Key Points

• It is reported that OpenAI plans to purchase Google's TPU processors.
• NVIDIA faces increasing competition in customizing AI chips
• Infrastructure spending of large tech companies could approach $600 billion

It is reported that OpenAI plans to purchase artificial intelligence chips directly from Google, which could grant one of NVIDIA's largest competitors an important customer. Meanwhile, the technology company is preparing to invest tens of billions of dollars in the infrastructure of AI.

The Information cited an informed source on October 27th stating that OpenAI intends to purchase Google's tensor processing units (TPU). The financial terms, procurement quantity, and delivery schedule have not yet been disclosed.

This potential transaction will deepen an unusual relationship between these two AI competitors. GPU is highly dependent on NVIDIA, while Google is increasingly providing its own processors to external customers seeking alternatives.

At the time of this report, Amazon, Microsoft, and Alphabet plan to spend a total of up to $600 billion this year on infrastructure in AI. This has also raised questions in the market about which chip manufacturers will receive this investment.

Google may acquire one of NVIDIA's largest AI customers

OpenAI already has experience in using Google's infrastructure.

Reuters reported in June 2025 that ChatGPT developers have begun to rent Google's AI chips, and OpenAI recently also announced plans to distribute its advanced models through Google Cloud.

Purchasing processors directly would mean a different business arrangement, which could give OpenAI greater control over the way its computing infrastructure is deployed.

According to The Information, the company is also recruiting an engineer to optimize the operation of cutting-edge models on TPU. However, this recruitment announcement does not prove that the hardware procurement has been completed.

Google's chips are designed specifically for high-intensity machine learning computations, providing an alternative to NVIDIA's more general-purpose and programmable GPU.

The comparison surrounding the customized AI chip also explains why dedicated processors can offer efficiency advantages in specific workloads, but they may not necessarily replace GPU in all scenarios.

Google's continuously expanding supplier network also includes Marvell; if the relevant conditions are met, its agreement with TPU could result in cumulative sales of up to $120 billion by 2033.

Oracle joins Google, TPU; NVIDIA's competition further intensifies

Time points are particularly important, as Oracle has announced plans to provide Google TPU infrastructure through its cloud platform.

At the same time, Oracle has also expanded its partnership with NVIDIA, indicating that large cloud service providers are adopting a variety of chip technologies rather than relying solely on a single supplier.

OpenAI is also adopting similar strategies.

In June this year, the company announced its first custom AI processor developed in collaboration with Broadcom, with plans to deploy it by the end of 2026.

This progress comes shortly after the market's interest in Broadcom inference chips has risen; such chips are designed to handle the massive computational demands required for running AI models.

Meanwhile, Anthropic has already achieved a diversified layout among the infrastructures of NVIDIA, Google, and Amazon, and the financing of Broadcom may also support an additional several billion dollars in TPU production capacity.

Competition is not just about processor performance. Customers also have to consider software compatibility, usability, power consumption, and total operating costs.

$600 billion AI spending boom raises the stakes for NVIDIA

This potential transformation occurred amidst a round of exceptionally large investment commitments.

According to the recent AI expenditure analysis by Coinpaper, Amazon, Microsoft, and Alphabet may collectively spend nearly $600 billion on AI infrastructure in 2026.

NVIDIA is still deeply tied to this round of expansion.

Its strategic relationship with OpenAI includes plans to deploy at least 10 gigawatts of NVIDIA systems, as well as a previously announced investment framework totaling up to $100 billion.

However, the continuously expanding supplier relationships of OpenAI indicate that future demand may be dispersed across more companies.

As the cost of infrastructure accelerates, the question of whether AI's massive investments can generate sufficient returns has also become more important.