Following Microsoft and Amazon in reducing reliance on NVIDIA, Google launches data center chip

Just like its old rivals Microsoft and Amazon in the cloud computing field, Google has also launched its own chips to reduce its reliance on NVIDIA chips. On Tuesday, April 9th, Eastern Time, Google announced at this year's annual cloud computing conference Cloud Next 2024 the launch of a data center chip based on Arm architecture, named Axion. Google plans to offer this CPU through Google Cloud, claiming that its performance exceeds x86 architecture chips and general Arm architecture chips running in the cloud. The performance of the Axion chip is 30% higher than general Arm chips and 50% higher than the current generation x86 chips produced by Intel. Axion is used for various Google services, such as YouTube ads in Google Cloud. The company plans to expand such uses and make it available to the public "later this year." Axion is suitable for a range of tasks, including supporting Google's search engine and artificial intelligence (AI) related work. Google officials stated that the chip can play an important supporting role in the field of AI by helping process large amounts of data and deploying services to billions of users. Google previously used Arm-based servers internally for running YouTube ads, BigTable and Spanner databases, as well as the BigQuery data analysis tool. A Google spokesperson mentioned that when Axion is available, the company will gradually transition them to cloud-based Arm instances. Companies like Datadog, Elastic, OpenX, and Snap are planning to adopt Axion. Some comments suggest that the introduction of the new chip means Google is catching up with cloud competitors like Amazon and Microsoft. Amazon and Microsoft had already introduced Arm architecture CPUs in 2021 as a way to provide differentiated computing services. Google had previously launched custom chips for YouTube, AI, and its smartphones, but had not built CPUs before. Other comments suggest that Google's development of new chips is to reduce reliance on external vendors amidst the intense competition in AI. However, Google officials do not see the new chip as a competitive move. Amin Vahdat, Google's Vice President in charge of self-developed chip business, said, "I think this is the foundation for making a bigger cake." Google's Tensor Processing Unit (TPU) is one of the few alternatives to NVIDIA's advanced AI chips, which developers can only use through Google's cloud platform and cannot purchase directly. Broadcom was Google's partner in producing the previous generations of TPU chips. Google did not comment on whether there are partners in designing Axion and whether Broadcom is involved in the launch of Google Cloud's fifth-generation TPU, TPU v5p. Data shows that while advertising is still Google's largest source of revenue, cloud computing is growing faster and accounts for an increasing share of Google's revenue, approaching 11% of the total company revenue. Gartner estimates that in 2022, Google holds a 7.5% share of the cloud infrastructure market, while Amazon and Microsoft collectively control around 62% of the market share Google has become the third tech giant, following Microsoft and Amazon, to introduce data center CPUs based on Arm architecture. This indicates a new trend, as previously, major enterprises operating servers mostly purchased CPUs from Intel and AMD