Google pays cloud providers to switch to its own accelerators

The battle for dominance in the AI accelerator market is reaching a new level. Google no longer relies exclusively on the technical superiority of its chips — the company is now prepared to pay cloud providers extra just so they choose its branded accelerators instead of competitors' products. A strategy that would have seemed unthinkable for a company of such scale just a couple of years ago now appears to be a forced, but entirely logical move in the face of intensifying competition.

To understand the context, it's worth recalling how the market for computing infrastructure for artificial intelligence has changed over the past two years. NVIDIA continues to hold the lion's share of the GPU market for training and inference of neural networks. Its accelerators from the H100 and B200 series have become the de facto standard, and delivery queues stretch for months.

In these circumstances, Google, which develops its own line of tensor processing units (TPU), finds itself in a paradoxical situation: it has a technically competitive product, but the ecosystem and market habits work against it. Cloud providers and their clients have spent years building workflows around CUDA and NVIDIA's architecture, and simply offering an alternative chip is not enough — you need to provide a compelling reason for migration.

This is where financial incentives enter the scene. According to available information, Google offers cloud providers various forms of financial support — from direct subsidies to favorable lending terms and discounts on cloud services — in exchange for a commitment to purchase and deploy its accelerators. The model resembles the so-called 'circular deals' that OpenAI actively practices: the company invests in startups, which in turn spend the funds received on cloud computing from OpenAI's partners, primarily Microsoft Azure. The money essentially makes a circle and returns to the ecosystem, but in the process creates the appearance of organic demand and strengthens the market positions of all participants in the chain.

Google has apparently decided to adapt this scheme to its needs. The difference is that the Mountain View company is promoting not a software platform, but specific 'hardware' — its TPU, which has so far been available primarily through Google Cloud. Expanding TPU availability among third-party cloud providers could radically shift the balance of power in the market, creating a real alternative to NVIDIA's monopoly and offering developers the choice they so desperately need.

However, competitors — particularly NVIDIA itself — view Google's initiative with skepticism. And there are good reasons for this. First, financial incentives work only as long as the company is willing to sustain them. Subsidies are, by definition, a temporary measure, and once the cash flow dries up, providers can quickly return to familiar solutions. Second, switching to a new hardware platform is not simply swapping one card for another. It requires retraining engineers, adapting the software stack, testing compatibility with existing models. The cost of such a transition often exceeds any subsidies.

There is also a third, purely practical problem: the global component shortage. Even if cloud providers want to purchase TPU in bulk, Google may face production constraints. The company orders its chips from TSMC, whose capacity is booked years in advance among Apple, NVIDIA, AMD, and dozens of other clients. Scaling TPU supplies when every wafer of silicon is worth its weight in gold is a non-trivial task.

Nevertheless, Google's strategy deserves attention because it reflects a fundamental shift in the industry. The AI infrastructure market is so overheated that the largest tech companies are literally willing to pay extra to expand their market share. This is no longer simply competition between products — it's competition between ecosystems and financial resources. Whoever succeeds in building the broadest partner network and creating a critical mass of users around their platform will gain a strategic advantage for decades to come.

For end users — AI system developers and companies implementing artificial intelligence — this battle between giants carries mostly positive consequences. More competition means more choice, more flexible terms, and ultimately more accessible computing resources. If Google manages to break NVIDIA's monopoly even partially, the entire industry wins. The only question is whether Mountain View has the patience and resources to see this game through to the end.