Google's AI Infrastructure Strained by Surging Demand for Gemini

Google’s artificial intelligence (AI) infrastructure is currently experiencing significant strain due to a surge in demand for its latest AI models. This challenge has led to limited availability for users accessing some of the company’s most advanced AI capabilities.

Logan Kilpatrick, a Google Product Manager, recently addressed user complaints regarding the restricted access to Gemini 2.5 Pro Deep Think. Kilpatrick explained that the model’s release is constrained because it requires “a boat load of compute to run.” This substantial processing demand comes at a time when Google’s custom-designed Tensor Processing Units (TPUs), essential for AI workloads, are already operating at peak capacity.

According to Kilpatrick, the TPUs are struggling to keep up with “massive growth” driven by several factors. These include the widespread adoption of Google’s new video generation model, Veo, increased usage of the standard Gemini 2.5 Pro, and the broader rollout of AI features to hundreds of millions of users.

User frustrations stem from the fact that, despite Gemini 2.5 Pro Deep Think’s strong benchmark performance, its practical use is hampered by access restrictions. Even subscribers to premium tiers, such as Gemini Ultra, report being limited to only a handful of requests per day as Google’s systems contend with the escalating computational demands.