Epoch AI recently estimated Google holding ~1/4 of global cumulative capacity as of Q4 2025. Google has the equivalent of 5+ million H100 GPUs in compute capacity. Nearly 4 million of that comes from custom TPU chips. Microsoft is second with the equivalent of 3.5 million H100.

When others depend heavily on Nvidia for AI infrastructure, Google built and scaled an alternative.

Thomas Kurian, Google Cloud CEO, explained:

“We’re not just a hyperscaler reselling other people’s technology. Our differentiation comes down to the fact that we own the IP - the model and the chips are ours….

For every dollar of revenue, we’re not shipping 80% of it to either a model or chip provider, which allows us to invest more.” (FT)

The full stack matters even more as the industry shifts to compute-hungry agentic AI.

At Cloud Next last week, Google unveiled two 8th-generation TPUs - one for training, the other for inference and agentic workloads. Those chips power Gemini models. Gemini powers the new Agent Platform. The Agent Platform connects to a new Agent Marketplace.

Compute → models → data layer → agent platform → marketplace. Each layer owned internally, including distribution.

Google is playing a platform-company model

It competes with Anthropic on models - while powering Anthropic’s training on TPUs. Google just agreed to additionally invest up to $40B in Anthropic and provide it with more computing capacity.

It competes with Apple via Android - while supplying Gemini to power the next generation of Siri.

Kurian: “Google’s a platform company. Different parts of your business compete with different players in the market.”

Whether Google wins on models or infrastructure, it generates demand across the stack.

Partnerships sit at the center of how Google plans to scale this

At Next, Google announced a $750M partner fund for agentic AI development. Google wants partners building on Gemini, extending Gemini Enterprise, and helping customers to move agentic AI from pilots into production.

The commercial results already back up the thesis.

Google Cloud is at a $70B+ annual run rate, growing 48% year over year, with $240B backlog. FT puts its market share at 14%, up from 7% when Kurian joined.

For alliance leaders:

• Google may still trail other hyperscalers overall, but AI creates a new decision layer.

• The early data - growth, backlog, partner commitments - says the market is responding.

• The Agent Marketplace is a new distribution category now with multiple surfaces. If your marketplace strategy only expects to include traditional listings, it may need a refresh.

What’s your current read on Google Cloud’s momentum?

Source: graph