OpenAI and Anthropic Orders Give Google's TPU Chip Business Visibility Through 2029, Analyst Says

TF International analyst Ming-Chi Kuo says strong new orders from OpenAI and Anthropic have extended Google's TPU shipment visibility to 2029, with the two AI labs taking about 26 million chips between them in 2028 and 2029.

Nov 17, 2026 - 22:23
 0  6
OpenAI and Anthropic Orders Give Google's TPU Chip Business Visibility Through 2029, Analyst Says
A long aisle in a Google data centre lined with tall racks of TPU server hardware.

OpenAI and Anthropic Orders Give Google's TPU Chip Business Visibility Through 2029, Analyst Says

Google's custom AI chip business has gained a much longer runway, according to one of the best-known supply chain analysts in the industry. TF International analyst Ming-Chi Kuo said in a note published on November 16 that strong new orders from OpenAI and Anthropic have extended shipment visibility for Google's Tensor Processing Units, or TPUs, through 2029, a horizon he says now runs further than that of Nvidia's GPUs and rival AI accelerators.

The two AI labs are the story. Kuo estimates that OpenAI and Anthropic will take about 26 million Google TPUs between them across 2028 and 2029, with demand from both growing year by year. OpenAI accounts for about 12 million of those chips, while Anthropic accounts for about 14 million, according to his latest industry survey.

OpenAI buys outright, Anthropic mixes leasing and purchases

The two customers are not buying the same chips, or in the same way. Kuo puts OpenAI's demand at about 12 million TPU 8t chips, the training-focused design co-developed with MediaTek, bought outright. Anthropic's roughly 14 million chips are mainly TPU 8i, the inference-focused design co-developed with Broadcom, taken through a mix of leasing and outright purchases.

The distinction matters because of what Google itself plans to run. From 2028, Kuo says, the next-generation v9 will become Google's main solution for its own in-house computing, which means orders for the v8 series can be viewed as coming from OpenAI and Anthropic rather than from Google's internal buildout.

Put against Kuo's shipment estimates, the two labs dominate the order book. He estimates total Google TPU shipments of 8 to 9 million units in 2027, 14 to 15 million in 2028 and 16 to 18 million in 2029, with supply chain capacity constraining how much of the demand can actually be delivered. On those figures, OpenAI and Anthropic together would account for roughly 80 percent of estimated TPU shipments in 2028 and 2029.

Shipments still growing in 2029, even as AI spending is expected to peak

The most striking part of the forecast is its shape. The market generally expects global AI capital spending to peak in 2028, but Kuo's numbers show TPU shipments still growing in 2029, driven by direct orders from leading AI service providers rather than by Google's own data centre expansion alone.

Kuo also says the uncertainty around Nvidia's leadership is helping Google's case with customers. With Nvidia in the middle of a chief executive succession after Jensen Huang's resignation, Google is now discussing post-2028 long-term plans with customers, according to the analyst's note.

MediaTek emerges as the biggest winner, Kuo says

The reshuffle reaches beyond the buyers. Kuo reports that Google's v10 TPU line has been significantly reorganised: the originally planned Icefish and Merope designs have been replaced by a new design called Quillfish, expected to enter mass production in 2029, with the design order secured by MediaTek.

MediaTek, Kuo writes, will become the sole supplier for the v9 designs, known as Humufish and Triggerfish, as well as for v10. He states bluntly that MediaTek will be the biggest winner in TPUs over the next few years. Broadcom's 8i chip carries a higher gross margin, Kuo notes, but Broadcom is shifting its focus toward non-TPU businesses.

That reading is not unanimous. A UBS supply chain survey in October concluded that Broadcom was deeply involved in Google's TPU projects for the two generations after v8, a different picture from Kuo's account of MediaTek winning v9 and v10 outright. Which version proves right will only become clear from shipment data as the new chips reach production.

What is confirmed, and what is not

Every unit count in this story is an estimate from Kuo's industry survey, not a figure disclosed by any of the companies involved. Google, OpenAI and Anthropic have not confirmed the 26 million figure, the split between the two labs, or the payment mix. Anthropic's public filing does point in the same direction in dollar terms, describing additional TPU compute expected to be delivered between 2028 and 2031, but it does not state chip counts.

What the survey does establish is the direction of travel: two of the largest AI labs are planning their compute around Google's chips years in advance, and Google's custom silicon business now has order visibility stretching further into the future than the rest of the AI accelerator market. For a business that began as an in-house project, that is a significant change in status.

Related reading: Google's TPU 8tx and 8i Max variants target confidential AI and long-context workloads.

Sources

What's Your Reaction?

Like Like 0
Dislike Dislike 0
Love Love 0
Funny Funny 0
Angry Angry 0
Sad Sad 0
Wow Wow 0
Nahid Joy I’m Nahid Hasan Joy, a technology writer, web developer, and digital enthusiast with a strong interest in the ever-changing world of technology.