Amazon's Cloud Grip: How GPU Cancellations Expose a Market-Power Problem

Epic Games CEO Tim Sweeney posted publicly that Amazon canceled his company's Nvidia GPU orders through AWS — a disclosure that crystallized a structural concern about cloud market power that analyst Gene Munster has been articulating with increasing urgency. Epic is not a marginal actor; it operates Fortnite, the Unreal Engine, and one of the most significant app stores in gaming. That Amazon, acting through its cloud infrastructure arm, could cancel the GPU capacity Epic had placed orders for — presumably redirecting that supply elsewhere — signals to every AWS customer that their infrastructure access is subject to Amazon's discretion rather than simply market availability.

Munster's argument is architectural: Amazon Web Services achieved dominance by presenting itself as a neutral utility provider, akin to electricity, on which startups and enterprises could build without fear that their infrastructure partner was also a competitor managing supply against them. But Amazon competes directly with significant numbers of its own cloud customers across gaming, entertainment, logistics, and retail. When it controls the supply of Nvidia GPUs — the critical hardware for AI workloads — it holds meaningful leverage over every company that depends on that compute to operate.

The antitrust question turns on a distinction courts have consistently found difficult to draw cleanly. Under Section 2 of the Sherman Antitrust Act, holding a large market share is not itself illegal; the violation requires using that market position to engage in exclusionary conduct that harms competition in a related market. The legal question in any eventual case would be whether Amazon canceled Epic's orders for legitimate business reasons — supply constraints, contract terms, pricing — or to disadvantage a company that competes with Amazon in an adjacent market. Market share alone does not establish the violation; the conduct must go beyond aggressive competing.

Nvidia's role adds another layer. The Financial Times reported that Nvidia has been asking cloud-computing clients to voluntarily cap sales of its chips into China — a request sitting between corporate compliance with export-control expectations and proactive regulatory anticipation. Nvidia already operates under significant export restrictions on its most powerful chips to China; extending that restraint downstream through cloud partners is an attempt to get ahead of potential further regulatory action, though analysts are divided on whether such restrictions accelerate Chinese domestic chip development by giving producers like SMIC and Huawei's HiSilicon additional breathing room.

Bank of America's note this week offered a counterintuitive thesis: if major hyperscalers begin rationalizing AI infrastructure spending, the companies that benefit paradoxically could be the chip suppliers themselves. With GPU supply currently so constrained, any reshuffling of demand could help chipmakers like Nvidia maintain pricing power and reduce customer-concentration risk. Less concentrated demand from a few mega-customers, the argument runs, could produce a healthier and more diversified chip market — even if the process is painful for Amazon's near-term market share.