Meta in Talks to Invest Over $10bn in Scale AI, Deepening Its Bet on Artificial Intelligence

Meta Platforms Inc. is reportedly in advanced talks to make a multibillion-dollar investment in Scale AI, the data infrastructure startup powering some of the most important players in the generative AI race.

According to Bloomberg, People familiar with the matter say the deal — still in flux — could exceed $10 billion, making it one of the largest private funding rounds ever seen in the tech industry.

While neither company has confirmed the potential investment, it would mark a major strategic pivot for Meta, which has traditionally relied on its own in-house AI research and open-source collaborations rather than funneling billions into outside startups, as some of its rivals have done.

Scale AI, co-founded in 2016 by CEO Alexandr Wang, has quietly grown into a cornerstone of the AI supply chain by providing data-labeling services essential to training machine learning models. With a client roster that includes OpenAI and Microsoft, the company made $870 million in revenue last year and expects that number to more than double to $2 billion by 2025.

Scale was valued at $14 billion during a 2024 funding round that included Meta and Microsoft. A subsequent tender offer reportedly valued the company at $25 billion.

A Rare Power Move by Meta

If finalized, this would be Meta’s biggest external AI investment to date, signaling a shift in strategy as the company ramps up its efforts to dominate the AI frontier. So far, Meta has been largely self-reliant in developing its Llama series of large language models, which now power chatbots used by over a billion people monthly across Facebook, Instagram, and WhatsApp. CEO Mark Zuckerberg has committed up to $65 billion for AI projects this year alone, and making Scale a deeper partner could help the company fill critical gaps in training data and infrastructure.

Unlike Microsoft, Amazon, or Google, Meta does not operate a public cloud business, and it remains unclear whether its deal with Scale would include infrastructure credits or take a different shape entirely. But the move would help Meta keep pace in an increasingly competitive arms race, where rivals have poured massive resources into AI development: Microsoft has invested more than $13 billion in OpenAI, while Amazon and Google have committed billions to Anthropic.

Strategic Alignment Beyond Commercial AI

Beyond the commercial AI space, Meta and Scale have already found common ground in military applications. Just last week, Meta announced a new partnership with defense contractor Anduril Industries to build military-grade technologies, including an AI-powered mixed-reality helmet. The deal confirmed Meta’s willingness to expand its AI ambitions beyond consumer tech and into defense infrastructure.

Scale, for its part, has already established itself as a key AI contractor to the U.S. government. It recently secured a Pentagon contract for developing advanced AI agents — a deal the company described as a “significant milestone in military advancement.” It is also working with Meta on “Defense Llama,” a customized version of Meta’s Llama large language model designed specifically for military use.

The emerging defense tech collaboration between Meta, Scale, and Anduril underlines a growing convergence between Silicon Valley and Washington, with AI now seen as central to both economic competition and national security. Meta has even begun licensing its AI models for use by U.S. agencies and defense contractors, marking a departure from its earlier more cautious stance on military applications.

Scale AI’s Strategic Importance

Scale’s business is built on the idea that powerful AI systems are only as good as the data that trains them. The company uses tens of thousands of contract workers to scrub, label, and structure massive datasets — a critical but often overlooked aspect of building reliable AI models.

Meta is not just investing in another AI unicorn by doubling down on Scale. It’s securing access to the foundational layer of training data infrastructure, which could prove to be as important to its long-term AI ambitions as GPUs or software models.