Big Tech Unites for Open Source AI - and Against Anthropic

July 28, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
July 28, 2026

HEADLINES

NVIDIA Backs Ilya Sutskever's Safe Superintelligence

A name that hasn't come up much in years is back. NVIDIA has made a substantial, undisclosed investment in Safe Superintelligence, the enigmatic lab Ilya Sutskever founded after leaving OpenAI in the wake of the late-2023 leadership dispute. The deal gives SSI access to next-generation Vera Rubin chips — a 10X increase in compute over the next 12 months, per the company — and sources indicate SSI had previously been leaning on Google TPUs. NVIDIA says it invested after gaining rare access to SSI's technology, which is about as close to a progress report as the outside world has gotten. The whole bet has always been that short-term commercial work is a distraction from the straight shot to superintelligence, so a compute deal, rather than an enterprise sales ramp, is exactly the signal to watch.

The Circular Financing Chart Is Back

NVIDIA stock fell 4% on Monday as roughly three-quarters of a trillion dollars in newly announced deals struck observers as a little too self-referential. The two headliners: a partnership with SK Group worth more than $500 billion in mutual business, and the $250 billion arrangement backstopping OpenAI's new data center construction. That old circular-deal chart from late last year has already been updated and is making the rounds again — expect it to keep resurfacing over the coming weeks.

China's Domestic Chip Push Goes From Encouragement to Ultimatum

The Wall Street Journal reports that CCP operatives are actively pushing adoption of domestic chips, with China's AI czar reportedly warning the country's largest AI users that anyone resisting domestic silicon was a traitor. The confidence came from a closed-door Huawei demo last summer showing dramatic catch-up progress. The numbers track: 90% dependence on US chips in 2021, 60% by 2025, and Morgan Stanley thinks it could hit 25% within five years. Huawei chips still lag NVIDIA significantly — Brookings fellow Kyle Chan likened it to chasing a bullet train — and China's entire supply chain produces around 28,000 advanced logic chips a month against TSMC's output. But Bernstein forecasts that number doubling annually for three years, Huawei alone expects to ship 1.5 million AI chips this year, and a Shanghai company has now begun mass-producing deep ultraviolet lithography machines it was never supposed to have. The debate over whether to restrict chips or flood the country with them is over. That ship has sailed.

Apple and Micron Are Fighting Over Chinese Memory

Apple has been lobbying the White House for approval to buy memory chips from Chinese manufacturers including CXMT, arguing that runaway memory prices lead to $5,000 iPhones. Micron — the only sizable US memory maker, with the rest headquartered in Korea — is now lobbying the other way, with CEO Sanjay Mehrotra warning that letting Chinese producers in would decimate the domestic industry the way steel went. Technically no approval is needed, since the Pentagon Entity List only covers the military supply chain, but nobody wants to route around this White House right now. The administration is caught between a rising inflation problem and its own domestic chipmaking agenda.

MAIN STORY

The Loudest Pro Open Source AI Coalition in History

The open versus closed debate has hit a new fever pitch. An open letter backing open-weight models drew signatures from dozens of companies across big and little tech, led by an NVIDIA CEO who chose it as the subject of his first-ever post on X — a post that pulled 62 million views. Roughly eight trillion dollars of market cap lined up behind open weights, and the two conspicuous holdouts became the story. This is now the single biggest conversation in AI, with real consequences for whatever policy Washington lands on.

EVERYBODY VS ANTHROPIC

How We Got Here
"Policymakers went from quietly concerned to full-on freak-out"
Washington's anxiety about Chinese models had been building for months. The GLM 5.2 release landed with headlines claiming parity with Mythos — including a Forbes entry about the bad guys now having a model as powerful as Mythos, which is the kind of thing that should get a podcast host unsubscribed from. It plainly wasn't Mythos-class for anyone who actually used it, but it was good enough for policymakers, and the first rumblings of an executive order banning open source models followed. Then Kimi K3 arrived as the largest open source model ever released at 2.8 trillion parameters, beating Fable on Arena's front-end code arena by a wide margin, and the freak-out went public with comments from Treasury Secretary Scott Bessent and senior tech official Michael Kratsios. Notably, those comments avoided raw model performance entirely and focused on distillation, with sanctions floated for Chinese firms known to have accessed NVIDIA chips.

The Little Tech Association
"A mass extinction event for US startups if banned"
The first industry pushback came from a trade group representing part of the startup community, arguing that an enormous amount of field work is built on top of Chinese models. Y Combinator was the most prominent member. As vaunted as YC is among founders, the newly formed group visibly lacked the political heft to block a policy with geopolitical stakes. Commerce Secretary Howard Lutnick, meanwhile, emerged as the sole administration official still arguing that competition — not prohibition — is the right answer to Chinese AI, and was reportedly taking meetings with multiple labs.

The Open Letter
"Openness could be a major contributor to AI safety"
The letter opens on the history of open source software, arguing that 1980s pioneers challenged the belief that software advances only under tight corporate control, and that open source created a shared foundation American engineers built on for generations. The United States now faces the same choice with AI: leadership will be judged not by one frontier model but by whether an open ecosystem diffuses into every sector. On the cybersecurity risk, the letter argues prohibition is the wrong response — defenders need models with capabilities comparable to attackers, and openness removes a single point of failure. Its most contested move is drawing a hard line between distillation and IP theft, framing distillation as a legitimate technique in a long tradition of building on existing technology, with unlawful extraction from closed models handled through targeted legal frameworks rather than sweeping bans.

Jensen Huang
"The world needs both frontier closed and frontier open models"
The first post he has ever made on X, and it did 62 million views. The argument: AI will transform every industry, power every company, and be built by every country, and open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

Big Tech Falls In Line
"One by one, the leaders of big tech co-signed"
Sundar Pichai retweeted Jensen on behalf of Google, pointing to the company's long history with open source and its Gemma open-weights releases from DeepMind. Satya Nadella framed open-weight models as essential to a healthy AI ecosystem and to American competitiveness. Mark Zuckerberg shared Nadella's post, calling open source a force for empowering people and preventing centralization. Elon Musk retweeted Jensen with full support. Palantir noted it was a new entrant 20 years ago and now stands against a future where a select few closed frontier models rule them all. Denny's got involved with a Venn diagram.

The Reaction
"Capitalism is just so beautiful, it brings a tear"
One whole strand of the discourse was simply gushing. Packy McCormick marveled at eight trillion dollars of market cap backing open weights because it's good for their business and makes them look good, and only coincidentally happens to be better for consumers and nearly every business except a couple of labs. Aravind Srinivas called it the most positive he has felt about the future of American AI, watching companies unite against regulatory capture. Aaron Levie argued nothing in tech has drawn this much broad-based alignment, and that open weights accelerate diffusion by giving customers more options, lowering costs on certain workloads, and enabling tuning for work a horizontal frontier model won't serve.

OpenAI
"Not a signatory, then pushed into becoming one"
Sam Altman responded within about half an hour, saying he wants the US to win in AI across both open source and proprietary models and was glad to see the letter — while OpenAI was not actually on it. Late Friday night, after a wave of pushback, the company was added to the signatory list. There is reason to think this isn't purely reputational: in a recent podcast appearance, Altman framed the fight of the current moment as AI authoritarianism versus liberty, argued that every historical trade of liberty for safety has been a long-term net loss, and said the technology is going into people's hands. Business Insider read that conversation as revealing a core Altman fear of a small number of companies controlling AI.

Anthropic
"And that left one"
Former AI czar David Sacks framed it as the entire tech industry save Anthropic coming out for open source, and predicted the gaslighting phase: nobody is trying to ban open source, just limit who can use it, who can contribute, how powerful it can be, and whether the guardrails can be removed — with the same net effect either way. The non-signature shouldn't surprise anyone given the company's stated views. Anthropic has openly said it is working with the government to prevent distillation attacks, and most assume it is lobbying for an open model ban behind closed doors. Dario Amodei told Bloomberg in June that Mythos-class cyber abilities available for anyone to download is a serious concern.

Talking Their Book
"The potential losers see an opening, or the door closing"
One clean interpretation: this is everyone talking their book. Bucco Capital argued that OpenAI and Anthropic are objectively beating everyone on revenue ramps, that the letter itself suggests they're doing even better than assumed, and that geopolitics and morality are mostly noise — corporations don't care about that stuff, they care about losing, and they are losing. Silicon Data's Steve Hou pushed on the logic: if the best open models are predominantly Chinese, then supporting open models cuts against the geopolitical and nationalist case. His alternative read is that signatories fear the federal government more than they fear the labs — a ban on Chinese open models on natsec grounds would cement two labs' dominance before US native open models get a chance to sprout.

The Aura Question
"Either the villain or the main recipient of aura"
Where this leaves Anthropic is a Rorschach test. Andrew Curran reshared Sacks and described a crusade against the House of Anthropic, increasingly framed as one company against the entire industry. Packy McCormick noted it would have been easy to sign the letter and lobby quietly behind the scenes, and respected that they didn't. Pseudonymous commentator Signull welcomed the divergence outright — a healthy ecosystem is one where people with fundamentally different visions compete to prove themselves right — and added that any policymaker or CEO who signed now owns the possibility that an open weight model ends up in a serious security incident, at which point Anthropic's position looks prescient. Even OpenAI's Roon weighed in, saying corporations meekly banding together to sign petitions against closed models don't realize how cool they're making Anthropic look. Ethan Mollick pointed at a different Roon post — that there is no consistent belief set where you're AGI-pilled and pro open source — to argue the real split is belief: most people discussing open weights simply don't hold the lab insiders' picture of AGI risk, and if you don't think those risks are real, you conclude the labs are manufacturing FUD to protect margins through regulation.

The Employee Tweet Problem
"Standing on principle works better without snark on Twitter"
Anthropic member of technical staff Julian Schrittwieser retweeted Jensen with a line about looking forward to the CUDA and GPU driver open source release, then added a follow-up on Nadella's post about waiting for Windows and Office to be open sourced. The responses were aghast. David Sacks countered that nobody is arguing all software must be open source — only that open weight AI should be allowed — and read the joke as rejecting the machinations to kneecap the open model ecosystem. Bill Gurley noted the real problem with open is that it competes with a corporate economic strategy, and that building the biggest private market cap ever means expecting competition. Matthew Berman called it a bad take and pointed out NVIDIA has put $20 billion into open source. A16Z's Justine Moore imagined the banker writing Anthropic's S1 adding massive aura loss from employee tweets to the risk section. Jeremy Howard summed it up: it boggles the mind how many otherwise functioning intellects lose all cognitive capacity when thinking about actions that impact the company paying them.