American AI Industry Swings in Favor of Open-Weight Models
An interesting thing happened last week, just a few days after the launch of two highly performant open-weight models from China: All of a sudden, the vast majority of the American AI industry came out in favor of open-weight models. Could this signify a permanent shift in our approach to AI?
The commitment to open-weight models was codified in an open letter signed by CEOs of Meta, Microsoft, Nvidia, IBM, Dell, Huggingface, Palantir, Perplexity, Mistral AI, Reflection and others. Nvidia CEO Jensen Huang’s very first post on X was about the letter, which is titled “Open Weights and American AI Leadership.”
“Our AI leadership will be judged not by one frontier AI model, but by whether the United States build a strong, open ecosystem that diffuses into every sector,” the letter reads. “This is essential for creating opportunities for innovation and prosperity across the country.”
Open-weight models are AI models where the final trained parameters are made available for anyone to run and fine-tune. They are not entirely open, as the exact training dataset and code usually remain private. But open-weight models are significantly more open than proprietary models from the likes of OpenAI, Anthropic, and Google, which have so far garnered the largest share of the market with their closed AI models, GPT, Claude, and Gemini, respectively.
Logos of companies backing the “Open Weights” letter (click to enlarge)
Open-weight models, the letter-writers say, are better because they can be downloaded, inspected, modified, and run by anyone on their own hardware. They’re important because they strengthen competition by preventing AI capabilities from being used by only a handful of companies or labs, which ultimately improve capabilities and drives down costs, they write.
But open-weight models carry risks, such as the potential to craft exploit code for vulnerable software, which is something that OpenAI’s Claude Code apparently did recently, when it broke out of its sandbox and attacked Huggingface. “But the right response to this risk is not to prohibit open weights,” they write. “Open models broaden defensive capability, increase transparency, and allow vulnerability to be discovered and remediated across many teams.” You can read the entire letter here.
Proprietary AI models up to this point have delivered the highest accuracy and performance. The latest release of the Stanford AI Index in April showed that the top American AI models are closed. However, the top two Chinese models, Alibaba, and DeepSeek, are open, and they have closed the performance gap with their American counterparts. Indeed, the Chinese have China has embraced openness. China President Xi Jinping gave a speech recently in which he portrayed the country as a champion of open to AI development.
The top Chinese AI developers upped the stakes in the battle for open-weight supremacy this month with the release of Qwen 3.8 tMax from Alibaba and Kimi K3 from Moonshot AI. Kimi K3 is the world’s first open-source model in the 3-trillion-parameter class, and is designed for frontier intelligence scenarios, including long-horizon coding, knowledge work, and agentic reasoning, Moonshot AI said. According to Moonshot AI’s internal suite of benchmark tests, Kimi K3 outperforms all other AI models except for two: Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol.
Following the news early last week, Michael Kratsios, the director of the White House Office of Science and Technology, publicly condemned the Chinese approach to developing AI, saying Kimi K3 had been distilled from Anthropic Fable.
“The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models,” Kratsios wrote in a post on X. “Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
Michael Kratsios, the director of the White House Office of Science and Technology
Kratsios, who is a senior advisor to the president, claimed that Moonshot AI used Nvidia B300 GPUs–the powerful Blackwell Ultra chips that shipped less than a year ago–to train the model, in violation of US export laws. US Treasury Secretary Scott Bessent this week said the administration would consider sanctions as a punishment for “industrial-scale distillation attacks.”
During the Genesis Mission Summit in Washington, D.C. last week, Darío Gil, the Department of Energy’s Undersecretary for Science, addressed the prospect of both Chinese and proprietary models being used in Genesis Mission projects. The U.S. government will not allow Chinese models in the National Labs, he told HPCwire, and it’s backing both proprietary and open-weight models developed in the U.S. and its allies.
During a panel at Genesis Mission Summit, Energy Secretary Chris Wright posed a question to Misha Laskin, the Reflection AI co-founder and CEO and a panelist. “Darío says we have to lead in open models and proprietary models,” Wright said. “Are we going to do it?”
“I think we will,” Laskin said. “I think we have all the capability here. There’s the right kind of commercial incentive and community incentives to build an open-source frontier intelligence laboratory.”
Another panelist was Karthik Narain, who leads product engineering and market capabilities for Google, which releases some of its Gemma models, but not Gemini, which is only accessible via a cloud API. “We’re a founding member [of Genesis Mission], but we are proud to be making a time commitment today of $40 million,” Narain said.
Meanwhile, other proprietary frontier model developers have stated their position on open-weight models, including Anthropic. Dario Amodei, Anthropic CEO, stated that while open-weight models carry additional risks, Anthropic has never called for a ban on open-weight models.
“Open-weights models–it does not matter whether they come from China or anywhere else—do potentially present a higher risk than closed models, because it is very difficult to apply guardrails to them or monitor their usage, and once weights are released they cannot be withdrawn,” Amodei wrote in a July 27 blog post titled “Our position on open-weights models.”
Interestingly, OpenAI is one of the signatories on the “open weights” letter. But OpenAI is also a signatory to another open letter, dubbed “Pacing the Frontier,” that calls for slowing the development of AI.
“To realize AI’s potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight,” states the letter, which was signed by Amodei and more than 1,200 employees at frontier labs (Altman was not among them).
OpenAI is taking another approach to getting its AI models in the hands of as many people as possible to improve humanity: It’s giving its top model away to 100,000 researchers and scientists as part of its new ChatGPT for Academic Researchers program.
“We believe the benefits of frontier AI should not be concentrated in a few companies and well-resourced labs,” OpenAI wrote in a blog post today. “Scientific progress depends on researchers asking the right questions, testing new ideas, and building on what others have discovered.”
AI as we know it emerged as proprietary software from OpenAI and Google, which still develop the biggest and best models in the industry. But now much of the momentum in AI appears to reside with open-weight models. The big proprietary model makers have responded by opening access to its tools—as Google is doing by giving researchers $40 million in tokens, and OpenAI is doing by giving free access to its tools to 100,000 researchers.
What a difference a week makes.
Related
NVIDIA Announces Financial Results for 1st Quarter Fiscal 2027
SANTA CLARA, Calif., May 21, 2026 — NVIDIA (NASDAQ: NVDA) has reported record revenue for…
Broadcom Announces VMware Cloud Foundation 9.1
PALO ALTO, Calif., May 5, 2026 — Broadcom Inc., a global technology leader that designs,…
Cerebras Systems Announces Launch of Initial Public Offering
SUNNYVALE, Calif., May 4, 2026 — Cerebras Systems, Inc. today announced that it plans to commence…
NVIDIA Launches Cosmos 3, the Open Frontier Foundation Model for Physical AI
TAIPEI, Taiwan, June 1, 2026 — NVIDIA today launched NVIDIA Cosmos 3, an open world…
OpenAI Launches Deployment Company to Scale Enterprise AI Adoption
OpenAI has agreed to acquire Tomoro, giving the OpenAI Deployment Company experienced Forward Deployed Engineers…
SpaceX, OpenAI, and Anthropic IPOs: Can Wall Street Finance the AI Boom?
Elon Musk’s SpaceX has filed for what could become the largest IPO in history. It…
American AI Industry Swings in Favor of Open-Weight Models
An interesting thing happened last week, just a few days after the launch of two…
Qualcomm Completes Acquisition of Modular
SAN DIEGO, July 29, 2026 — Qualcomm Incorporated today announced that it has completed its…
Encore AI Lands $30M Series A for AI That Replicates Top Sales Performers
NEW YORK, July 29, 2026 — Encore AI today announced a $30 million Series A…
groundcover Raises $100M Series C to Create the Observability Platform Built for the AI Era
SAN FRANCISCO, July 29, 2026 — groundcover, a leading bring-your-own-cloud (BYOC), eBPF and OpenTelemetry (OTel)-native observability…
NSF Awards 1st CyberAICorps Scholarships to Train AI Cybersecurity Professionals
July 29, 2026 — The U.S. National Science Foundation‘s Directorate for STEM Education has announced the…
You Cannot Govern What You Cannot See: Closing the Visibility Gap in AI Agents
IT governance has relied on the rule that if you can’t see it, you can’t…
