NVIDIA doubled its commitment to rent cloud-based compute capacity, revealing in a recent regulatory filing that it plans to spend $26 billion over the next six years on servers hosted by major cloud providers. This is a dramatic increase, given that three months ago NVIDIA disclosed half that amount, and it highlights how deeply the chipmaker is investing to secure GPU access for its own internal projects.
The spending arc is steep: $1 billion in the current fiscal year, rising to $6 billion annually in 2027 and 2028, and tapering to $4 billion in both 2030 and 2031. By the end of this period, NVIDIA will likely rank among the world’s largest buyers of cloud infrastructure, despite being the industry’s dominant supplier of the underlying AI chips.
That, to be sure, is a unique position in the tech sector. And it’s also subject to additional variables: the company said portions of these commitments may be reduced or reassigned by its cloud partners, an acknowledgment that demand forecasts can shift quickly in today’s AI cycle.
A Circular Flow
What’s striking is that NVIDIA is expanding its cloud commitments even as it pulls back from the original vision for DGX Cloud, its in-house GPU rental service that once positioned the company as a direct rival to Amazon Web Services and other hyperscalers. Reports suggest that NVIDIA found limited appetite for a service priced above mainstream cloud offerings. Instead, DGX Cloud has become an internal resource, powering everything from chip design to model development.
One example is Nvidia’s arrangement with Lambda, a rising so-called neocloud that rents GPU servers to both tech giants and startups. NVIDIA has agreed to rent 10,000 of its own chips back from Lambda, a circular flow of silicon that reflects how competitive the GPU market has become. These deals help smaller cloud firms bulk up capacity, while giving NVIDIA a guaranteed pathway to the compute it needs. NVIDIA has similar relationships with CoreWeave and Oracle, making the company both supplier and customer in a tight, interdependent ecosystem.
While some industry experts find fault with these circular business arrangements, others see the collaborative agreements as the foundation for a new level of competitiveness in the AI sector, the demands of which require many and varied partnerships.
A Volatile Environment
Behind the aggressive cloud spending is a regulatory environment that remains volatile. NVIDIA’s latest 10-Q filing details how U.S. export controls on its H20 data center chip forced a $4.5 billion charge earlier this year, largely tied to inventory and purchase commitments for a product that suddenly faced strict licensing requirements. Although U.S. officials later granted limited permissions, the company has generated only about $50 million in H20 revenue under those licenses. Washington is also rewriting its broader AI Diffusion rule, leaving uncertainty around how future restrictions might affect Nvidia’s flagship GPUs.
Meanwhile, the global trade picture is becoming more complex. Rapid shifts in tariffs and export rules are prompting NVIDIA to expand U.S.-based manufacturing, a move it says will add resilience to its supply chain. But the company cautioned that these efforts depend on domestic suppliers scaling production quickly, which is certainly not a guaranteed outcome.
Capacity is the Constraint
Amid these pressures, NVIDIA’s enterprise outlook remains buoyant. Revenue jumped 62% year over year in the most recent quarter, driven largely by demand for its Blackwell processors. CEO Jensen Huang said cloud GPUs remain sold out, an indication that the generative AI boom is maintaining its momentum.
Still, the company is navigating a delicate balancing act. Its biggest customers, including Amazon, Microsoft and Google, are buying record quantities of NVIDIA hardware while simultaneously developing their own competing AI chips. NVIDIA’s cloud commitments help ensure access to compute, but they also bind the company more tightly to cloud providers whose long-term strategies may diverge from its own.
If anything, NVIDIA’s $26 billion cloud investment reflects the current reality of the AI arms race: compute is essential but capacity tends to be a constraint, and even the world’s leading chipmaker must secure its spot in line.
