OpenAI’s plunge into public markets highlighted a crisis of affordability that AI firms have long masked. In 2025 the company reported a $38.5 billion loss, powered by a $200‑a‑month subscription that lets users burn $14,000 in tokens. The subsidy model inflates demand while squeezing margins and forcing enterprises to confront the true cost of compute today.

Ed Zitron’s recent analysis exposed that Anthropic subsidizes its enterprise tier up to 40‑times, while OpenAI does so up to 70‑times. Even though OpenAI’s subsidy is higher, Anthropic’s business adoption has surged, revealing a mismatch between marketing spend—44 percent of revenue—and actual product uptake. This gap forces firms to rethink token‑based billing and reassess ROI today.

Microsoft’s pivot to token‑based pricing for GitHub Copilot mirrors the trend, as leaked documents show the company will pause new signups and tighten limits to curb rising per‑prompt costs. The shift signals that high‑volume users face costs that can outpace employee salaries, a reality that has already shocked small enterprises and prompting reexamination of budget.

With OpenAI’s cash reserves near $50 billion, the company must raise prices soon to cover its $13 billion revenue gap. As OpenAI, Anthropic, and Microsoft transition to true cost‑based models, businesses will need to balance productivity gains against the steep token burn that now rivals hiring additional staff for their operations and prioritizing efficiency in their.