The arithmetic is not subtle. If the consulting firm's estimate of total yearly AI infrastructure costs runs to something in the region of $6 trillion, then $4.2 trillion of that—the figure cited in the MarketWatch report published 29 September 2026—is currently uncovered by any accountable revenue stream. That is not a rounding error. That is the dominant share of the bill.
The infrastructure in question is the data centres, chips, power capacity and network buildout that hyperscalers and their suppliers have been racing to complete. The capital commitments are real and largely already made. What remains unresolved is who pays for it on the other side of the ledger, and when.
This is not the first time the question has been raised, but attaching a dollar figure to the gap is a different exercise from voicing general scepticism. A $4.2 trillion shortfall implies that current AI revenue—enterprise contracts, API fees, productivity tools—covers roughly one dollar in three of the cost structure being assembled to produce it.
The unresolved question is whether demand catches up to capacity before the capital cycle turns. History suggests those two curves do not always arrive in the same decade.
Gabriel Fenech 
Isla Camilleri 
Alexandre Noir