TECH Signal 482
AI capex buildout reportedly needs $1.2 trillion in annual token revenue to cover hardware alone
↗
Illustration only Photo by Sebastian Schuster on Unsplash
An analysis argues that even assuming AI models work as promised, current capex spending by major tech companies requires roughly $1.2 trillion per year in token purchases just to cover hardware costs, rising toward $2.5 trillion by 2031, with no clear customer base large enough to provide that revenue.
If the thesis holds, the AI infrastructure buildout is a bet on replacing payroll spending with token spending at a scale that may not materialize. Engineers building on these platforms should understand that pricing pressure, capacity rationalization, or provider consolidation could follow if demand falls short of the revenue hurdle.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Goldman Sachs models AI capex at $765 billion this year rising to $1.6 trillion by 2031, with the big four plus Oracle guiding to roughly $830 to 840 billion combined.
At a 65% gross margin on compute, $765 billion in annual capex requires about $1.18 trillion in annual revenue just to cover hardware, before salaries, interest, or returns.
Much of today's demand is the supply side buying from itself, with Microsoft investing in OpenAI which spends back on Microsoft's cloud, and similar patterns at Amazon and Google.
THE CLUSTER