Short-seller Jim Chanos questions Nvidia's AI infrastructure cost estimates as significantly higher than public company guidance.

Short-seller Jim Chanos has publicly questioned the accuracy of NVIDIA CEO Jensen Huang's cost estimates for building large-scale AI data centers, asserting they are significantly higher than what other industry players are reporting to investors. Chanos highlighted Huang's projection that a one-gigawatt (1GW) "AI factory" would cost between $20 to $30 billion before the cost of GPUs is factored in, characterizing this figure as "well above what many AI data center companies are currently telling investors their costs will be."

The critique centers on what has been termed "Jensen's math," which estimates a total 1GW facility cost of $60–$80 billion, with $40–$50 billion representing the "compute cost," which constitutes NVIDIA's potential revenue. Huang has stated that "every gigawatt is about $40 billion, $50 billion to Nvidia," a figure that shapes expectations around the company's revenue potential. This scrutiny arrives at a critical juncture, as the newly announced NVIDIA-OpenAI partnership aims to deploy at least 10 gigawatts of AI systems, a project Huang himself called "giant."

According to projections from asset management giant Brookfield, total AI data center capacity is expected to surge from 7 GW in 2024 to 82 GW by 2034, representing a 28 percent compound annual growth rate. When considering AI workloads specifically, capacity is projected to expand from 44GW to 156GW between 2025 and 2030. Based on "Jensen's math," the 156GW capacity expected by 2030 multiplied by the $40 billion figure stated by Huang yields a market opportunity of $6.2 trillion for NVIDIA—a figure that would still reach $3.1 trillion even if halved.

Chanos's skepticism raises a fundamental question for investors: if Huang's non-GPU cost estimates are accurate, data center operators and infrastructure companies may be significantly underestimating future capital expenditures, potentially squeezing their margins. This casts doubt over the industry's growth narrative, suggesting that the true cost to power the AI revolution may prove substantially higher than currently anticipated.