Nvidia CEO Huang Estimates $60 Billion Price Tag for Gigawatt AI Facilities

NEW YORK — Nvidia Corp. Chief Executive Officer Jensen Huang stated on Monday that constructing a single gigawatt-scale artificial intelligence facility requires an investment between $50 billion and $60 billion, a figure he said is reshaping the financial strategies of emerging cloud providers.

Speaking at the G20 Summit on September 7, 2026, Huang outlined the staggering capital requirements driving the next phase of global AI infrastructure expansion. The estimate highlights the immense scale needed to power advanced neural networks, moving beyond traditional data center models toward utility-scale energy consumption dedicated entirely to computing workloads.

The high cost of entry has forced neocloud providers, including Nebius Group and Iren, to rethink their funding mechanisms. Huang noted that the sheer magnitude of these capital expenditures is enabling a shift in how these companies secure liquidity. Rather than relying solely on traditional debt markets or equity dilution, neocloud operators are increasingly turning to substantial prepayments from enterprise customers and specialized financing structures to bridge the gap between construction costs and operational revenue.

Nebius Group, a European provider that has positioned itself as a key alternative to major hyperscalers, is among the firms adapting to this new economic reality. The company has begun structuring deals where large-scale AI clients commit funds upfront to secure capacity in facilities still under development. This approach allows providers to lock in revenue streams before breaking ground, effectively using customer capital to leverage construction financing.

Iren, an Italian energy and technology firm, is similarly navigating the intersection of power generation and data center economics. As a partner in building AI-ready infrastructure, Iren faces the challenge of securing the massive upfront capital required for gigawatt-scale projects while managing long-term energy contracts. The financial models proposed by Huang suggest that without such prepayment strategies, the timeline for deploying these critical facilities could extend significantly, delaying the global rollout of next-generation AI capabilities.

Huang's comments underscore a broader industry trend where the bottleneck for AI growth is shifting from chip availability to the ability to fund and power the massive physical infrastructure required to run them. The $50 billion to $60 billion price tag represents not just hardware costs, but the integration of power grids, cooling systems, and site development necessary to sustain continuous high-performance computing.

While the cost estimates provide a clearer picture of the investment landscape, questions remain regarding the long-term sustainability of prepayment models if demand for AI compute fluctuates. Industry observers are also monitoring whether traditional financial institutions will continue to offer favorable terms for such capital-intensive projects or if the burden will fall entirely on customer advances. As the race for AI dominance accelerates, the ability to finance these gigawatt facilities may become the defining factor for market leadership in the coming decade.