AI's Economics Don't Make Sense
GitHub Copilot is shifting to usage-based pricing in 2026, ending its subsidized subscription model as rising AI compute costs make current pricing unsustainable. This reflects a broader crisis in the generative AI industry, where companies have long operated at a loss by undercharging users relative to the actual cost of running AI models. As services move toward charging based on token usage, customers face potentially steep price increases and must adapt to new, less predictable billing models. The shift exposes the flawed economics of flat-rate subscriptions for AI-powered tools, which have masked the true cost of AI inference for years.
- ▪GitHub Copilot will transition to usage-based pricing on June 1, 2026, charging users based on the actual compute cost of AI models rather than a fixed number of requests.
- ▪Microsoft previously lost an average of over $20 per Copilot user monthly, with some users costing up to $80 per month in compute expenses.
- ▪AI startups like OpenAI and Anthropic have allowed users to consume far more in compute than their subscription fees cover, sometimes by a factor of 8-to-1.
- ▪The shift to usage-based pricing reveals the unsustainability of flat-rate subscriptions for LLM-powered services due to wildly variable inference costs.
- ▪Users are reacting negatively, citing confusion and frustration over suddenly bearing the real costs of AI usage they were previously shielded from.
Opening excerpt (first ~120 words) tap to expand
If you liked this piece, please subscribe to my premium newsletter. It’s $70 a year, or $7 a month, and in return you get a weekly newsletter that’s usually anywhere from 5,000 to 18,000 words, including vast, detailed analyses of NVIDIA, Anthropic and OpenAI’s finances, and the AI bubble writ large. I recently put out the timely and important Hater’s Guide To The SaaSpocalypse, another on How AI Isn't Too Big To Fail, a deep (17,500 word) Hater’s Guide To OpenAI, and just last week put out the massive Hater’s Guide To Private Credit.I also just did a piece about how OpenAI will kill Oracle, and I’ve used some of the materials in today’s piece.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Ed Zitron's Where's Your Ed At.
Discussion
0 commentsMore from Ed Zitron's Where's Your Ed At
