyone can download, inspect, modify and run on their own infrastructure.
It argued distillation — using one model's outputs to improve another — is a legitimate, long-standing technique, while unlawful extraction from closed models should be handled through targeted legal frameworks, not sweeping bans. It even claimed openness aids safety by removing single points of failure.
— Jensen Huang, NVIDIA CEO. Jensen Huang led the charge with his first-ever X post backing the letter, arguing open models strengthen safety and cybersecurity, accelerate diffusion, and enable sovereignty.
Google's Sundar Pichai (citing Gemma), Microsoft's Satya Nadella, Meta's Mark Zuckerberg ('open source is a positive and important force... preventing centralization'), and Elon Musk ('This has my full support. Jensen is right') all backed the letter alongside dozens of firms.
— Palantir. Palantir, a onetime new entrant now standing with tech's biggest players, framed open-weight AI as essential to sovereignty, competition, and keeping customers in control.
— Packy McCormick, Not Boring. Packy McCormick's gleeful read: capitalism aligning corporate self-interest with consumer benefit. Others like Aravind Srinivas and Aaron Levie called it the biggest, most broad-based show of alignment tech has ever seen.
Sam Altman initially responded warmly without signing ('I want the US to win in AI, both in open source and proprietary models'), but after significant pushback OpenAI was added as a signatory — leaving Anthropic as the sole frontier-lab holdout.
— David Sacks. Former AI czar David Sacks predicted 'the gaslighting begins' — that Anthropic won't stop until it kneecaps open source — and urged the industry to watch them like a hawk.
— Dario Amodei, Anthropic CEO. Anthropic's refusal to sign tracks with CEO Dario Amodei's repeated warnings about the risks of powerful open models, especially Chinese ones — comments made in a June Bloomberg interview.
— Buco Capital. Buco Capital's cynical read: OpenAI and Anthropic are objectively kicking everyone's butt on revenue, and the coalition is scared rivals talking their book. The open letter, they argued, suggests the two labs are doing even better than thought.
— Sam Altman, OpenAI CEO. Sam Altman's framing suggests OpenAI's support isn't purely reputational — his stated fear of a small number of companies controlling AI intersects with the open-weight case.
Several commentators praised Anthropic for standing on principle rather than signing and lobbying quietly. Signal noted that if an open-weight model is ever involved in a serious security incident, Anthropic's position 'will look a lot more prescient' — the nature of a principled stance.
Anthropic technical-staff member Julian Schrittwieser mocked the coalition ('Can't wait for the open sourcing of Windows and MS Office'), drawing aghast responses. Bill Gurley, Matthew Berman and A16Z's Justine Moore piled on — Moore quipping about adding 'aura loss from employee tweets' to Anthropic's S-1 risk section.
— Rune, OpenAI (via Ethan Mollick). OpenAI's Rune argued the two positions are incompatible; Ethan Mollick countered that most open-weight advocates simply don't share lab insiders' belief in grave semi-autonomous AI risks — and therefore suspect the labs of spreading FUD to protect margins via regulation.