Amazon, Microsoft, and Google parent Alphabet face historic capital expenditure scales, with cumulative AI infrastructure spending projections reaching USD 600 billion this current year. Major cloud providers invest tens of billions quarterly to secure compute capacities, making inference cost optimization a critical operational priority for key market players. According to a report by The Information citing an informed source, OpenAI plans to procure AI chips from Google to expand its compute base and reduce its dependence on NVIDIA.
Google manufactures Tensor Processing Units, specialized processors engineered for artificial intelligence training and execution. A potential deal could strengthen commercial ties between AI competitors amid surging demand for advanced processors across the technology sector. OpenAI already utilizes substantial volumes of NVIDIA graphics processing units to develop and operate its models. Nevertheless, according to earlier reports by Reuters, the company began renting Google AI chips in 2025. Direct hardware purchasing would deepen this partnership, provide an alternative compute source, and increase pressure on NVIDIA's dominant margins.
Google is expanding direct sales of its proprietary chips as market demand for AI infrastructure grows. Agreements with clients, including competing AI developer Anthropic, drive increased demand for Google hardware as companies actively seek replacements for widely deployed NVIDIA processors.
Custom silicon and market integration
Although hardware rental delivers immediate operational relief, long-term infrastructure strategies shift toward proprietary semiconductor development. OpenAI is also designing specialized chips in partnership with Broadcom. The company unveiled its first proprietary AI chip design in June and expects to deploy the processor before the end of the year, enabling hardware base diversification across multiple vendors as services scale.
Google manufactures Tensor Processing Units (TPUs), specialized chips designed for artificial intelligence model training and execution. They attract AI companies seeking alternatives to dominant NVIDIA processors.
Neither company has confirmed the rumored transaction. The report and available materials lack financial terms, pricing data, shipment volumes, or timelines for estimating potential savings. Google chip procurement remains unverified, while financial parameters and timeframes remain undefined. Any official agreement will mark another step for OpenAI toward diversifying the hardware powering its AI systems.
For comparison
The push to erode NVIDIA’s dominance forces competitors into uneasy alliances: while vendor exclusivity began crumbling with cloud rentals in October, procuring Google’s TPU silicon outright makes a key adversary a core infrastructure supplier. Commercial pragmatism trumps competitive friction as soon as hardware concentration begins eroding operating margins.
Our stories: OpenAI Joins Google Cloud: Death of Vendor Exclusivity (October 21).