December 2, 2025
ROI or R.I.P.?
IBM CEO says there is 'no way' spending on AI data centers will pay off
IBM boss rains on the AI gold rush—commenters clap back, demand receipts
TLDR: IBM’s CEO says today’s costs make massive AI data centers a money pit, citing $80B per gigawatt and trillions in spending. Comments split between roasting IBM’s judgment, demanding a clear cost breakdown, and arguing targeted AI projects could still make money—crucial as tech bets billions on AI infrastructure.
IBM CEO Arvind Krishna just splashed cold water on the AI bonfire, telling the Decoder podcast there’s “no way” today’s AI data center spending pays off. His napkin math: about $80 billion per gigawatt of capacity; across Big Tech, that’s potentially $8 trillion of spending needing $800 billion in profits just to cover interest. He even put the chance of AGI—artificial general intelligence, a machine as smart as a human—at 0–1%. Cue comment-section fireworks.
The loudest mic-drop came from one user who called IBM a "broken ship" that missed the AI wave, alleging talent exodus and bungled data center experience—pure drama and plenty of schadenfreude. In the “show me the receipts” corner, another reader begged for a line-item breakdown: if a gigawatt of compute really costs $80B, what’s power, land, buildings, chips, and people? Skeptics chimed in that $8T sounds inflated, noting not every lab spends like OpenAI.
Others offered a cool-down: it can pay off—if it’s targeted, not a blind gold rush. The depreciation debate got spicy too: will today’s pricey GPUs be junk in five years, or are performance jumps slowing? Meanwhile, folks memed Meta’s obsession with “capacity” as “capacity for debt,” and joked that Google’s space data centers will need “interstellar ROI.” Bottom line: the community isn’t arguing whether AI matters—they’re roasting the spreadsheet and asking who’s actually getting paid.
Key Points
- •IBM CEO Arvind Krishna said current AI data center capex is unlikely to yield returns at today’s costs.
- •He estimated about $80 billion is required to build and equip a 1‑gigawatt data center.
- •He said 20–30 gigawatts for a single company implies roughly $1.5 trillion in capex.
- •Krishna noted AI chips depreciate over about five years, necessitating rapid utilization.
- •He claimed $8 trillion of capex would require about $800 billion in profit just to cover interest; he also put the chance of current tech reaching AGI at 0–1%.