Google deploys TPU push against Nvidia in AI chip dealmaking war

Google is making a coordinated push to displace Nvidia's dominance in AI chips by targeting the specialized cloud providers known as "neo-clouds"—firms that often began as cryptocurrency miners and now operate GPU clouds. While hundreds of such firms exist, only about half a dozen carry real weight in the market, and Google has been in talks with them about adding TPUs to their service offerings.

Google's pitch to these providers rests on two pillars. The first is diversification: neo-clouds can reduce their total dependence on Nvidia by offering TPU capacity alongside GPU capacity. The second is technical: TPU designs have remained stable over time, whereas Nvidia's architecture changes radically with each new generation, creating genuine operational friction for data centre operators who must manage deployments, integration, and ongoing maintenance of the shifting platform.

Google is backing this pitch with financial creativity. The company is considering offering what amounts to financial guarantees to lenders—arrangements where Google would backstop payments if businesses that borrow money to buy TPUs later cannot find renters or buyers for the chips. This approach mirrors Nvidia's long-standing practice of using financial muscle to support its largest customers, but with a structural advantage: if TPUs remain unsold, Google can absorb them into its own cloud operations at scale, whereas Nvidia lacks a cloud business large enough to serve as a safety valve for stranded hardware. Google is also pursuing a joint venture with investment firm Blackstone to build a TPU-based cloud provider directly, extending the same logic into the market infrastructure itself.

The competitive intensity is already visible. Reports indicate that Nvidia became aware of Google's discussions with neo-cloud provider Nscale and may have offered additional incentives to discourage TPU adoption—though Nscale has publicly stated this is not its position. Jensen Huang, Nvidia's CEO, is said to monitor Google's chip programme closely and regards Google as a significant competitive threat.

Yet the rivalry exists within a framework of mutual dependence. Google remains one of Nvidia's largest customers, purchasing GPUs at scale for a cloud business that serves external clients, and both companies currently need each other's supply. On Google's side, the external TPU business is gaining real traction: Anthropic, Apple, and Meta are among the clients, with Meta emerging as a significant customer. The next bottleneck is manufacturing capacity. TSMC is the essential intermediary through which all chip ambitions must pass. Google secures its capacity via Broadcom as intermediary. Allocations for 2027 production are being determined now, and the capacity Google wins will indicate how seriously TSMC takes the TPU business against competing demands from Nvidia and others. With gigawatt data centres costing $50 billion（約8兆円） to $60 billion（約9.6兆円） and rising, the companies able to guarantee that spending will shape the infrastructure of the entire AI economy.

Google's push into the neo-cloud market represents a strategic shift from competing on chip performance alone to competing on operational stability and financial support. The body identifies a genuine technical friction point: Nvidia's frequent architectural changes impose real costs on the data centre operators who must manage those deployments. By contrast, TPU designs have remained stable, offering a lower-friction alternative—but only if those providers can be convinced to diversify away from Nvidia's near-monopoly.

The financial dimension of this competition is crucial. Google is matching Nvidia's long-standing practice of using balance-sheet muscle to support customers, but with a structural twist: Google can absorb unsold TPUs into its own cloud business, creating a natural floor under the asset. Nvidia, lacking a cloud operation at comparable scale, cannot offer the same guarantee. This asymmetry potentially shifts the risk calculus for lenders and cloud operators considering TPU adoption.

However, the rivalry is constrained by mutual dependence. Google remains one of Nvidia's largest customers, and both companies currently need each other's supply. The real constraint—and the metric to watch—is TSMC manufacturing capacity. Because Google secures capacity through Broadcom as intermediary, the amount TSMC allocates to Google in 2027 production decisions will reveal how seriously TSMC weighs the TPU business against competing demands. That allocation decision will partly determine which company shapes the trillion-dollar AI infrastructure economy.

No comments yet. Be the first to share your thoughts!