Google has a clear lead over its competitors in sales of artificial intelligence accelerators. how report Google executive Thomas Kurian told a Goldman Sachs investor conference that the company’s TPU sales business is twice that of its nearest hyperscale competitor. according to Report The company’s cloud computing revenue increased 82% year-on-year in the second quarter of fiscal 2026, reaching $24.77 billion. At the same time, Google included the sales of TPU artificial intelligence accelerators into the cloud computing revenue of third-party companies for the first time.
Kurian said that the company provides the best AI infrastructure and offers three types of chips – NVIDIA GPUs, proprietary TPUs and customized Arm processors axion. “We offer 2.7 times the training price, 80% the inference price, 30% the processor price. All of this allows us to differentiate our product portfolio from other vendors. This allows us to provide solutions for financial markets and capital markets.” – Kurian said.
According to Kurian, more than 90% of Fortune 100 companies and thousands of small businesses use the Gemini Enterprise platform. He noted that the company hopes to use it as an inference agent and eventually automate workflows. At the same time, they are interested in the ability to audit AI work, securely control it, and manage AI costs and budgets. “We have all these controls. And we enable people to use the right model to solve the right problem.” said a Google executive.
Image source: Google
Kurian said that Google uses three business models to monetize TPU systems: leasing TPUs in the cloud and selling TPUs directly for deployment in customer data centers (like Anthropic selection) and sell TPU through joint ventures с black stone. Morgan Stanley reported in August: “We raise our TPU sales estimate to $27 billion with a gross margin of 30%. Overall, we expect Google to sell 0.3 GW of TPU in the second half of 2026, 3.2 GW in 2027, and 4.2 GW in 2028. This implies $20 billion and $2 billion in 2027 and PU.
As the resource points out techblog.comsoc.orgOther cloud providers are also developing their own AI-based programs to reduce reliance on NVIDIA GPUs and optimize cost, power consumption and performance based on the workload. Amazon has training courseуMicrosoft— maya. Analysts say Amazon leads Microsoft sell Artificial intelligence chips for external customers. Yuan✴ Implement an AI accelerator miaOracle decided not to develop its own chips, but instead used NVIDIA and AMD accelerators to create large-scale AI clusters.
By mid-2026✴Amazon, Microsoft and OpenAI have closed the gap in three key components of artificial intelligence – specialized chips, computing power and models – and by 2021 only Google has it. Five of the six major artificial intelligence players—Google, Meta✴Amazon, Microsoft, and OpenAI—will now control at least two of the three most important components of artificial intelligence (chips, power supplies, models).
This vertical integration trend is changing the AI infrastructure market, with hyperscale enterprises increasingly using customized chips to optimize cost and performance for specific workloads while maintaining strategic flexibility by sourcing GPUs from multiple vendors.
source: