What happens when the world's best-funded AI companies still can't figure out how to make money?
That question is no longer hypothetical. OpenAI reportedly spends more than it earns, despite 300 million weekly active users. Anthropic has raised more than $9 billion in funding - including a $4 billion commitment from Amazon - but remains unprofitable. Google is weaving AI into every product while watching whether the cost justifies the returns. The assumption that someone would crack the business model before the money ran out is being tested in real time.
The math is genuinely hard. Training a frontier AI model - one that competes at the top of capability benchmarks - costs hundreds of millions of dollars per run. Running it for users, called inference (meaning processing each query through the model), isn't cheap either. ChatGPT's $20/month Plus tier almost certainly loses money on heavy users. The revenue math only works in aggregate, at massive scale, with strong conversion from free to paid tiers. That conversion has been slower than anyone hoped.
Enterprise Revenue Is the Real Bet
Both OpenAI and Anthropic have bet that selling to companies is where sustainable revenue lives. API access, enterprise contracts, custom deployments - these carry higher margins than consumer subscriptions. But enterprise software sales cycles are slow, IT procurement is conservative, and many companies are still running pilots that haven't converted to committed deployments.
There's pressure from below too. Meta's Llama 4, released in April 2026, is competitive with many paid models and available free under an open license. Companies with engineering teams can self-host it for near-zero marginal cost per query. That doesn't threaten ChatGPT's consumer base, but it competes directly with API revenue - the part of the business that was supposed to carry the financials.
The Funding Runway Has Limits
The deeper issue is timing. AI labs burned through capital on the bet that these models would become essential business infrastructure before the runway ran out. Investors have stayed patient - AI is still treated as one of the most significant technology bets in decades - but "show us the path to profitability" is a different conversation than "here's another billion."
For users of these tools, the monetization pressure is already showing up in practical ways: more aggressive upsells, features gated behind higher tiers, API pricing that changes without much warning. Companies building products on top of AI APIs are particularly exposed - when their model provider adjusts pricing, their margins move overnight.
The major AI labs aren't disappearing. But the shift from "grow at any cost" to "demonstrate a real business" is happening now. The next 18 months will likely determine which companies have a durable revenue model and which were always dependent on the next funding round.