Google accelerates TPU deployment, escalating competition with Nvidia in the AI chip sector.

```

Google is one of NVIDIA's largest buyers of AI chips and rents these chips to Google Cloud customers such as OpenAI and Meta Platforms. However, Google's ambition to develop its own AI chips has not slowed down.

According to seven people involved in the relevant negotiations who spoke to the media, Google has recently approached some small cloud service providers that mainly rent NVIDIA chips, proposing that their data centers also deploy Google's AI chips.

Company representatives involved in the deal told the media privately that Google has already reached an agreement with at least one cloud service provider, including London-based Fluidstack, which will deploy Google’s Tensor Processing Units (TPUs) in its New York data center.

In addition, Google has also tried to reach similar agreements with other cloud service providers focused on NVIDIA chips, such as Crusoe, which is building a data center for OpenAI to deploy a large number of NVIDIA chips, and CoreWeave, which rents NVIDIA chips to Microsoft and OpenAI.

The media reported that it is currently unclear why Google has chosen, for the first time, to deploy TPUs in other cloud service providers' data centers. Analysts believe this may be because the speed of Google’s own data center construction cannot keep up with the growing demand for chips, or perhaps Google hopes to find more new customers for its TPUs through other cloud service providers, such as AI app developers. This approach is similar to the model of cloud service providers renting out NVIDIA GPUs.

Analysts say that if it is the latter case, Google's approach would mean it is competing more directly with NVIDIA, as NVIDIA mainly sells chips to these cloud service providers. No matter the reason, deploying TPUs in other cloud service providers’ data centers will mean a reduction in the number of NVIDIA GPUs used in these facilities.

The equity research analyst team at investment firm D.A. Davidson, led by Gil Luria, told the media that more and more cloud service providers and large AI developers are interested in TPUs and hope to use them to reduce their dependence on NVIDIA. After communicating with researchers and engineers from several cutting-edge AI labs, they found that Google’s AI- and machine learning-accelerated chip receives positive reviews in the industry.

Therefore, the analyst team believes that if Google merges its TPU business with its AI research institute DeepMind and spins them off for a separate listing, there will be strong market demand. According to the Luria team’s estimates, the potential valuation of this business is about $900 billion. Earlier this year, their valuation was $717 billion.

“No one wants to have only one source... to be completely controlled by a single supplier for key components.”

“If this business is really spun off, investors will get both a leading AI accelerator chip supplier and a top AI lab, which may become one of Alphabet’s most valuable assets.”

However, NVIDIA CEO Jensen Huang downplayed such competitive chip projects. He told the media that AI app developers prefer GPUs because they are more versatile and have stronger software support.

Winning Over NVIDIA’s “Friends”

Media reports indicate that Google's negotiations show it is seeking to approach emerging cloud service providers that NVIDIA is actively supporting. Unlike Google Cloud and Amazon Web Services, these companies almost exclusively rent NVIDIA chips and are more willing than traditional cloud providers to purchase a variety of NVIDIA products. NVIDIA has also invested in these companies and given them priority supply for the most popular chips.

Google mainly uses TPUs to develop its own AI models, such as the Gemini series, and in recent years internal demand for TPUs has soared.

However, Google has also long rented TPUs to other companies. For example, Apple and Midjourney both rent TPUs through Google Cloud. Earlier this summer, Google even attracted OpenAI’s interest in renting TPUs for a time, but OpenAI suddenly changed its mind.

Internally, Google has discussed expanding its TPU business to increase revenue and reduce its cloud computing division's dependence on expensive NVIDIA chips. According to two former executives, higher management has also considered selling TPUs directly to customers outside Google Cloud.

Analysts believe that small cloud service providers like CoreWeave and Fluidstack (for example, Fluidstack provides NVIDIA GPUs for startups like Mistral) have a strong commercial incentive to prioritize servers with NVIDIA chips, since AI developers generally prefer NVIDIA products.

But Google seems to have found a way to encourage Fluidstack to support its TPU expansion plan: if Fluidstack cannot afford the leasing costs for its soon-to-open New York data center, Google will provide up to $3.2 billion in backing. This commitment helps Fluidstack and its data center partners raise debt financing to build facilities.

TPU Demand Is Rising

The media report that since Google opened its sixth-generation Trillium TPU chips to external customers in December of last year, demand has been strong. Analysts expect demand for the seventh-generation Ironwood TPU to “rise significantly.” Ironwood is Google’s first chip designed specifically for large-scale AI inference tasks (post-training model deployment and operation).

Analysts note that Google’s TPU chips can reach a computing power of up to 42.5 exaflops, and have greatly increased high-bandwidth memory capacity. These chips are also “significantly more cost-efficient,” which is one of the main reasons attracting the attention of cutting-edge research labs.

The startup Anthropic previously used TPUs on a small scale, but analysts note the company is now hiring TPU kernel engineers, which may mean they are considering switching from Amazon Web Services’ Trainium chips to TPUs. Trainium is Amazon's chip for AI training; Amazon has invested $8 billion in Anthropic.

Analysts also note that Elon Musk’s xAI is showing interest in purchasing TPUs, in part due to the “significant advances in JAX-TPU tool support” this year. JAX is a high-performance computing Python library developed by Google that allows programs to run efficiently on TPUs. Until recently, the JAX ecosystem had limited the possibility of large-scale TPU deployment outside Google.

According to D.A. Davidson's DaVinci developer dataset, developer activity related to TPUs on Google Cloud grew by about 96% in the six months from February to August 2025.

Risk Warning and DisclaimerThe market carries risks, and investment should be made cautiously. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment decisions based on this article are at your own risk. ```