According to The Information, OpenAI is preparing to acquire Google Tensor Processing Units (TPUs) — computing accelerators that Google's parent company, Alphabet, began selling directly to select buyers this year. The decision comes amid a surge in technology corporation spending driven by increased demand for AI computing. The financial terms of the future deal, as well as its exact volumes, were not disclosed at the time of publication, which is typical for the early stages of negotiations in the AI infrastructure sector.
The Race for Compute and Investor Questions
The context for the deal is set by macro industry indicators: this year, Amazon, Microsoft, and Alphabet intend to spend a combined total of about $600 billion on AI infrastructure. Such unprecedented capital expenditures are increasingly raising questions among investors, who are looking for answers as to when large-scale spending will begin to yield measurable financial returns. For Alphabet, which will soon release its quarterly earnings, the market will monitor particularly closely whether active investments in AI are driving growth in its cloud business.
From Competition to Partnership: OpenAI and Google
Implementing the plan to purchase TPUs will strengthen ties between OpenAI and Google — companies that compete in AI products and search markets while working increasingly closely on joint commercial projects. Last week, OpenAI announced it will provide access to its advanced AI models via the Google Cloud platform. Reports indicate that as early as June 2025, it was reported that OpenAI had already begun leasing Google computing power, although detailed information about the types of chips used was absent at the time. The company is already among the largest buyers of NVIDIA AI accelerators, which are used to train and operate its models.
TPUs as an NVIDIA Alternative and Circular Financial Schemes
Google’s Tensor Processing Units have gained traction as an alternative to NVIDIA accelerators as many AI market participants look for less expensive solutions. According to The Information, information recently emerged that Oracle also intends to buy Google TPUs. OpenAI is not the only one relying on complex financial structures to obtain compute resources: according to Bloomberg, as reported by 3dnews, Apollo Global Management and Blackstone Inc. intend to attract additional investors to channel $36 billion in borrowed funds to buy Google TPUs for rival Anthropic. Under this arrangement, the chips do not become the startup's property but are leased, with Broadcom ready to provide a significant portion of the resources; Anthropic's management openly acknowledges a lack of computing power as demand outstrips the company's capabilities.
OpenAI's Own Chips and Alphabet's Revenue Horizons
Simultaneously, OpenAI is pursuing its own semiconductor project: in June, the company unveiled a chip developed jointly with Broadcom and reportedly planned to begin rolling it out by the end of this year as part of a multi-generation custom semiconductor program. Thus, purchasing Google TPUs does not cancel but rather complements the compute base diversification strategy. For Alphabet, client deals including Anthropic have fueled demand for TPUs: the company expects to recognize a small portion of revenue from agreements to sell these processors by the end of this year, while the bulk of the revenue is expected to be recognized in 2027.
