Token Demand Surges Even As Prices Fall

The market for tokens is huge and growing — but that growth reflects demand accelerating as prices plunge. Spending on AI models will roughly double from $15.5 billion in 2025 to $32.6 billion in 2026, per Gartner.

Anthropic drives faster growth. Its annualized first-quarter 2026 revenue rose fivefold to $45 billion from roughly $9 billion at the end of 2025, according to Sacra.

Demand for tokens is rising so fast that price cuts have a less negative effect on revenue growth — and may even encourage more demand. Token consumption is forecast to rise 24-fold by 2030, to 120 quadrillion tokens a month, projects Goldman Sachs Research, which notes chipmakers are lowering annual token costs by 60% to 70%.

Simply put, volume growth is running at twice spend growth as per-token prices fall. Business token usage grew 1,001% from January 2025 to April 2026, while spend jumped 497%, reported Ramp.

Big AI Budgets Keep Growing — But ROI Still Lags

AI is not paying off for many companies, with agentic AI proving especially wasteful. In April, companies had already gone three times over their annual token budgets due to agentic AI, “which makes dozens of token-burning calls per task,” noted The FinOps Foundation. Another problem: For every 10 hours AI saves, employees lose nearly four fixing its output, found Workday.

While individual productivity gains are significant — as high as fivefold — generative AI’s ROI remains elusive. A mere 29% of organizations see significant ROI from generative AI, and just 23% from agents, noted Writer’s 2026 enterprise survey.

A more optimistic take comes from Boston Consulting Group, which claims 25% of its 2025 revenue came from AI-related services. So-called AI visionaries generated 1.7 times more revenue growth than laggards, only 6% of executives would cut AI investment even if 2026 returns disappoint, and agentic workflows capable of true labor substitution are on the way, noted BCG’s AI Radar.

How Token Price Cuts Could Hit The AI Industry

Token price cuts have already happened, and more could be on the way. Anthropic cut Claude Opus pricing 67% at the Opus 4.5 launch in November 2025, and OpenAI already offers Flex processing at a 50% discount.

While the magnitude of future token price cuts remains unknown, their damage to participants throughout the AI industry value chain could be significant. Here’s how:

• AI chatbot margins could fall. Despite cost cuts from AI chip providers — Nvidia’s Blackwell plus its NVFP4 format cut cost per million tokens 75% to 5 cents, noted Nvidia — margins could drop further. Anthropic already lowered its gross margin projection to 40% because of inference costs on Google and Amazon servers.
• Nvidia’s market share could slip. To survive lower token prices, AI chatbot providers must slash cost per token. By 2030, custom chips will run 40% of AI servers, growing about three times faster than GPUs, according to TrendForce. Custom chips — including Google TPUs, AWS Trainium, Microsoft Maia and Meta MTIA — cut inference costs between 30% and 50% compared to Nvidia GPUs.
• Neocloud debt risk may rise. Up to $300 billion in bonds will finance AI capital expenditures of $3 trillion, notes Morgan Stanley. Lower token prices could mean lower revenue for neoclouds. Since their debt service is fixed, the cash-flow squeeze could tighten for CoreWeave — which holds $29 billion in total liabilities with $9.7 billion due within twelve months, noted Yahoo Finance, and pays 9.75% interest on senior notes — and Oracle, which now carries more than $130 billion in debt plus $248 billion in lease commitments, according to Quinn Emmanuel.

Two Scenarios Investors Should Watch

Lower token prices could boost consumption, or AI’s limited ROI could drive companies to cut back.

A token price war could mean a lower valuation for OpenAI and Anthropic. But if consumption rises, AI capital expenditures could keep climbing even as chip demand shifts from Nvidia to custom silicon providers that supply the cheapest cost per token.

If demand drops, thin AI chatbot margins will narrow further, the $700 billion AI data-center buildout will search more urgently for revenue, and a trillion-dollar-scale debt stack priced in tokens could put borrowers and lenders at greater risk.