Google told Meta in March 2026 that it could not supply the full volume of Gemini model capacity Meta had sought to purchase. The restriction, first reported on Sunday by sources familiar with the matter, delayed several of Meta’s internal AI projects and forced the company to instruct employees to use AI tokens more efficiently. Other Google Cloud customers also faced capacity limits, but Meta – whose demand for Gemini was exceptional in scale – was the most significantly affected. Gemini had become a key operational tool for Meta across content moderation, advertiser chatbots, coding assistance, and scam detection, specifically because it outperformed Meta’s own Llama open-source models for those tasks. The capacity restriction has been in place since at least March. Google and Meta both declined to comment. YourNewsClub views the combination of non-comment from both companies as itself a commercially legible signal – neither party benefits from public discussion of the arrangement, for different reasons: Google because it acknowledges an inability to serve a major customer at the contracted scale; Meta because it reveals an operational dependency on a competitor.
The numbers around Google’s own infrastructure make the restriction structurally credible. Google Cloud generated more than $20 billion in revenue in Q1 2026, up 63% year-on-year, with Sundar Pichai acknowledging that compute constraints held back stronger results and pushed the order backlog to approximately $460 billion. Despite $180 billion to $190 billion in guided 2026 capital expenditure, demand materially exceeded supply. Google signed an agreement to lease capacity from SpaceX at approximately $920 million per month as bridge capacity while its own data centres expand. That a company spending $180 billion building its own infrastructure simultaneously needs to rent capacity from a competitor illustrates the scale of the demand-supply gap.
The Meta situation reveals something specific about how AI compute is now allocated: access is not purely commercial. A company can attempt to purchase more compute than a provider can supply, and the provider will ration rather than expand in real time. That is a fundamentally different market structure than traditional cloud computing. Meta has no cloud business of its own. Its response to the Gemini restriction was to push its own Muse Spark model to take on more of the content moderation and safety work that Gemini was performing. Meta has also projected $115 billion to $135 billion in capital expenditure for 2026 and is accelerating its own data centre buildout.
YourNewsClub logs the Gemini restriction as an accelerant for Meta’s vertical integration strategy – the episode will be cited in every future internal argument about whether to depend on external model providers.
Owen Radner, who models digital infrastructure as energy-information transport systems, draws the architectural implication: “What the Google-Meta situation reveals is that AI compute is now a rationed commodity even at hyperscale. A company the size of Meta cannot buy its way to unlimited capacity from Google, because Google does not have unlimited capacity to sell. Every enterprise building AI workflows on external model APIs should treat this episode as a supply chain risk disclosure. The assumption that cloud capacity is effectively unlimited is no longer defensible.” Alex Reinhardt, who tracks financial systems and settlement infrastructure through digital protocols, places the competitive logic: “Meta is simultaneously Google’s customer, Google’s advertising competitor, and now a competitor in AI model development. Google’s rationing of Gemini capacity to Meta is rational given that constraint – it prioritises customers who do not compete with it for the infrastructure margin. Understanding which customers a hyperscaler prioritises in a capacity crunch tells you a lot about its actual strategy.”
Your News Club pins the Gemini capacity restriction as the most concrete demonstration yet that AI compute is now allocated by priority and relationship as much as by price.
For enterprise buyers currently running production AI workflows on external model APIs, the Meta-Google episode is a supply chain risk disclosure that requires a contingency plan. YourNewsClub marks whether Google publicly discloses its customer prioritisation methodology in any forthcoming regulatory or investor communications as the accountability moment the industry needs.