Google is rationing Gemini access to Meta because it cannot provide enough compute

At a glance:

• Google has imposed compute limits on Meta's access to Gemini AI models
• Meta is shifting workloads to its internal Muse Spark model
• The restrictions reflect a broader industry-wide AI compute shortage

What happened

Google has placed strict limits on Meta's use of its Gemini AI models due to insufficient computing capacity, as reported by the Financial Times. This move has disproportionately affected Meta, which had relied on Gemini for critical tasks like content moderation and scam detection. The restrictions emerged as Google itself faces surging demand for its AI infrastructure, forcing it to allocate resources unevenly. While Google owns one of the largest AI infrastructure portfolios, it is now renting 110,000 Nvidia GPUs from SpaceX at $920 million monthly to meet Gemini Enterprise demands. This paradox highlights a systemic issue: AI compute supply cannot keep pace with consumption despite massive investments.

Meta's reliance on Gemini was strategic. The model outperformed its own Llama open-source alternatives in safety processes, making it ideal for automating harmful content removal. However, the compute crunch has forced Meta to prioritize efficiency. Internal directives now require staff to optimize AI token usage, a euphemism for reducing reliance on external models. This shift accelerates Meta's pivot to Muse Spark, a new internal model developed under its Superintelligence Labs division. The company has already reallocated 7,000 workers to AI roles and committed $115-135 billion to 2026 capex, signaling its urgency to build self-sufficient capabilities.

Meta's internal restructuring

Meta's response to the Gemini limitations reveals a broader strategic recalibration. The company cut 8,000 jobs in May, redirecting funds toward AI infrastructure. This move aligns with its 2026 capex guidance, which prioritizes building internal AI models over third-party dependencies. Muse Spark, while still in development, is positioned as a long-term solution for high-stakes workloads. By reducing Gemini usage, Meta aims to mitigate risks associated with depending on a competitor's infrastructure. This approach mirrors similar trends at companies like Anthropic, which recently rented an entire data center from SpaceX to address compute shortages.

Industry-wide compute crisis

The Google-Meta dynamic is symptomatic of a larger AI infrastructure bottleneck. Demand for GPUs and TPUs has outpaced supply, creating a competitive scramble for capacity. Google's partnership with SpaceX exemplifies this trend, as does Anthropic's data center rental. Even Meta, despite its aggressive AI investments, is constrained by its cloud provider's limits. This scarcity is reshaping corporate relationships, with companies increasingly seeking alternatives to avoid vendor lock-in. The situation underscores a fundamental challenge: AI's exponential growth demands physical infrastructure that current supply chains cannot sustain.

The EU tech scene's reaction

The EU has taken particular interest in this compute crunch, which could influence regulatory approaches to AI infrastructure. While the region's tech ecosystem is still developing, the Gemini-Meta case highlights vulnerabilities in global AI supply chains. EU policymakers may scrutinize how compute access affects market competition, especially as major players like Google and Meta navigate these constraints. The EU's focus on ethical AI and data sovereignty could intersect with these technical challenges, though no direct regulatory action has been announced.

What to watch next

The coming months will reveal whether Meta's shift to Muse Spark can close the performance gap with Gemini. If successful, it would mark a significant step toward AI self-sufficiency for large tech firms. Meanwhile, Google's reliance on SpaceX GPU rentals may set a precedent for cross-industry infrastructure partnerships. Investors should monitor capex reports from both companies, as well as any announcements about new AI hardware or data center deals. The broader implication is clear: the AI boom's next phase will be defined by who can secure and optimize compute resources most effectively.

Conclusion

The Gemini access restrictions between Google and Meta are more than a technical issue—they signal a transformative phase in AI development. As compute becomes the new bottleneck, companies will prioritize vertical integration over open ecosystems. This trend could accelerate consolidation in AI infrastructure or spur innovation in alternative computing paradigms. For now, the story serves as a cautionary tale about the fragility of AI progress when physical constraints outpace technological ambition.