AI Giants Pivot to Token Billing as Enterprise Costs Surge

Major artificial intelligence providers, including OpenAI, Anthropic, and Microsoft, are transitioning from subscription-based models to token-based billing, a move that has significantly increased costs for enterprise users. This structural shift follows revelations that platforms had been running massive subsidies, with some incurring costs up to 70 times the revenue generated. The change marks the end of an era where flat-rate subscriptions sheltered users from the true expense of compute, exposing a severe affordability crisis across the sector.

The financial strain driving this pivot is evident in OpenAI’s 2025 financial results, which reported a net loss of $38.5 billion. Total costs and expenses reached $34 billion against revenue of $13.07 billion. Independent analysis highlighted the scale of the distortion, with tests showing that a $200-a-month Anthropic subscription could burn $8,000 in tokens, while an equivalent OpenAI plan could consume $14,000. This subsidy model, described by analysts as akin to a "drug-dealer's algorithm" designed to lock in users before raising prices, has proven financially unsustainable for the providers.

Microsoft is actively managing its exposure by restricting internal access to rival tools and consolidating its engineering teams onto its own infrastructure. The company plans to cancel most Claude Code access for engineers in its Experiences and Devices division by June 30, 2026, shifting them to GitHub Copilot CLI. This decision follows reports that the week-over-week cost of running GitHub Copilot nearly doubled since January, prompting Microsoft to pause new signups for student and paid individual tiers and tighten rate limits.

For enterprise customers, the transition to token-based pricing has resulted in immediate and steep cost increases. One company CEO reported that their spend increased sevenfold on the first day of the switch, noting that large language model companies had previously subsidised all usage. With compute costs now exceeding human labour expenses in some cases, businesses are reassessing their AI adoption strategies, with some finding that it is now more expensive to use AI than to hire staff for comparable tasks.

The industry is also grappling with the broader implications of these costs on capital markets and future profitability. With the AI sector accumulating an estimated $3 trillion in debt, servicing this debt requires hundreds of billions in annual profit. Analysts note that for the industry to remain viable, it must either replace human labour at a colossal scale or achieve profitability margins that currently remain out of reach, leading to a reevaluation of the technology's economic value proposition.