OpenAI to Buy AI Chips from Google: The Compute Race Heats 2026

OpenAI is reportedly turning to Google for AI chips as it seeks to reduce its dependence on Nvidia and secure the massive computing power needed to train and run its models. The move signals a significant shift in the AI hardware landscape, where even fierce competitors may become suppliers.

Why OpenAI Needs More Chips

OpenAI's AI models, such as GPT-4, require enormous computational resources for both training and inference. Training involves feeding vast amounts of data through the model to learn patterns, while inference is the process of running the model to generate responses. Both tasks demand specialized hardware, primarily GPUs from Nvidia.

The data centers OpenAI operates with partners like Oracle, SoftBank, and CoreWeave are packed with Nvidia GPUs. However, the company is building new facilities in Texas, New Mexico, Wisconsin, Ohio, and Michigan that are likely to incorporate chips from OpenAI itself and from Google.

OpenAI's Own Chip: Jalapeño

In late August, OpenAI unveiled Jalapeño, a chip optimized for inference tasks. While this demonstrates OpenAI's ambition to design its own silicon, it doesn't mean the company will stop buying Nvidia GPUs anytime soon. Jalapeño is specifically for inference, so Nvidia's GPUs remain crucial for training.

Nevertheless, OpenAI's push into custom chips is a clear effort to reduce reliance on Nvidia. Nvidia's GPUs are expensive and their availability fluctuates due to manufacturing constraints at TSMC, which produces them. By controlling the design of its own chips, OpenAI could optimize them for its specific workloads and potentially lower costs.

Google's Role as a Chip Supplier

Google has its own AI chips, the Tensor Processing Units (TPUs), which are designed for machine learning workloads. According to reports, OpenAI may purchase TPUs from Google to supplement its computing needs. This would be a notable development, as Google is a major competitor in the AI space with its own models like Gemini.

However, the AI industry is no stranger to strange bedfellows. The compute race is so intense that companies may set aside rivalries to secure the resources they need. Google's TPUs are known for their efficiency in inference and could offer OpenAI an alternative to Nvidia's GPUs.

What This Means for the AI Industry

The potential deal highlights the growing demand for AI chips and the strain on supply chains. Nvidia currently dominates the market for AI accelerators, but alternatives like Google's TPUs and custom chips from OpenAI could reshape the landscape.

For OpenAI, diversifying its chip suppliers is a strategic move to ensure it can continue to scale its AI models. For Google, selling chips to a competitor could be a lucrative business opportunity and a way to monetize its TPU technology. The background to this story is set out in PlayStation 6 Memory Race: Samsung and SK hynix Compete for GDDR7 Supply. A related thread runs through OpenAI's Jalapeño AI Chip: How AI Designed Its Own Silicon. Earlier coverage in Google Sends AI Chips to Space in Project Suncatcher Test shows how the argument developed.

• OpenAI is reportedly buying AI chips from Google to reduce reliance on Nvidia.
• OpenAI's own Jalapeño chip is optimized for inference, not training.
• Nvidia GPUs remain essential for training OpenAI's models.
• Google's TPUs could provide an alternative to Nvidia's GPUs.
• The AI compute race is driving companies to seek diverse chip sources.

Comparison of AI Chips

What to Watch Next

Whether OpenAI officially confirms the purchase of Google TPUs remains to be seen. If the deal goes through, it could set a precedent for other AI companies to diversify their chip suppliers. It also underscores the importance of custom silicon in the AI arms race. A closer look at the same subject appears in Anthropic Pursues Custom AI Chips to Cut Nvidia Reliance. Anyone following the wider picture can start with Microsoft AI Chip Shortage Threatens Data Center Expansion. The same theme surfaces in AI Regulation Debate Heats Up 2028 Democratic Primary Race, which takes a different angle.

Meanwhile, Nvidia is likely to maintain its dominant position, but the emergence of alternatives could lead to more competitive pricing and innovation. The AI compute race is far from over, and the winners will be those who can secure the most efficient and cost-effective hardware. Context on how this played out before sits in South Africa 2036 Olympics Bid: Africa's First Games Race Heats. A companion piece, F1 Title Race Heats Up: Wolff Warns of Norris and Hamilton Threats, approaches the same subject from another angle.