In the large-model world, consumer-grade chips have no price-performance at all — you’re better off buying a token plan. It sounds like complaining. It’s arithmetic.
Ledger, this side: a five-dollar subscription, fifteen billion tokens a month; I alone burn a billion a day. Ledger, that side: one consumer card crawling at a few dozen tokens per second, with the electricity, the VRAM, and the maintenance all yours. Data-center output has hit two yuan per million tokens. Your card sleeps in the study while theirs runs at capacity. Your card didn’t get slower. The commodity price simply fell below the cost line of self-hosting at home.
Then look at the billing unit — the clearer clue. Compute used to be sold as time: one VPS, one month, a list price. Now compute is sold as tokens. Why switch units? Because time is unfair to models — inputs vary in length, generation varies in speed; billed by the clock, who wins and who loses is luck. Tokens track actual computation; that’s what makes them read like a utility meter. Buying a bare-metal box means paying last era’s prices; wanting data-center prices means accepting what sits between — pooled scheduling and amortized utilization, the magic you don’t have at home.
Someone will say: cards are for privacy, for offline work, for saturated workloads. Right — that’s the boundary. Run it hot, need it offline, willing to maintain it: then a card makes sense. Everywhere else, buying a card is donating electricity to the data center.
The GPU isn’t dead. What’s dead is the position of “an individual stacking compute at home.” Compute became a utility — and nobody generates their own electricity.
