Google has restricted Meta’s access to its Gemini AI models after the social media giant requested more computing capacity than Google could supply, the Financial Times reported on June 28, 2026. The move has disrupted and delayed multiple internal Meta AI projects and signals a deepening global crisis in artificial intelligence infrastructure that is now affecting even the largest players in the industry.
What Was Announced
According to the Financial Times and subsequent reports from CNBC and other outlets, Google informed Meta around March 2026 that it was unable to fulfill the full volume of Gemini AI computing capacity Meta had sought to purchase. Meta, which had become one of Google’s largest enterprise Gemini customers, found its AI operations constrained as a result.
The fallout was immediate for Meta’s internal teams. The company instructed employees to use AI tokens more sparingly and to improve efficiency in how they consume computing resources. Meta has also begun shifting internal workloads from Google’s Gemini to its own internally developed Muse Spark model, a move that signals a strategic pivot toward reducing dependency on external AI providers.
The situation extends beyond Meta. Several other Google Cloud customers have reportedly been affected by compute constraints, though to a lesser extent than Meta. Google declined to comment on the specifics of any individual customer relationship, but the scope of the shortage is reflected in the company’s own financial disclosures and executive commentary.
Google Cloud posted more than $20 billion in quarterly revenue, a year-over-year increase of 63 percent. Despite that staggering growth, the company faces an estimated $460 billion in unmet infrastructure demand. Google CEO Sundar Pichai publicly acknowledged the challenge, stating: “We are compute-constrained in the near term.”
Technical Details
The core bottleneck is GPU supply. Training and serving large AI models requires massive quantities of specialized hardware, primarily NVIDIA GPUs, which remain in critically short supply across the industry. Google has committed $180 to $190 billion toward AI infrastructure investment in 2026, a figure that reflects the scale of the problem rather than a solution to it.
To bridge the gap between existing capacity and skyrocketing customer demand, Google has entered into an extraordinary arrangement with SpaceX, paying approximately $920 million per month for access to 110,000 NVIDIA GPUs. Google describes this as “bridge capacity,” a temporary measure to supplement its own data center buildout while new facilities come online. The SpaceX deal alone represents an annualized spend of roughly $11 billion on externally sourced compute.
For Meta specifically, the compute squeeze arrived at a difficult moment. The company has simultaneously been undergoing significant internal restructuring, including a reduction of approximately 8,000 positions, while also planning to invest up to $135 billion in its own AI infrastructure. Meta’s reliance on Google’s Gemini API for internal tooling made the compute limits particularly disruptive to engineering workflows that had been built around consistent access to that capacity.
Industry Impact and Reactions
The Google and Meta situation is being closely watched across the AI industry as a concrete example of the infrastructure constraints that have until recently been discussed in mostly theoretical terms. For months, analysts and executives have warned that demand for AI compute would outstrip supply. This episode confirms that the gap has become wide enough to affect major commercial relationships between two of the largest technology companies on the planet.
The competitive implications are significant. Meta’s accelerated investment in its own Muse Spark model and internal compute suggests that large-scale AI consumers are drawing lessons from this episode and moving toward greater self-sufficiency. Other hyperscalers and enterprise AI adopters who rely on third-party API access for critical workflows may now reconsider their dependence on any single compute provider.
For Google, the situation presents a paradox: its Gemini models are generating intense commercial demand, yet infrastructure limits are forcing the company to ration access to paying customers. While Google Cloud’s revenue growth is exceptional, the ability to translate that demand into revenue is constrained by hardware availability. Competitors including Microsoft Azure, AWS, and Oracle Cloud are facing similar pressures, though each has structured its infrastructure investments differently.
What Comes Next
Google has provided no specific public timeline for when compute capacity constraints will ease. The company’s bridge arrangement with SpaceX is expected to persist into late 2026 at minimum, as new Google-owned data center capacity requires 18 to 24 months from groundbreaking to full operation. The $180 to $190 billion infrastructure commitment suggests that Google is building toward a significant expansion of capacity, but the benefits of that investment are unlikely to reach enterprise customers in the near term.
Meta, for its part, has signaled that its long-term strategy involves far greater self-reliance on internally developed models and owned infrastructure. The Muse Spark transition and the planned $135 billion infrastructure investment are likely to reduce the company’s exposure to third-party compute rationing going forward. Whether Google can retain Meta as a major customer once its own capacity is online will be one of the more consequential enterprise AI business storylines of the next 12 months.
Conclusion
The restriction of Meta’s Gemini AI access is a milestone moment in the evolution of the AI industry, marking the first widely reported instance of a major provider rationing compute to a major customer due to infrastructure scarcity. As demand for AI services continues to accelerate faster than new data center capacity can be built, the industry should expect rationing, strategic pivots toward internal models, and intensified competition for GPU supply to become defining features of the AI landscape through 2026 and beyond.
Stay updated on the latest AI news at Evolve Digital.
Leave a Reply