Nvidia is facing renewed scrutiny over the sustainability of the artificial intelligence spending boom after technology critic Ed Zitron accused the company of encouraging artificial scarcity in the GPU market. Zitron, chief executive of public relations firm EZPR, argues that Nvidia’s sales strategy and the concentration of its customer base raise questions about whether demand for AI computing infrastructure can continue supporting the industry’s rapid expansion.
A central concern is Nvidia’s customer concentration. According to figures cited in the source material, five companies accounted for approximately 70% of Nvidia’s accounts receivable during the first half of the year. This concentration exposes the company to potential payment and demand risks if a small number of major customers reduce spending, delay purchases or encounter financial difficulties. Investor Steve Eisman has also highlighted the concentration risk, reinforcing the debate over how dependent Nvidia’s growth has become on a relatively narrow group of buyers.
Zitron went further, alleging that Nvidia CEO Jensen Huang is engaging in channel stuffing by encouraging customers to purchase GPUs ahead of potential future shortages. He argues that this approach could sustain elevated prices and accelerate purchasing commitments even when customers may not immediately need all the computing capacity being acquired. These are Zitron’s allegations, not established findings of misconduct, and the available material does not independently demonstrate that Nvidia is deliberately manufacturing scarcity.
The broader question is whether the rush to secure AI infrastructure reflects durable end-user demand or a purchasing cycle increasingly driven by expectations of future shortages. Semiconductor supply chains can involve lengthy lead times, giving customers legitimate reasons to reserve capacity in advance. However, if companies accumulate computing equipment faster than they can deploy it or generate sufficient returns from it, current sales could overstate the durability of underlying demand.
Zitron also questioned the pace of data-center construction and how quickly planned capacity becomes operational. He cited a report indicating that Microsoft had approximately two gigawatts of AI computing capacity online out of a reported total of 12 gigawatts. He also pointed to Amazon’s cancellation of a proposed data center in Loudoun County, Virginia. These examples raise questions about the timing of infrastructure deployment, although neither figure nor cancellation alone establishes that the broader AI investment cycle is deteriorating.
Another concern is the interconnected financial exposure surrounding major AI developers and cloud infrastructure providers. Zitron argues that OpenAI and Anthropic have become critical to the growth expectations of companies including Microsoft, Alphabet, Amazon, CoreWeave and Nebius. If AI developers fail to generate enough revenue to justify their computing commitments, infrastructure providers could face weaker demand, while investors may reassess the earnings assumptions supporting AI-related valuations.
Zitron’s argument ultimately centers on GPU sales as a key driver of the current investment cycle. If spending continues to depend on customers purchasing ever-larger quantities of computing hardware at rising prices, any slowdown could affect Nvidia and companies investing heavily in data centers. Government support might cushion some losses, but it would not necessarily resolve a mismatch between infrastructure supply and economically sustainable demand.
The counterargument is that advanced AI infrastructure may require substantial upfront investment before its full commercial benefits become visible. Long deployment schedules and concentrated purchasing do not automatically indicate excess capacity or artificial scarcity. The decisive issue is whether computing demand translates into productive use, recurring revenue and returns sufficient to support continued investment.
For Nvidia, the next test is not simply whether customers keep ordering GPUs, but whether those purchases remain financially sustainable as more computing capacity comes online. Evidence of slowing orders, delayed deployments or weaker spending by major customers would strengthen concerns about concentration and overinvestment. Continued deployment and demonstrable commercial returns would weaken the bearish case. For now, Zitron’s criticism highlights a genuine question facing the AI sector: whether the infrastructure boom is being driven by durable economic demand or by expectations that become increasingly difficult to sustain.
