More than 75 big tech companies have joined Nvidia CEO Jensen Huang's call to promote the idea that AI models should be open-source – available for anyone to download and run.
The argument is that U.S. leadership in AI will come from "diffusing it into the workflows of factories, hospitals, farms, classrooms and main street businesses," reads the letter, which software and cloud-computing giant Microsoft has now posted on its website.
For at least one prominent developer of open-source AI agents, the principles outlined – however worthy – should be read with extreme cynicism.
Shaw Walters of Eliza Labs, whose agent project ElizaOS broke out to a billion-dollar-plus valuation in crypto markets last year, says the tech firms' avowed principles may come secondary to their own commercial interests. He says everything he builds is open-source, because it's his ideology, "but I'm also a brutal capitalist who's gonna win."
"Nobody here has any ideology," Walters told DeAI News in an interview in San Francisco. "They say they do, but it is very convenient that it's the most profitable angle for them."
The episode exposed a massive rift among some of the biggest AI firms while also appearing to counter the protectionist and national-security interests asserted by U.S. President Donald Trump's administration, typically seen as friendly to the AI and crypto industries.
Anthropic, maker of Claude and some of the world's most powerful AI models, has notably refused to sign on to the effort, with CEO Dario Amodei publishing a blog post insisting the company isn't calling for a ban of open-source models, but that the models should go through "mandatory safety testing" to ensure they don't jeopardize "military superiority" or unleash a wave of cyberattacks.
The AI firm has not open-sourced its own models, including the powerful Mythos, now known as Fable, whose release earlier this year was delayed by the Trump administration on the grounds that it could threaten the security of the nation's internet and banking infrastructure. Anthropic makes money from charging Claude users.
"Anthropic is just defending their financial position," Walters said.
Huang's letter garnered about 25 signers, a number that has since swelled past 75, pulling in even OpenAI and Google, two big AI labs that initially sat out.
The timing isn't an accident. The Trump administration has been leaning on labs to slow their releases, asking OpenAI to limit its latest model to a small set of government-approved partners, and Washington is now weighing a ban on Chinese models after Moonshot AI dropped Kimi K3, a Chinese open-weights system that rivals the best American models at roughly 40% less cost.
Even former Trump AI czar David Sacks, who still acts as a White House science and technology adviser, came down on the open-source side, warning that federal pre-approval of models is "how you lose the AI race."
Huang's Nvidia, as a giant maker of chips for AI, sells the picks and shovels, so he wants the technology everywhere.
Mark Zuckerberg of Meta, which signed the letter, might be one of the most aligned, according to Walters, since the company's Llama models were "the only open source game in town for a long time." In April, Meta reversed course and released its new Muse Spark model as a closed-source version, but reports have indicated the firm could still release open-source models.
Anthropic's bioweapon fear
Walters saves his sharpest words for the safety argument, which he sees as a convenient rationale for not releasing its AI models to the public.
Anthropic CEO Amodei, in his response, said that "sufficiently capable models may be able to quickly weaponize pandemic-level viruses with widely available materials, whereas defense against these agents is a multi-year operational task in the best case."
Walters says it's extremely unlikely that a user would be able to pull this off.
"Interpol and FBI would show up at your door within a few hours because they're tracking all of the precursors and like materials and chemicals for bioweapons and nuclear weapons," he said.
Walters likened the brouhaha to the 1990s fight over encryption, when Washington warned that strong cryptography would empower criminals, right up until it became the thing securing everyone's bank account.
What's far more worrisome, according to Walters, is the prospect that the most powerful AI models end up concentrated in a few hands.
"Think about going to school and one person has unlimited Claude of the best possible and the other kid has no subscription," he said. "Who's going to do better in school?" Wall it off behind a subscription, he argues, and you manufacture "a permanent underclass, permanent overclass," where a few people own billions of robots making value and everyone else lives on welfare. "I just don't want to live in that world."
Strip out the gatekeepers, he says, and you get "the cheapest, most efficient market of global intelligence," instead of "relying on a very small amount of people to be the arbiters of knowledge, information and truth."
"There are a lot of ideological people, but we also have to align the ideology with the capitalist incentive," he said.
(HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: This draft was written on Claude Desktop from a transcript and summary of the author's interview with Shaw Walters. Background facts about the open-weights letter, the signatories, the Trump administration's model-release restrictions, Kimi K3, David Sacks' comments and Anthropic's position were researched via web search and checked against the linked original sources before publication.)