Meta is reportedly running into a problem many AI buyers now face: even Google may not have enough AI capacity to go around.
According to a Financial Times report citing people familiar with Google’s operations, Google has limited Meta’s access to its AI models because of computing constraints. The limits have reportedly delayed some Meta AI projects and pushed the company to use tokens more efficiently or consider alternative models.
The bigger wrinkle is that Meta is not just another customer. Its reported reliance on Google Gemini and other frontier models suggests that even the companies building their own AI systems may still need outside help for coding, agentic tasks, and large-scale development work.
Google struggling with how best to allocate capacity
Google has been struggling to build out capacity quickly enough to satisfy both internal projects and external buyers. It recently signed a $920 million-a-month deal with SpaceX for access to its Colossus 1 data center and raised $80 billion to accelerate its AI infrastructure buildout.
Google CEO Sundar Pichai said on a recent quarterly earnings call that the company could be generating far more revenue from Google Cloud if it were not using some of that capacity for its own projects. Google has not said how much revenue it is currently making from AI services, but it appears to be all-in on AI as the future of its business, as shown by its recent changes to search.
Even though Google is seen as a distant third to Anthropic and OpenAI in coding and agentic services, it still has major customers spending millions of dollars on its services. Its announcements at Google I/O in June also looked to strengthen its enterprise portfolio, with a major focus on building AI agents that can operate independently.
Cost reductions may be on the cards, despite demand
The next move for the industry appears to be reducing AI expenditure, either through more targeted and focused AI usage, or by cutting overall usage in some cases. Meta, Uber, Coinbase, and several other major tech companies have all said they are looking for ways to improve efficiency and reduce costs.
To help with this, several vendors are reportedly looking to reduce the price per token, including OpenAI and Anthropic, while also making their AI models more efficient so users can do more while spending less.
However, even these pure-play AI model makers are facing massive demand for their services, making cost reduction difficult. With demand expected to outweigh supply in several key areas, including data center buildout, capacity, and components, for the rest of the decade, these cost reductions may be little more than temporary attempts to win over more customers.
For now, the report points to a broader problem facing the AI market: demand for frontier models is rising faster than infrastructure can comfortably support it.
That gives Google a valuable business opportunity, but also a difficult balancing act. The company has to serve cloud customers, support its own AI products, and keep pace with competitors building models just as aggressively.
Also read: Our AI pricing cheat sheet compares what major chat, image, video, and voice platforms cost as AI usage gets more expensive.