Google limits Meta’s access to Gemini AI capacity

Google has limited Meta’s access to Gemini AI capacity after Meta sought more computing power than Google could supply, exposing a sharp contradiction in the AI market. The two companies compete in social platforms, advertising and increasingly AI, yet Meta depends on Google for model access that now appears to be constrained by infrastructure, not just software.

Google told Meta around March 2026 that it could not meet the full Gemini capacity Meta wanted to buy. The shortfall disrupted and delayed some of Meta’s internal AI projects, turning what would ordinarily be a procurement issue into a direct operational setback for one of the industry’s largest companies.

The limits were not confined to Meta. Several other Google customers also faced restrictions, though to a lesser extent, pointing to a broader squeeze on premium AI compute rather than a single dispute between rivals. Alphabet said on its first-quarter 2026 earnings call that Google Cloud’s backlog had reached $462 billion, a figure that underscores how much demand is pressing against available capacity.

Google has also been signaling how quickly that demand is rising. At Cloud Next in April 2026, the company said its first-party models were processing more than 16 billion tokens per minute through direct API use by customers, up from 10 billion in the previous quarter. Alphabet later said more than half of its 2026 machine-learning compute investment was expected to go to the Cloud business, while its expected 2026 capital expenditures were set at $180 billion to $190 billion, roughly double 2025 spending and about six times 2022 levels.

The pressure is visible inside Gemini itself. Google shifted its consumer app in 2026 to compute-based usage limits, replacing fixed daily prompt caps with limits that vary by prompt complexity, features used and chat length. That change reflected the same underlying constraint now shaping enterprise access: capacity is finite, and the company is rationing it.

For Meta, the episode shows how the AI race has moved beyond model quality alone. Access to frontier systems now depends on chips, cloud resources, and the willingness of a supplier to reserve enough compute, even when that supplier is also a rival.