Nvidia CEO Huang Says AI Factory Cost Per Gigawatt Has Risen to $80 Billion-$100 Billion
Nvidia Chief Executive Officer Jensen Huang said AI-factory capex intensity is rising dramatically, with the cost of building one gigawatt of capacity increasing to about $80 billion to $100 billion from roughly $50 billion.
Related comments described newer systems as delivering more tokens per watt, with providers getting about five times more throughput per watt on average even as cost per watt doubles. That suggests efficiency gains may come with a higher upfront bill, leaving model providers to decide whether to absorb the increase or pass it through to customers.
From the sources (6 posts)
@firstadopterJensen says the cost per gigawatt is going higher: $80-$100 billion per gigawatt into AI factory (from $50 billion) $NVDA
@firstadopterNot exactly sure why the cost is going up. Something to follow up on later but this should be bullish for AI infra companies
@zephyr_z9Memory prices Jensen raising server prices as they can deliver exponentially more Tokens per Watt
@dnystedtRT @TheTranscript_: $NVDA CEO: AI factory capex intensity is rising dramatically "Each one of these, at the one-gigawatt level, started at…
@zephyr_z9Now the big question is, will model providers absorb the cost or pass it to the consumers Providers will get a 5x thruput boost per watt on average, but the cost will be 2x per watt
@zephyr_z9@nrehiew_ Jensen said that the cost of building 1GW of compute is going up from $50B to $80B-$100B