Google TPU Dilemma: Opening the Market, but New Cloud Providers like Nebius and CoreWeave Choose to Wait and See

```

Google is aggressively pushing TPU into the market, but has hit a wall at the most critical distribution phase.

According to the latest report from The Information, in the same week that Google announced plans to sell TPU chips directly to customers, three major U.S. cloud service providers—Nebius, Lambda, and CoreWeave—stated at a public event: they do not intend to use TPUs in the short term.

Google CEO Sundar Pichai said on the earnings call that the company plans to sell TPUs to "specific customer groups," focusing on financial services and frontier AI sectors. This statement already reveals Google's strategic adjustment—not pursuing GPU-like widespread adoption, but focusing on a select few willing to cooperate deeply.

However, the collective refusal by the three cloud service providers reveals the real difficulties Google faces in building its TPU ecosystem.

"We bleed green"

When asked about TPUs, Lambda CFO Chuck Fisher gave a one-sentence answer: "We bleed green." Green is Nvidia’s signature color.

Nebius Chief Revenue Officer Marc Boroditsky’s data is even more direct: 99% of their customers’ demands come from Nvidia GPUs. Occasionally, customers ask about TPUs, but it's usually ex-Google employees—because they've used the chip before.

"Unfortunately, the circles of ex-Google employees cannot sustain a market," said Boroditsky.

CoreWeave's VP of Corporate Development Nick Robbins explains from the perspective of return on investment. He points out that the main TPU users—Google itself, Anthropic, and Meta—are also major GPU buyers. That means even if TPUs gain more market share, GPU demand won’t disappear.

"For every dollar we invest, every megawatt we allocate, it’s a risk-adjusted bet to see what brings the longest and highest return," Robbins said. "If 99% of the market wants the same thing—even if that number drops to 90%—it’s still hard not to focus on that 90%."

The stance of these three companies is not surprising. Nebius, Lambda, and CoreWeave are all deeply tied to Nvidia—not only as their largest supplier and key investor, but sometimes as a major GPU customer.

Google's Response: Bypassing the Mainstream, Finding New Paths

Google has tried to collaborate with mainstream cloud service providers. Reports say that last year, Google explored closer partnerships with CoreWeave, Crusoe and others, hoping to deploy TPUs alongside GPUs in their data centers. But most of these negotiations fell through.

Ultimately, Google turned to a lesser-known partner—software and data center startup Fluidstack, providing it with multi-billion dollar leases and debt guarantees for deploying TPUs to Anthropic.

From aiming to enter multiple major cloud providers, to signing with up-and-coming players like Fluidstack, this shift shows that the TPU market outside Google Cloud remains highly concentrated.

On the financing side, Google is also actively planning. According to a February report from The Information, Google has signed an agreement with a large undisclosed investment institution to jointly fund a joint venture to lease TPUs to other customers. Google is also in talks with potential financing partners, planning to use special-purpose entities to borrow and buy TPUs and lease them out.

Unlike Nvidia's approach of supporting early cloud providers, Google has to take a more active role in financing—since there is no ready-made market for TPUs.

Long-term Variable: Can a Multi-chip Ecosystem Break the Pattern?

Despite a clear short-term stance, Nebius’s Boroditsky admits that TPUs may gain greater market share in the future. He says that many companies are now developing cloud software that, if successful, will make it easier for developers to use multiple chip types, reducing dependency on single hardware.

"It's hard for a single brand to form a monopoly—there will be more diversity in the future," he said. "This happened in the CPU space, and I expect the same story will repeat here."

This insight is meaningful for the entire AI chip market. Currently, Nvidia dominates the AI training and inference chip market, but the maturity of multi-chip compatible software may be the key variable in changing this pattern.

Risk DisclaimerThe market has risks and investment requires caution. This article does not constitute individual investment advice, nor does it take into account the specific investment objectives, financial circumstances, or needs of any particular user. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their particular situation. Investing based on this is at your own risk. ```