Omdia: Google Cloud's TPU has an advantage in competing with NVIDIA's AI chips
Google Cloud's TPU has a competitive advantage in the competition with NVIDIA's AI chips, leading among major cloud providers. Omdia found that Google Cloud's TPU outperforms competitors such as Microsoft Azure, Amazon Web Services, and Meta Platforms. Google Cloud's TPU may be deployed to the company's data centers to support internal projects and customer workloads. Large-scale enterprises all have custom AI acceleration chips, but the details of commercial success are unknown. Omdia believes that the advantage of Google Cloud's TPU may explain the profitability of the Google Cloud platform. Meanwhile, the ecosystem of semi-custom chips is also evolving, supporting the industry's trend towards custom chips
According to the Zhītōng Finance and Economics APP, Omdia's latest research report "Investigation of Artificial Intelligence Chips for Super-Large Enterprises: Spring 2024" found that in the competition for artificial intelligence hardware with NVIDIA (NVDA.US), Google (GOOGL.US) is in a clear leading position among super-large cloud providers. In 2024, Google Cloud's TPU worth $6 billion may be shipped to the company's data centers to support internal projects such as Gemini, Gemma, and Search, and to provide support for customer workloads through the Google Cloud platform.
Currently, all three major super-large enterprises have custom artificial intelligence acceleration chips, but the details of their commercial success are often unknown. However, these super-large enterprises have all used at least one company specializing in semi-custom chip projects, such as Broadcom, Marvell, Alchip, or Arm plc's Neoverse CSS service. By carefully studying the financial reports and public statements of these companies, these customers can be identified and linked to the revenue figures of these outsourcing partners.
Therefore, Omdia found that Google Cloud TPU outperforms competitors such as Microsoft Azure, Amazon Web Services, or Meta Platforms. Alexander Harrowell, Senior Computing Chief Analyst at Omdia, said: "This may explain how the Google Cloud platform itself has recently become profitable. At the same time, the semi-custom chip ecosystem itself is constantly evolving and deepening its products, supporting the industry-wide trend towards custom chips."
However, one thing remains unresolved, which is the identity of "Customer C." Omdia stated that "Customer C" is a cloud computing company based in the United States, and its artificial intelligence chip will be mass-produced in 2026, but it is not one of the three tech giants. Harrowell explained: "For example, Broadcom claims that from signing, it can complete an artificial intelligence chip project within a year, so the long preparation time means it is a qualitative innovation."