Key Takeaways
- Google told Meta in March it could not meet full Gemini capacity demands, disrupting Meta's AI projects
- Meta is now pushing staff to use AI tokens more efficiently due to the restrictions
- Google Cloud's compute constraints cost it growth despite hitting $20 billion in Q1 revenue
Google has restricted Meta's access to its Gemini AI models after Meta requested more computing capacity than Google could deliver, according to a Financial Times report. The limits, communicated around March 2025, have delayed some of Meta's internal AI projects and forced the company to ration its AI usage.
The shortage affects other Google Cloud clients too, though Meta felt the impact hardest because of its unusually high demand for Gemini capacity. In response, Meta has instructed employees to be more efficient with AI tokens, the units that measure AI model usage.
Why is Google limiting Gemini access?
The answer is simple: demand outstrips supply. Even trillion-dollar companies cannot build AI infrastructure fast enough. Google Cloud posted $20 billion in Q1 2025 revenue, but CEO Sundar Pichai acknowledged that compute constraints prevented higher growth and nearly doubled the cloud unit's backlog quarter over quarter.
Google is spending $75 billion on capital expenditures in 2025, primarily for AI data centers. Meta has committed to spending $37 billion or more this year on similar infrastructure. Yet both remain capacity-constrained. The GPU shortage that began in 2023 has evolved into a broader compute crunch affecting the entire AI supply chain.
The strange bedfellows of AI competition
Meta and Google compete directly in online advertising, their primary revenue source. They also compete in AI, with Meta's Llama models challenging Google's Gemini. Yet Meta has been exploring partnerships with Google Cloud since at least September 2025, looking to fine-tune Gemini and the open-source Gemma models with its own advertising data.
The Information reported that Meta was also considering working with OpenAI to improve features in Meta AI, its chatbot, and across its social apps. The company has been hedging its bets, building massive internal AI infrastructure while simultaneously shopping for external capacity.
Meta operates one of the world's largest GPU clusters, reportedly exceeding 600,000 H100 chips. That it still needs Google's capacity speaks to the scale of AI compute demands in 2025.
What this means for enterprise AI projects
If Google cannot fully serve Meta, a company with virtually unlimited resources, smaller enterprises should expect their own capacity requests to face scrutiny. Cloud providers are prioritizing workloads, and customers without leverage may find themselves waiting.
The token efficiency push at Meta offers a preview of what other organizations will face. Companies that optimize their AI workflows now, reducing unnecessary model calls, caching responses, and choosing appropriately sized models for each task, will fare better in a supply-constrained environment.
Logicity's Take
This dispute reveals that even Big Tech cannot buy its way out of the AI compute shortage. For enterprise buyers evaluating cloud AI services, the lesson is clear: capacity guarantees matter as much as pricing. Google Cloud, AWS Bedrock, and Azure AI all offer Gemini or competing models, but their SLAs differ significantly. Organizations running production AI workloads should negotiate committed capacity pools rather than relying on on-demand access. The alternative, as Meta discovered, is watching projects stall when the provider cannot deliver.
Will the shortage ease in 2026?
Probably not. NVIDIA's next-generation Blackwell chips are ramping production, but demand continues to outpace supply. New data centers take 18 to 24 months to build. The hyperscalers are all expanding, but so is the appetite for AI compute.
Google's backlog nearly doubling in a single quarter suggests the gap is widening, not closing. Companies planning major AI initiatives in 2026 should secure capacity commitments now rather than assume availability later.
Frequently Asked Questions
Why did Google limit Meta's Gemini AI access?
Google could not meet Meta's full capacity request due to compute constraints affecting its cloud infrastructure. The company has a growing backlog of AI demand it cannot immediately serve.
How is Meta responding to the Gemini restrictions?
Meta has asked employees to use AI tokens more efficiently. The company is also exploring partnerships with Google, OpenAI, and others to supplement its internal AI infrastructure.
Are other Google Cloud customers affected?
Yes, several other clients face similar constraints, though to a lesser extent than Meta due to Meta's exceptionally high demand.
Why would Meta use Google's AI when they compete?
Meta was exploring Gemini to improve advertising effectiveness and power features in Meta AI. Despite competing, large tech companies often use each other's infrastructure where it offers advantages.
Samsung's massive investment includes chip manufacturing capacity that could eventually ease AI compute constraints
Need Help Implementing This?
If your organization is navigating AI cloud capacity planning or evaluating multi-provider strategies, reach out to the Logicity team for guidance on enterprise AI infrastructure decisions.
Source: mint
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
