OpenAI reportedly plans purchases of Google AI chips

By Sophie Lindqvist

Reuters, citing The Information, reported on October 28, 2026, that OpenAI is “planning to buy Google’s AI chips.” The plan has no disclosed order value, unit count, chip generation or delivery date.

The two-source report points to two stated pressures: rapid expansion of autonomous systems and rising demand for AI computing. It does not quantify either. No capacity, revenue, shipment, pricing or analyst-forecast figure accompanies the procurement claim.

That evidentiary limit matters. Reuters attributes the story to The Information, while the source presents no comment from OpenAI or Google. The wording “planning to buy” establishes intent, not a completed transaction.

What exactly did the report establish?

Reuters relayed The Information’s account that OpenAI plans to purchase Google’s AI chips. The source offers no signed agreement, purchase order, named product, delivery schedule or executive statement. It also does not define the transaction. Direct hardware purchases, cloud capacity commitments and multiyear supply contracts would have different commercial consequences.

No delivery schedule appears. That omission leaves OpenAI’s near-term compute position uncertain. A chip announcement without timing does not prove that additional machines will enter service this year or next.

The report further gives no spending comparison. It does not state how much OpenAI currently spends on accelerated computing, what share Google would supply, or whether the goal is lower unit cost, faster deployment, supplier diversification or access to capacity already built by Google.

Which Google chips are under discussion?

Google’s Tensor Processing Units, or TPUs, are the relevant product family. Reuters described TPUs as a globally recognized alternative to Nvidia’s AI chips. A confirmed purchase would therefore place Google-designed accelerators inside OpenAI’s infrastructure discussions.

The source does not identify a specific TPU generation. It lists no process node, wafer size, accelerator configuration, memory capacity, packaging, interconnect or power envelope. It also provides no benchmark comparing Google’s hardware with Nvidia’s.

That lack of specification prevents responsible claims about performance, energy efficiency or cost per unit of AI work. Such conclusions would require comparable model workloads and system-level pricing, neither of which the source supplies.

Supply-chain context remains equally thin. The report does not say whether Google has reserved production capacity for OpenAI, which manufacturing sites would serve the order, or whether delivery would depend on new fabrication and packaging. It offers no lead-time estimate.

Without a volume and schedule, the report cannot establish whether the purchase would materially change OpenAI’s training or inference capacity. It also does not quantify any effect on Nvidia demand. The commercial signal is an intention to add a second named accelerator source, not a measured supply shift.

Why does Google Cloud matter?

OpenAI has separately announced that its frontier AI models, including GPT-6 Astra, will be available through Google Cloud. That confirms a distribution path, but it is not a hardware-capacity disclosure.

The source does not say that Google Cloud has allocated TPUs to those models. It also does not connect the reported chip purchase to the model announcement. Model access through a cloud platform and direct accelerator procurement may support different commercial strategies, and the available account does not merge them.

The distinction limits the immediate pricing implication. There is no quoted cloud price, reserved-instance commitment or discount attached to GPT-6 Astra availability. Buyers therefore cannot use the announcement alone to infer OpenAI’s unit economics.

How does the custom chip fit?

The source headline also states that OpenAI unveiled its first custom AI chip in June. It provides no accompanying specifications. The chip’s foundry, process node, design partner, packaging, expected volume and deployment role all remain undisclosed.

Its timing suggests OpenAI is pursuing more than one route to compute, but that is an inference, not a stated procurement strategy. The source never says whether the custom silicon would complement Google TPUs, replace some external accelerators or remain outside data-center deployment.

No public figure links the three developments into a capacity plan. OpenAI’s June chip announcement, the reported Google purchase and the Google Cloud distribution decision appear as separate items. Treating them as one confirmed supply chain would assign relationships the source does not establish.

What would make the plan material?

A firm order would require at least a named chip family, unit volume, delivery window, pricing structure and deployment location. Confirmation of reserved Google Cloud capacity would provide a different answer. Neither appears here.

The report contains no confirmed capacity increase, revenue contribution, shipment count or manufacturing roadmap. Its central claim remains a reported plan, backed by a named publication but lacking company confirmation and commercial terms.

The defensible conclusion is limited. OpenAI reportedly plans to buy Google AI chips amid rising AI-computing demand and autonomous-system expansion. GPT-6 Astra will be available through Google Cloud. Whether either move changes capacity, pricing or competition with Nvidia depends on disclosures that neither OpenAI nor Google has provided.