Meta gives away Llama AI models to weaken $965B rival Anthropic

Meta is spending billions on AI infrastructure only to give its most powerful models away — a strategy that could undermine the premium-pricing model of market leader Anthropic.

Meta's decision to release its Llama models as open weights while rivals charge premium prices for proprietary AI represents a calculated bet that platform dominance will prove more valuable than model exclusivity.

"Open-weight models expand competition, lower costs, and speed AI adoption across industries," Nvidia Chief Executive Officer Jensen Huang said in a July 25 open letter backed by Meta, Microsoft, IBM and others, arguing the approach strengthens American AI leadership.

The strategy puts Meta in direct opposition to Anthropic, now valued at $965 billion and commanding more than 60% of combined OpenAI-Anthropic enterprise subscription spend, according to Ramp data. Anthropic has overtaken OpenAI in US business adoption for the first time, with Claude models leading on real-world agentic tasks. OpenAI, which initially resisted open-source releases, has since signed the same open-weights letter and released its gpt-oss models after Chief Executive Sam Altman said the company was on the "wrong side of history."

Meta spent an estimated $37 billion on AI infrastructure in 2025, according to company filings, and turning Llama into the default foundation for third-party development could generate returns through platform lock-in rather than direct model sales. For investors, the question is whether Meta can sustain the infrastructure spending required to keep Llama competitive while forgoing the subscription revenue that rivals generate.

Why Open Weights Threaten Anthropic's Premium

Anthropic's valuation thesis rests on maintaining a performance gap wide enough to justify premium enterprise pricing. A world where capable open-weight models from Meta, Mistral and Chinese labs like DeepSeek narrow that gap threatens the company's pricing power directly. Microsoft AI Chief Executive Mustafa Suleyman has said openly that Microsoft wants to reduce and eventually eliminate what it pays Anthropic, despite being one of its largest enterprise customers. Microsoft Chief Executive Satya Nadella has separately argued that foundation models are getting commoditized as a category.

The dynamic creates a classic prisoner's dilemma. Every AI lab except the market leader has an incentive to support open weights, even at some cost to their own future pricing power, because commoditization hurts the leader most. OpenAI, Meta, Microsoft and Nvidia all signed the open-weights letter. Anthropic did not. Journalist Tae Kim called the OpenAI signing "an amazing turn of events," given the company had previously warned the US government about foreign labs distilling its outputs into cheaper open alternatives.

Nvidia's Hidden Incentive

Nvidia's support for open weights is the most straightforward of any signatory. The company sells the computing infrastructure needed to train, fine-tune and deploy AI models, regardless of which lab builds them. Open-weight models shift GPU demand from a handful of hyperscalers to startups, governments and enterprises running models on their own infrastructure, expanding Nvidia's total addressable market dramatically.

The strategy creates an uncomfortable tension for US policymakers. DeepSeek, one of China's most capable open-model developers, operates roughly 20,000 H100-equivalent GPUs and expects to receive large batches of Nvidia-powered systems in coming months, according to leaked comments from founder Liang Wenfeng. Broader chip access strengthens overseas competitors even as it expands Nvidia's addressable market. Huang's letter frames the issue as one of American competitiveness against China, but the same argument could be used to justify chip sales that ultimately benefit Chinese AI labs.

Investor Impact

Meta shares, which fell 1.8% on July 25, trade at roughly 22 times forward earnings — a discount to the broader tech sector that reflects market skepticism about heavy AI spending without clear near-term returns. If the open-weights strategy succeeds in making Llama the default platform for enterprise AI development, the payoff could come through increased engagement across Meta's advertising business and new revenue from cloud partnerships. If it fails, Meta will have spent tens of billions subsidizing a commodity that rivals monetize directly.

For Nvidia, the calculus is simpler. Every new model, whether built in Silicon Valley or Beijing, creates demand for its GPUs. The company's data center revenue reached $47.5 billion in its most recent fiscal year, and open-weight proliferation could accelerate that growth by broadening the customer base beyond the current handful of hyperscale buyers.

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