Huang Renxun just said:
As the pace of growing computing demand outstrips the ability of frontier labs to secure infrastructure, the next bottleneck for AI is “land, power, and enclosures.”
$NVDA is currently using its balance sheet to lock in long-term sites like PORTS-Pike. Each compute generation may support about 1.5 million GPUs, and could bring up to $200 billion in NVIDIA revenue.
OpenAI’s broader commitment could reach about 16 GW, representing computing capacity of roughly $600 billion by 2030, as Nvidia expands from supplying chips to securing the entire infrastructure around them.
As the pace of growing computing demand outstrips the ability of frontier labs to secure infrastructure, the next bottleneck for AI is “land, power, and enclosures.”
$NVDA is currently using its balance sheet to lock in long-term sites like PORTS-Pike. Each compute generation may support about 1.5 million GPUs, and could bring up to $200 billion in NVIDIA revenue.
OpenAI’s broader commitment could reach about 16 GW, representing computing capacity of roughly $600 billion by 2030, as Nvidia expands from supplying chips to securing the entire infrastructure around them.