Google's artificial intelligence semiconductor, the Tensor Processing Unit (TPU), is widening its sales reach into the external cloud market, drawing attention to whether it will reshape an Nvidia-centric AI chip market. After Google decided to provide TPUs to Oracle Cloud customers, the possibility has been raised that OpenAI could also expand its use of TPUs, emerging as a factor that could challenge Nvidia's market dominance.

Business Insider reported on Oct. 28 that Google and Oracle announced a contract that will allow Oracle Cloud customers to use Google's TPUs starting next year. The two companies said the deal is the first case of providing TPUs on another hyperscaler's platform beyond Google's own platform.

TPUs are Google's in-house AI chips, developed over more than a decade to boost computing performance for its services, including search. Until now, some assessments said the growth potential of the TPU business largely depended on demand from Anthropic. With the new deal expanding the external customer base, forecasts say Google's TPU business could enter a new growth phase.

Nvidia Chief Executive Jensen Huang (젠슨 황) earlier this year gave a low assessment of Google's TPU growth potential in an interview with podcaster Dwarkesh Patel. He said, "If there is no Anthropic, is there any reason for TPUs to grow?" He added, "It is 100 percent because of Anthropic." Recently, though, more companies have sought to use TPUs, showing a different trend from his remarks at the time.

OpenAI may also move to use TPUs. OpenAI posted a job listing to hire software engineers to optimize its AI models for TPUs. The Information, an IT news outlet, reported that OpenAI plans to buy TPUs from Google.

Several analysts also forecast Google's TPU business would grow rapidly as companies adopt a "multi-chip strategy" that uses multiple types of chips to reduce dependence on a specific semiconductor supplier. This trend could expand opportunities for Google and other competitors to enter Nvidia's core markets.

Still, Nvidia's scale in the current AI semiconductor market remains far ahead of competitors including Google. Google is also a customer that uses Nvidia graphics processing units in large volumes for its AI infrastructure. Even so, the possibility is increasing that TPUs, which Google has developed for a long time, will grow in earnest in external markets.

According to IPO-related filings submitted by Anthropic this week, Anthropic plans to spend $114.5 billion on Google's computing resources from April 2026 to July 2033. It did not disclose the amount directly allocated to TPUs, but it estimated it would spend $125.0 billion to $150.0 billion on TPU compute capacity secured from Google and other providers.

Anthropic also said in the same filing it uses not only chips from Google and Nvidia, but also Amazon's Trainium chips and AMD GPUs. It suggests a clearer trend among major AI companies to use semiconductors from multiple suppliers together.

Some analysis also says the spread of this multi-chip strategy could weaken Nvidia's pricing power. If companies do not insist only on Nvidia products and use chips from Google or other companies together, Nvidia could find it harder to lead pricing.

Forecasts for growth in Google's TPU business are also steep. Japan's financial group Mizuho said in a report released this month it expects Google's TPU revenue to reach $28.0 billion in 2026 and $67.0 billion in 2027. It forecast the figure could rise to as much as $100.0 billion in 2028. TD Cowen analysts projected 2028 TPU revenue at $138.0 billion, or 184 trillion won, which could account for 34 percent of Google Cloud revenue.

Even those estimates still show a wide gap compared with Nvidia's current performance. Nvidia posted revenue of $96.2 billion in the latest quarter alone. The market did not show an immediate reaction to the TPU growth outlook, and Nvidia shares were flat on Wednesday.

Some analysts think the market is underestimating the potential of Google's TPU business. Mizuho analysts said in an Oct. 18 report, "Google's TPU hardware business is emerging as a much bigger opportunity than what is reflected in the current market consensus." They added that if Google sells chips directly to external customers in addition to renting TPUs at its own data centers, it could partly ease the burden of large capital expenditures.

As investment in Google's AI infrastructure is expected to keep rising, Bank of America analysts assessed Google's large-scale investment approach as "relatively efficient" compared with competitors. They cited the fact that TPUs are used not only by external customers but also for Google's own AI services.

For Google, semiconductor technology it has invested in for more than a decade to enhance search functions is increasingly likely to become a new source of revenue. Mizuho also analyzed that, in a situation where AI agents could encroach on existing search services, the TPU business could serve to cushion potential damage to Google's core search business.