Sooner or later, they're going to make us actually pay for AI
According to estimates by SemiAnalysis, the true cost of serving a single ChatGPT Plus user is roughly $700 per month. For a $200 Pro subscription, it climbs to $14,000 per month. Yes, you read that right: fourteen thousand. And Anthropic is no different.
Our $20 subscription is being kept artificially low — very, very low — by a historic subsidy operation. And indeed, the two startups with the highest market valuations ever are burning tens of billions of dollars every year.
The latest numbers are staggering: OpenAI is projected to spend $22 billion in 2026 against $13 billion in revenue, with an operating loss of $14 billion and a break-even point tentatively set for 2030 (maybe). Anthropic, supposedly more efficient, still burns $2.16 for every dollar of revenue and just surpassed OpenAI in market valuation.
The question is no longer "is it ethical to use AI?" The question is: will we be able to afford it?
Because the bill is coming, sooner or later. And when VCs stop subsidizing our prompts, the true price of tokens will surface. Agents that today seem "magical" consume 5-30x more tokens than a simple chatbot, and companies like Uber have already blown through their annual AI budget in 4 months thanks to Claude Code.
"Ah, but open-weight and Chinese models exist, they're much cheaper!"
True. DeepSeek V4 Flash costs $0.14 per million tokens (about two orders of magnitude less than GPT-5.5), GLM 5.2 beats GPT-5.5 on coding benchmarks at one-sixth the price, and Chinese models have slashed inference costs by 90% compared to their American competitors.
But there's a big, fat "but": those are subsidized too. DeepSeek made a 75% discount permanent — a promotion that became the market price. GLM 5.2 is cheaper, but requires 800GB of VRAM for self-hosting. And nobody knows how long these prices last when providers actually have to cover their costs.
The point is: we're living in an artificial price bubble. Today we use AI like it's water, tomorrow we'll find out it's champagne. And the debate "is it right or wrong to use AI?" will be settled not by ethical consensus, but by economic impossibility.
The next crisis won't be "AI is stealing our jobs." It will be "AI is too expensive for anyone but Big Tech."
Sooner or later, they're going to make us actually pay for it.
