SignalCommodities
Goldman's Eric Sheridan prices a gigawatt of AI compute three ways: $40-50B/yr on short-term deals, $20-30B long-term, and only ~$12B to a hyperscaler needing a decent return. That gap is the entire A
High-priority alert based on source significance and keyword relevance.
Context
What happened: A Goldman note cited by zerohedge gives three annual rental values for a gigawatt of AI capacity: short-term, long-term, and hyperscaler return-based.
Key numbers: $40-50 billion a year short-term; $20-30 billion long-term; roughly $12 billion for a hyperscaler; $1+ trillion a year of capex.
Why it matters: The spread between the highest and lowest figures frames the debate over AI revenues, with spot pricing far above the return a hyperscaler needs.
What changed
A Goldman note cited by zerohedge gives three annual rental values for a gigawatt of AI capacity: short-term, long-term, and hyperscaler return-based.
Why it matters
The spread between the highest and lowest figures frames the debate over AI revenues, with spot pricing far above the return a hyperscaler needs.