Comment by yogthos
20 hours ago
We don't have to take Ed's word for it. Anybody who's capable of doing grade school math can see that the numbers simply don't work. These companies are literally spending orders of magnitude more money than they're actually bringing in. Cursor, who've been renting Claude, estimated just recently that a $200-per-month Claude Code subscription could use up to $2,000 in compute. https://www.forbes.com/sites/annatong/2026/03/05/cursor-goes...
Interesting story. Here's what it says:
> According to a person familiar with the company’s internal analysis, Cursor estimated last year that a $200-per-month Claude Code subscription could use up to $2,000 in compute, suggesting significant subsidization by Anthropic. Today, that subsidization appears to be even more aggressive, with that $200 plan able to consume about $5,000 in compute, according to a different person who has seen analyses on the company’s compute spend patterns.
The load-bearing detail here is if that means $2,000 of internal server+electricity costs, or $2,000 if they were to charge at their API pricing instead of the subscription cost.
The latter is how I understand these things to work right now. If it's the former then yeah, Anthropic are losing a TON of money on those subscriptions.
That's the big question, and nobody really knows what the operating costs actually are right now.
Frankly, everyone in the industry knows. When people make these statements without additional clarity they always talk about API prices. You can look at the NVL72 specs and make estimates for electricity and ownership costs rather easily. Inference at data-center scale is dirt cheap, even with public codes using dynamo and sglang. The mystery is why the early misconceptions about inefficient inference persisted even after NVIDIA was very open about everything they did to help reduce costs dramatically in the last two years.
1 reply →