Meta reportedly exceeded Google's Gemini token limits and was cut off

At a glance:

• Meta was using large volumes of Google's Gemini AI tokens in March.
• Google imposed usage caps on Meta, disrupting and delaying internal AI projects.
• The strain on Google's resources may have contributed to its early-June $920 million per month compute deal with SpaceX's xAI.

What happened

In March, Meta emerged as one of the biggest participants in the "tokenmaxxing" trend, where employees were evaluated by how many AI tokens they consumed. The practice coincided with interest in agentic AI platforms such as OpenClaw, which promised heightened workplace efficiency. Internal metrics reportedly tracked token usage per engineer to gauge productivity.

According to three anonymous sources cited by the Financial Times, Google told Meta it could no longer sustain the company's AI consumption and placed limits on its use of Gemini models. The FT quoted the sources: “…disrupted and delayed some of Meta's internal AI projects. Owing to the restrictions, which remain in place, as well as a broader push to streamline AI costs, Meta has encouraged staff to be more efficient with AI tokens — the units that measure AI usage, several people said.” The limits remain in effect as of the FT's reporting.

Impact on Meta's AI initiatives

The token caps have forced Meta to shift from unrestricted tokenmaxxing to judicious token‑counting, affecting several internal AI efforts that relied on heavy Gemini usage. Projects that depended on large‑scale model inference or fine‑tuning have reportedly faced delays. Teams experimenting with generative content creation and code‑assist tools are among those reporting slower iteration cycles.

Meta and Google both declined to comment on the FT's report, leaving the exact scale of the restriction and the affected teams unclear. Nevertheless, the move signals a broader tightening of access to Google's AI infrastructure among large corporate users. Analysts warn that similar caps could emerge for other heavyweight customers if demand continues to outstrip supply.

Broader industry context

The Financial Times notes that other large companies also strained Google's AI capacity and received caps, but Meta's usage was described as exceptional even among those high‑rollers. This suggests Google's Gemini service is under pressure from enterprise demand that rivals its current allocation. Industry observers speculate that Google may need to expand its TPU and GPU farms to accommodate growing token consumption.

The article speculates that the burden on Google's resources may have influenced its early June decision to rent compute capacity from SpaceX, the parent company of xAI, for a reported $920 million per month. That deal would provide Google with additional AI‑training hardware amid growing token consumption across the industry. If finalized, the arrangement could mark one of the largest single‑month cloud‑compute commitments in the AI sector.