Nvidia CEO Jensen Huang said longer-lived systems and quicker installation could help operators borrow against equipment as they fund AI infrastructure amid power and supply constraints.
In an Oct. 21 interview with Yahoo Finance, Huang said a gigawatt-scale data center can cost about $50 billion. He estimated that wiring and starting delivered computers takes four to six months, during which the equipment has yet to generate revenue.
Suppliers might take on more assembly and setup before equipment reaches customer sites, Huang said. In discussing Nvidia’s DSX pod, he said quicker deployment could bring forward the start of commercial service for cloud operators.
Nvidia systems previously assigned a four-year useful-life assumption could remain productive for eight years or more, Huang said. He estimated Vera Rubin could serve for seven to eight years, citing software improvements and CUDA. Equipment may retain value for borrowing if another business can use capacity an initial customer no longer needs, he said.
Huang said demand is increasing for open and closed AI models as more people use them for practical tasks. He cited limited power, sites and buildings, delayed generator deliveries and grid connections, and a shortage of construction workers. Production pressure has moved among stages: Huang cited wafers, connectors, silicon photonics, high-bandwidth memory and advanced packaging. He said supply strains may last for another couple of years.
Differences in power availability and permitting allow Nvidia to direct equipment to customers with ready sites while it works with others on local obstacles, Huang said. He said customer demand gives the industry grounds to address supply and financing constraints, while capital markets and varied funding tools would be needed to pay for infrastructure and broaden customers’ access to borrowing.
