Google Now Controls One Quarter of the World's AI Computing Power

A new report from research institute Epoch AI reveals that Google holds roughly 25 percent of the entire world's artificial intelligence computing capacity. The finding puts a number on what many in the tech industry have long suspected: Google's early and aggressive bet on building its own custom chips has given it a commanding lead in the global AI race.

Custom Chips Over Off-the-Shelf GPUs

Unlike competitors such as Microsoft, Meta, and Amazon, Google did not rely mainly on NVIDIA graphics processing units to build its AI infrastructure. Instead, the company developed its own Tensor Processing Units, commonly called TPUs. These are specialized chips designed from the ground up to handle the math-heavy work that powers AI tools and models. A large language model, or LLM, like Google Gemini needs enormous computing power to process and generate text, and TPUs are built specifically for that kind of workload. According to the Epoch AI analysis, Google now operates roughly 3.8 million TPUs alongside 1.3 million GPUs, giving it more total AI compute than any other single organization on the planet.

A Growing Gap Between Tech Giants

The report uses data from the fourth quarter of 2025 and measures computing power by converting each chip type into a standard unit based on a popular NVIDIA chip called the H100. By that measure, global AI compute has passed the equivalent of more than 16.5 million H100 chips. The five largest cloud companies — Google, Amazon, Microsoft, Meta, and Oracle — together control an estimated 71 percent of the world's total AI computing capacity. Google Cloud chief executive Thomas Kurian has said that rising customer demand and revenue growth justify the enormous spending required to maintain this infrastructure at scale.

Google's strategy stands out because building custom chips is expensive and risky. Most companies prefer to buy proven hardware from established suppliers like NVIDIA. But Google started its TPU program years ago, and the payoff is now clear. As global demand for AI compute continues to surge, owning proprietary chip technology gives Google both a cost advantage and greater control over its supply chain, making it less dependent on external chip makers during a time of fierce competition for AI hardware.