Meta, Nvidia open AI models as Chinese rivals take 6.1% share

Meta and Nvidia are releasing open-weight AI models this week to counter Chinese rivals that have captured 6.1% of business AI spending.

Meta and Nvidia are returning to open-weight AI models this week, a strategic pivot aimed at countering Chinese rivals whose low-cost models have captured 6.1% of business AI spending, up from 4.5% a year earlier. The moves reflect a growing recognition that open-weight models will play a bigger role in making AI widely available as companies shift from heavy spending to scrutinizing efficiency.

"For so many of the basic operational tasks that are being done, you don't need the most cutting frontier model. So it becomes a return on investment question for these companies," Marc Bhargava, managing director at General Catalyst, said.

Meta on Monday unveiled Muse Glimmer and said it would open the weights for Muse Spark 1.2, its most advanced model. Nvidia followed Tuesday with Nemotron 3.5 Lightning, which the chipmaker said is "truly open source" because it publishes training datasets, techniques, and model weights. Meta plans to spend up to $145 billion on AI infrastructure this year, giving it a scale advantage few open-weight companies can match.

The moves threaten IPO-bound OpenAI and Anthropic, which must convince investors their technological edge justifies the staggering cost of developing and operating frontier models. Meta's $1.5 trillion market cap provides computing capacity that most open-weight competitors cannot replicate.

Chinese models' cost advantage

Chinese open-weight models from Moonshot AI, Z.ai, and Alibaba's Qwen have taken Silicon Valley by storm this year, generating code nearly as well as the best from Anthropic and OpenAI at a fraction of the cost. Demand for American alternatives is already visible on AI developer platforms such as OpenRouter, where Nvidia's Nemotron 3 Ultra ranks above Chinese models including MiniMax's M3 despite trailing them on performance benchmarks.

"People don't want to use Chinese models. That's their default position. But they are so cheap that the economics drive them to use it," said Ameya Kanitkar, co-founder and chief technology officer at Larridin, a San Francisco-based startup that helps businesses optimize AI adoption. "So the natural default position to win here is an American open-weight model."

Data from fintech provider Ramp released Wednesday showed about 6.1% of businesses spending on AI used platforms offering open-weight and Chinese-developed models in July, up from 4.5% a year earlier. Ramp lead economist Ara Kharazian said the trend has yet to dent spending on OpenAI and Anthropic, but their adoption is slowing, particularly for the ChatGPT maker.

Trust and transparency hurdles

Chinese models are still viewed with suspicion by many American companies worried about data security and biases absorbed during training. Running Chinese open-weight models on American platforms addresses data-security concerns by keeping customer prompts within the cloud, but unease persists because the models are not fully open source — companies release weights while keeping training data and code private.

Box CEO Aaron Levie, a signatory of last month's open letter urging policymakers not to place "premature restrictions" on open-weight AI models, called Zuckerberg's plan for Muse Spark 1.2 a "very big deal" because it rivals top foundation models from Anthropic and OpenAI. "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 before with Llama, but the release of Llama 4 in April 2025 left developers unimpressed. The company spent billions overhauling its AI unit, installing Scale AI CEO Alexandr Wang as division leader. Umesh Sachdev, CEO of business AI startup Uniphore, said Meta burned bridges with third-party developers when it shifted from open weight to proprietary models. "The emotion of my developers at Uniphore, they almost feel betrayed," Sachdev said.

Forrester analyst Charlie Dai called Meta's latest move strategically important because it restores a major U.S. frontier AI vendor to the open-weight community. The company must now prove it can cultivate a durable developer base beyond releasing competitive models, Dai said.

Meta's commitment to open models remains unclear. The company plans to release weights for Muse Spark 1.2 but has not said whether it will do the same for Watermelon, a more powerful system reportedly under development. That decision could show whether Meta is testing the waters or, as Zuckerberg put it, working to make "American open source models to be the best globally."

For investors, the open-weight shift carries direct implications. OpenAI and Anthropic face pressure on pricing power as cheaper alternatives gain adoption, while Nvidia stands to benefit from increased demand for inference infrastructure. Together AI's $240 million Nvidia-powered AI cluster on IBM Cloud is expected to sell out two to three months before launch because of strong demand, chief revenue officer Kai Mak said.

This article is for informational purposes only and does not constitute investment advice.