メタとNvidia、中国企業主導のオープンウェイトAI競争に明確な旗を掲げる

(L to R) Jensen Huang, President and CEO of Nvidia Corporation, Google CEO Sundar Pichai and Mark Zuckerberg, CEO of Meta, arrive for a U.S. Senate bipartisan Artificial Intelligence Insight Forum at the US Capitol in Washington, D.C., on Sept. 13, 2023.

Andrew Caballero-reynolds | Afp | Getty Images

Last month, American tech giants came together to urge policymakers not to place “premature restrictions” on open-weight AI models, even if they’re from China. Now, two of those companies are making a concerted effort to compete by introducing their own open offerings.

Meta and Nvidia both released artificial intelligence models this week that are available for developers to download for free via the open-source ecosystem, a contrast to the popular proprietary models from OpenAI and Anthropic.

Open-source AI has become a contentious topic from Silicon Valley to Washington, D.C., with critics raising concerns about the potential national security risks of Chinese models, and of the AI training practice called distillation, which can be viewed as a form of intellectual property theft. Meanwhile, most of the industry’s leading players contend that restricting use of the models would be to our own detriment and would place too much power in the hands of too few companies.

“The age of AI can be one of prosperity,” the consortium of tech companies wrote in an open letter on July 24. “With the right choices, open weight AI can expand opportunity, strengthen competition, extend American technological leadership, mitigate risk, and ensure that the benefits of this extraordinary technology are shared broadly across our economy.”

As for distillation, they call it “a widely used technique for model improvement, evaluation, and validation.”

watch now

Meta on Monday released Muse Glimmer as part of a strategy to release its most powerful AI models to the open-source community. CEO Mark Zuckerberg said the company would open the weights for its latest AI model, Muse Spark 1.2. Weights refer to the calculations and rules that determine how the AI works and behaves.

A day later, Nvidia debuted Nemotron 3.5 Lightning. The model stems from the company’s Nemotron 3 family of modelsreleased in December. The chipmaker said its models are “truly open source,” because the company publishes therelated “training datasets, techniques, and model weights” for developers to inspect.

Both companies still have to prove there’s an audience for their offerings in a market featuring popular models from Chinese AI labs like Moonshot AI and DeepSeek, as well as Alibaba’s Qwen.

Box CEO Aaron Levie, one of the signatories of last month’s letter, is optimistic. He said Zuckerberg’s plan for Muse Spark 1.2 is a “very big deal” because it’s a powerful model that rivals top foundation models from Anthropic and OpenAI. The models this week from Meta and Nvidia are smaller and intended to run on laptops for tasks like powering on-device digital agents.

“There’s a very firm flag in the ground that America will have near-frontier open-source models,” Levie said.

Meta has tried this route in the past with Llama. That was Zuckerberg’s initial entry into the foundation AI market, but the release of Llama 4 in April 2025 left developers unimpressed. Meta followed by spending billions of dollars to overhaul its AI unit, installing Scale AI CEOAlexandr Wang as the division’s leader.

Recently, Wang’s group has been rolling outproprietary models under the Muse branding to try and develop new revenue streams.

‘Tremendous amount of potential’

Levie said that companies put off by using Chinese open-weight AI models would be more inclined to experiment with Meta’s upcoming variant.

“You probably wouldn’t be able to put a non-domestic open-source model in a major government agency, as an example, and you wouldn’t be able to use it at very large banks most likely,” Levie said. “If you think about the kind of use cases that now Muse can be used in, it actually opens up a tremendous amount of potential.”

Still, Meta in particular faces some headwinds as it pursues yet another open-source strategy. Umesh Sachdev, CEO of business AI startup Uniphore, said Meta burned bridges with third-party developers when it shifted from open weight to proprietary AI models.

“I think it’s going to take more than a 3,500 worded article from Zuck to convince developers,” Sachdev said regarding Zuckerberg’s accompanying manifesto this week. “The emotion of my developers at Uniphore, they almost feel betrayed.”

But Sachdev said he’s rooting for domestic companies to succeed, “because more competition will drive down token cost, and will drive up innovation, and it’s always good for consumers.”

It’s a sentiment shared by Forrester analyst Charlie Dai. He called Meta’s latest move “strategically important because it restores a major U.S. frontier AI vendor to the open ecosystem.”

“Developers and enterprises will likely welcome Meta’s shift back toward open weights because it improves transparency, customization, deployment flexibility, and data sovereignty,” Dai said. Now, the company must “prove it can cultivate a durable ecosystem beyond releasing competitive models,” he said.

WATCH: Meta’s new AI model is a “positive development for the ecosystem.”

watch now