# AI agents drive unpredictable token costs as Anthropic revenue hits $65bn annualised

Digest AI · Business & Funding · published 2026-09-29T20:20:00Z · updated 2026-10-01T06:30:01Z

Canonical: https://digestai.news/story/analyst-reiterates-amazon-buy-rating-on-220b-ai-capex

## Summary

AI agents are rewriting enterprise computing economics by consuming tokens at unpredictable rates, pushing companies like Uber to impose spending caps. Uber exhausted its full-year AI budget in four months and now limits coding tool spend to about $1,500 per month per engineer. An Amazon project ran 860% over budget, costing more than $1 million after background tasks ran for five months.

SemiAnalysis estimates roughly 90% of Anthropic's business now comes from agentic AI. The Financial Times reported Anthropic's annualised revenue surged from $9 billion at end-2025 to $65 billion in July 2026, with backers forecasting over $120 billion by year-end. However, the company's IPO prospectus reportedly shows nearly 25% of 2025 revenue came from just two clients. OpenAI data shows Codex, used by 17% of customers, generates 63% of all output tokens. A study found coding agents consume 1,000 times more tokens than simpler interactions.

Labs are shifting from token-based to outcome-based pricing. OpenAI president Greg Brockman said "price per task is what matters." OpenAI, Salesforce and Sierra are testing service-based models. Meanwhile, cheaper open models from Chinese labs including Alibaba, DeepSeek, Moonshot AI and Zhipu AI are gaining share. Anthropic's new Opus 5.5 will cost 40% less than its predecessor; OpenAI's Sol and Luna models will be 50% cheaper.

## Key points

- Uber exhausted full-year AI budget in four months, now caps coding tool spend at ~$1,500/month per engineer
- Anthropic annualised revenue reached $65bn in July 2026, ~90% from agentic AI per SemiAnalysis
- Coding agents consume ~1,000x more tokens than simple interactions; Codex drives 63% of OpenAI output tokens

## Why it matters

Agentic AI shifts cost risk to customers via usage-based pricing, creating unpredictable bills. Labs face pressure to move to outcome-based pricing as cheaper open models erode margins ahead of anticipated IPOs.

## Sources

1. [SemiAnalysis estimates ~90% of Anthropic's business comes from agentic AI, while sources say nearly 25% of its revenue in 2025 came from just two clients](https://ig.ft.com/ai-tokens?accessToken=zwAAAaD2RgF7kc8jmwPtJzdNudOxLaAd9rZ2wQ.MEYCIQC-ZdVD-VVohnvB-TG5LBZHkrDDkOgK68dXBdNNhMT5YAIhAJYK7OMibZyEV_cPmte5KO0rf6yk9xNdv9-DVm7vvkOC&sharetype=gift&syn-25a6b1a6=1&token=32e5bd39-bd06-404f-981e-2adc8458c6f5) (ig.ft.com, 2026-10-01)
2. [Amazon: Anthropic's $110 Billion Commitment Changes The AI Capex Debate (Rating Reiterated)](https://seekingalpha.com/article/4951077-amazon-anthropics-110-billion-commitment-changes-the-ai-capex-debate) (seekingalpha.com, 2026-09-29)

Part of the developing story: [Investors Rebalance Portfolios Toward AI Leaders](https://digestai.news/thread/ackman-cuts-amazon-boosts-microsoft-and-meta-stakes) (3 stories)

## Cite

Digest AI, "AI agents drive unpredictable token costs as Anthropic revenue hits $65bn annualised", 29 September 2026, https://digestai.news/story/analyst-reiterates-amazon-buy-rating-on-220b-ai-capex

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/analyst-reiterates-amazon-buy-rating-on-220b-ai-capex.json