OpenAI, Anthropic weigh token price cuts as enterprise AI spend

OpenAI is reportedly preparing “sharp” token‑price reductions to stay ahead of Anthropic, while both firms face enterprise budget caps that could compress margins and reshape their looming trillion‑dollar IPOs【2】. The move matters because corporate AI spend is already being reined in—Uber limited its AI coding budget to $1,500 per month per tool—prompting a shift toward cheaper open‑source models and threatening the profitability of the two leading chatbot providers.

Enterprise pressure drives price war

Enterprises that subscribe to ChatGPT and Claude are flagging “too high” token costs, with Uber’s 2026 AI coding budget exhausted in four months and per‑engineer expenses ranging $500‑$2,000 a month for Anthropic’s Claude Code and Cursor【1】. To curb spending, firms are routing simple tasks to lower‑cost Chinese models such as Alibaba and DeepSeek, reserving the most expensive models for complex work【1】. This “cost‑optimizing” behavior has already forced Anthropic to cut Claude Opus pricing by 67% at its Opus 4.5 launch in November 2025, and OpenAI now offers a Flex processing tier at a 50% discount【1】.

Token demand rises despite lower prices

Even as prices fall, token consumption is accelerating. Business token usage grew 1,001% from Jan 2025 to Apr 2026, outpacing spend growth of 497%【1】. Gartner projects AI model spend to double from $15.5 bn in 2025 to $32.6 bn in 2026, while Goldman Sachs forecasts a 24‑fold rise in monthly token volume to 120 quadrillion by 2030【1】. The surge in volume mitigates the revenue impact of price cuts, but it also pressures chip makers: Nvidia’s Blackwell and NVFP4 format have already cut per‑million‑token costs by 75% to 5 cents【1】, and custom AI chips are expected to handle 40% of AI servers by 2030, growing three times faster than GPUs【1】.

Competitive ripple effects

Lower token fees could erode margins for chatbot providers, with Anthropic already lowering its gross‑margin outlook to 40% due to inference costs on Google and Amazon servers【1】. Nvidia’s market share may slip as custom silicon—Google TPUs, AWS Trainium, Microsoft Maia, Meta MTIA—offers 30‑50% cheaper inference than Nvidia GPUs【1】. Meanwhile, “neocloud” operators like CoreWeave, which carries $9.7 bn of debt due within a year, could face tighter cash flows if token‑price cuts depress revenue streams【1】.

What to watch

• OpenAI and Anthropic pricing announcements – any formal token‑price revisions in the next quarter.
• Enterprise AI budget caps – further limits from large users such as Uber that could accelerate the shift to open‑source models.
• Custom‑chip adoption rates – the pace at which data‑center operators replace Nvidia GPUs with cheaper silicon, affecting AI hardware demand.

The emerging price war underscores a paradox: token demand is booming even as enterprises demand tighter ROI, forcing the sector’s biggest players to balance growth against shrinking margins and an uncertain path to profitable IPOs.