Huang: 3-month 1 GW data center delay costs $10 billion

Huang's warning came before his departure from Nvidia, which followed a board investigation into reports he was targeted by a Chinese spy operation and his admission of an affair with a former executive at Megaspeed. He resigned on Nov. 12, but his cost estimates for data center construction remain a useful benchmark for understanding the economic stakes of delays.

The gap between a $50 billion construction cost and $40 billion in annual revenue potential explains why the industry is so focused on speed. Nvidia itself is trying to reduce construction hurdles by shifting more assembly work to the supply chain rather than on-site workers, which could let components enter operation faster. Yet the bigger long-term bottleneck is power, not assembly.

McKinsey & Co., Morgan Stanley, J.P. Morgan and Goldman Sachs all describe energy as the binding constraint on AI growth, with McKinsey noting that incumbents cannot meet demand for power. That framing suggests that investors seeking AI exposure might consider power stocks, either high-risk SMR plays like NuScale and Oklo or more established names like Constellation Energy and Bloom Energy, the latter of which is expected to see sales growth exceed 100% this year.