To save folks looking this up:
>At current 2026 API rates, 500 billion AI tokens would cost roughly $100,000–$750,000 depending on the model, with most flagship models in the $150–$400 per million input tokens range
>At current 2026 API rates, 500 billion AI tokens would cost roughly $100,000–$750,000 depending on the model, with most flagship models in the $150–$400 per million input tokens range
Yet, it is now profitable.
The bet is that people realize how valuable these services are, and despite complaining, they still would pay the higher price. This realization would not happen without this initial subsidy from investors.
It isn't too different from drug dealer's first sample free...
Also, Uber was peak ZIRP + COVID, money was cheap and growth was easy. I think the mountain to climb is a lot steeper now.
Taxis were an established profitable business model and the uber subsidisation wasn’t anywhere near as much as AI subsidies.
You think developers won't/don't stretch subscriptions to the absolute limit? The AI subscription model is like the gym model except a large percentage of the customers work out 24/7/365.
The price these companies charge per token is irrelevant to a discussion about subscriptions. You're comparing two selling prices, when what matters is the actual cost. It's like those ads saying, “Normally $20 million, but today it's only $99!” for something that costs $10 to make.
Sure, they're losing money now. But the question is whether the tokens you use through your subscription will eventually cost them less to generate than you pay for that subscription. Their per-token selling price doesn't tell you that.
????
Uber and Lyft were paying drivers $X but charging users way less than $X so they were literally burning investor money to get market share.
Why does this point come up over and over again, pretending that you can truly separate training and inference costs. Yes, they are separate things but the value OpenAI and Anthropic are presenting to the world is "we're the absolute best, no one else comes close." Well, to keep that up you can't just not train for extended periods of time. You have to keep that engine going non-stop when there's free Chinese models nipping at your heels. You can be profitable on "just inference" all you want but if training expenses dwarf that, you're not going to be profitable overall, and that's the bottom line.
So my conclusion is that it only works because the majority of users doen't fully utilize their allowance. Last month I used almost nothing of my Claude 20x plan (did use Codex though).
https://developers.openai.com/api/docs/pricing https://platform.claude.com/docs/en/about-claude/pricing
OpenAI and Anthropic are both 10/mil in.
https://openrouter.ai/z-ai/glm-5.3#providers https://openrouter.ai/moonshotai/kimi-k3#providers Other frontier models are like 1-3/mil in.
Also, 500 billion is 500,000 millions. At the lower end of your 140/mil estimate that's 70 million dollars. Even at my 2/mil lookup for Chinese frontier models that's one million. Show your math for 100k-750k, please.
Reading between the lines:
"The avid reader of my blog might have noticed that I had previously written two posts that have since been removed. Everyone else might now be wondering which game I am talking about. To both of you I can only say that corporate America was here to ruin our fun."
Keeping it private likely wasn't the original plan...
The tone of the blog suggests they are using the game to build out tooling and expertise; the capability to decompile large projects on the scale of AAA games is not worth nothing, e.g. security or "competitor analysis" that lays trade secrets bare.
AI is a great at laundering code, I thought open-source software was the prime prey, but I should have thought bigger.