The $200 plan that lets you burn $8,000 in tokens, and the bill coming due
SemiAnalysis found a $200/month Anthropic subscriber can burn $8,000 in tokens; an OpenAI one, $14,000. Rosenthal's question is what happens to all of us when that subsidy stops.
David Rosenthal calls it the drug-dealer's algorithm: subsidize usage to build dependence, then move customers from flat subscriptions to metered token pricing. He cites OpenAI's 2025 numbers — $13.07B revenue against $34B in costs.
The thread's top objection: nobody knows the per-token margin, so the 40x/70x 'subsidy' could just be API prices that are wildly inflated. But the corporate anecdotes were brutal — Fortune 200 engineers watched the AI taps shut off the moment the bills landed.
The drug-dealer's algorithm
David Rosenthal's post frames the AI business on the drug-dealer's algorithm: give away enough of the product cheaply to build dependence, then raise prices once switching is painful. The mechanism is the migration from flat-rate subscriptions to per-token pricing, which Anthropic, OpenAI and Microsoft have all begun pushing on enterprise customers.
The number that makes the case vivid comes from SemiAnalysis testing: on a $200/month Anthropic plan a user can burn roughly $8,000 in tokens, and on a $200 ChatGPT plan about $14,000. That implies subsidies of "up to 40 times" for Anthropic and "up to 70 times" for OpenAI, if you assume the token price itself isn't also subsidized. Rosenthal stacks that against OpenAI's 2025 financials: $13.07B in revenue, $34B in costs, a $20.92B loss, and $5.73B (44% of revenue) on sales and marketing. His closing arithmetic is the alarming part: to service roughly $309B in annual debt, the industry would need to displace 46.8 million jobs, about 27% of all US employment.
"You can't tell the subsidy from a 70x markup"
The most-upvoted technical objection attacked the load-bearing assumption. The 40x/70x gap between subscription and API pricing tells you nothing about cost, because you don't know the API margin:
"This author even says it's implausible that the api charges 4x marginal cost when I think it's very likely even higher than that. The entire rest of the post sits on this faulty assumption. Fixed costs don't matter when marginal revenue is profitable and growing rapidly." — HDThoreaun
An engineer who hosts these models at scale put rough numbers on the unknown, and they weren't the 75% gross margin SemiAnalysis assumed:
"Given my experience with hosting these models at scale… I don't think the margins are nearly as high as 75% if the models are as big as people claim… Context sizes have fucked us all." — an engineer who hosts large models at scale
A detailed rebuttal pointed out that the labs don't even sell the $200 plan to enterprises, undercutting the framing that enterprise customers are the ones being subsidized, and traced the post's "75% margin" figure back to a single SemiAnalysis tweet that was "a total shot in the dark" — 827a, in a line-by-line takedown of the sourcing.
When the token bill hits a Fortune 200
Whatever the true margin, the demand side is already flinching, and that, several argued, is the real crisis. An engineer at a Fortune 200 company described the whole arc in one month:
"At first, it was the Wild West. Need an LLM? You got it… Then they signed contracts with Anthropic and Google… One month later? They completely cut off access to everybody for both Claude and Gemini. If you wanted access? Suddenly it was several forms… a waiting list that was thousands of people long." — an engineer at a Fortune 200 company
A consultant to Fortune 100s saw the same reflex: token allowances capped at $500/month, with developers told to justify the spend against deliverables:
"Over the last month I have seen companies scrambling to measure deliverables against cost… giving devs a small allowance ($500 a month) and then making them prove their own productivity increases." — a consultant who does client work for Fortune 100s
The WebAssembly co-founder reframed where the squeeze ends up — not in the bill, but in the product:
"The coming AI enshittification is going to be epic… just wait until every stupid thing is powered by AI and is subtly trying to manipulate you to buy/watch/believe some crap all the time." — titzer, co-founder of WebAssembly
What makes the thread land is that the optimists and pessimists converged on the same destination by opposite roads: whether the unit economics are secretly fine or secretly broken, the customer behavior that token pricing just unlocked (monitoring, rationing, escalation for "holding it wrong") does not look like a product anyone can quietly 10x. The exploration phase is over; ROI is now the only question, and a lot of the answers are arriving as suspended Copilot licenses.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
"My job is to write loops": the coding shift Armin Ronacher resents but expects
Boris Cherny, who built Claude Code, says he no longer prompts Claude — he writes loops that prompt it for him. Armin Ronacher thinks that future is coming whether he likes it, and he does not.