Google may hold the largest AI computing capacity in the world, according to an analysis by AI research firm Epoch AI. The reason is not just access to Nvidia hardware, but Google's own custom-built Tensor Processing Units, known as TPUs.
The finding highlights a central split in the AI infrastructure race: one company may have the biggest internal compute fleet, while another still dominates the chip market that supplies much of the industry.
Google's edge comes from TPUs
Epoch AI's analysis says Google likely has AI computing power equivalent to at least 600,000 Nvidia H100 GPUs through its TPU fleet. That figure matters because TPUs are not a side project in this comparison. They are the main reason Google appears to stand apart from other companies on total AI compute capacity.
The researchers also considered Google's Nvidia GPUs alongside its custom chips. Their conclusion was that this combined infrastructure probably places Google ahead of any other single company in AI compute capacity.
"Given this large TPU fleet, combined with their NVIDIA GPUs, Google probably has the most AI compute capacity of any single company," Epoch AI researchers noted.
That does not mean every part of the AI chip market is moving toward Google. It means Google's long-running investment in custom processors may have given it a different kind of advantage: the ability to build large internal capacity without relying only on the same commercial GPU supply chain as many rivals.
Nvidia still dominates AI chip sales
Even if Google leads in total capacity, Nvidia remains the central supplier in the broader AI chip market. Since early 2022, Nvidia has sold AI chips with computing power equal to about 3 million H100 GPUs, according to the source article's summary of Epoch AI's analysis.
Most of that capacity went to four major cloud providers, with Microsoft identified as the largest customer. This helps explain why Nvidia's position remains so strong: its chips are not concentrated in one internal fleet, but distributed across major infrastructure buyers.
The buyer list extends beyond the biggest cloud providers. The source article names Oracle and CoreWeave among cloud companies, xAI and Tesla among AI firms, Chinese tech companies, and governments building national AI infrastructure.
That spread shows how AI compute has become a resource for several different groups at once:
- Cloud providers that rent infrastructure to customers.
- AI companies training and running large models.
- Technology companies building their own systems.
- Governments investing in national AI infrastructure.
Blackwell demand adds more pressure
Nvidia's next generation, Blackwell, is also described as being in heavy demand. According to Barron's, citing conversations between Morgan Stanley analysts and Nvidia CEO Jensen Huang, the next 12 months of Blackwell supply is already sold out.
That detail points to a market where supply remains a major constraint. If companies are already lined up for future Nvidia GPUs, access to AI chips is not just a technical matter. It is also a planning issue for cloud providers, AI labs, and other large buyers that need enough compute to support their work.
The reported Blackwell demand also fits with the broader picture: even as Google builds strength through TPUs, Nvidia continues to shape the pace of AI infrastructure expansion through its commercial GPU roadmap.
The chip race is getting wider
The source article also notes that the field is not limited to Google and Nvidia. AMD, Intel, Huawei, Amazon, Meta, OpenAI, and Microsoft are developing their own AI chips.
That list matters because it suggests the AI compute race is becoming more diversified. Companies that buy chips may also want more control over cost, supply, and architecture by building processors of their own.
China is also reportedly encouraging domestic firms to buy more chips from Chinese suppliers like Huawei. That adds another layer to the market: AI chip decisions can reflect supply needs, corporate strategy, and national infrastructure priorities at the same time.
Epoch AI cautions that its estimates have significant uncertainty. The AI chip landscape is changing quickly as companies design new processors and increase production. For now, the clearest takeaway is that Google's TPU fleet may give it the largest single-company AI compute position, while Nvidia remains the dominant force in AI chip sales.