June 7, 2026
Silicon Valley’s tab is showing
Anthropic/OpenAI may be spending more than $1000 for every $100 you pay them
AI’s cheap-looking price tag has commenters screaming that the real bill is still coming
TLDR: The article argues some AI firms may be spending far more to run their services than customers pay, raising doubts about whether today’s prices make any sense. Commenters are split between “get ready for a brutal price hike” and “wait for real financial proof,” with jokes that AI is basically in its cheap-Uber phase.
A new blog post is pouring cold water on the shiny promise of chatbot-powered coding, arguing that companies like OpenAI and Anthropic may be burning more than $1,000 behind the scenes for every $100 customers pay. The writer also side-eyes the industry’s favorite hype line — that artificial intelligence could soon “build itself” — and basically says: not so fast. In plain English, the claim is that these tools may look magical, but they may be wildly expensive, messy, and nowhere near as efficient as the marketing suggests.
And the comments? Absolute popcorn material. One camp is convinced we’re still in the “too cheap to be true” phase, with one reader warning, “Hang on to your wallet”, saying the real prices won’t show up until users are fully dependent. Another commenter compared today’s AI boom to the old days of absurdly subsidized Uber rides, joking that we’re in the “$6 Uber ride era of AI” — cheap for now, painful later. That line alone pretty much became the thread’s unofficial meme.
Meanwhile, the practical dreamers are already fantasizing about a future where people skip subscriptions entirely and buy one-time AI hardware cards for their computers instead. On the calmer side, a few commenters urged everyone to stop doom-posting and wait for hard numbers, especially if Anthropic ever files to go public. So yes, the article questions the business math — but the community turned it into a bigger drama: Is AI a revolution, or just a luxury free trial with terrifying fine print?
Key Points
- •The article introduces a new discussion of coding with large language models after a 15-month gap in the author’s generative AI coverage.
- •It critiques Anthropic’s blog post “When AI builds itself,” arguing that caveats are present but overshadowed by stronger claims.
- •The author questions whether coding benchmarks such as 50% or 80% success rates are meaningful for fully agentic coding without human oversight.
- •The article argues that increases in lines of code checked in per day may reflect overhead rather than true productivity gains.
- •The author reports an ongoing Claude Code experiment that has produced about 40,000 lines of code and a functional but incomplete application, while also surfacing cost concerns.