Evidence correction, 14 October 2026: earlier versions of this post cited a JPMorgan research note and several analyst, Datadog and Josh Bersin claims that the named sources do not contain, misstated Anthropic's revenue mix, its $1 million customer milestone, the meaning of its 80x growth figure and Opus 4.8's benchmark, got details of the Uber, Microsoft and Meta reports wrong, described an API reseller's year-over-year comparison as a same-period gap, and contained an arithmetic error in the routing model. Those claims are removed or corrected below.
Anthropic filed for an IPO on June 1. The numbers tell the story.
On June 1, 2026, Anthropic confidentially filed an S-1 with the SEC. The company raised $65 billion in its Series H at a $965 billion post-money valuation, eclipsing OpenAI for the first time. Fortune reported the listing could come as soon as this fall.
Source: TechCrunch, "Anthropic files to go public," June 2026
The revenue trajectory behind the valuation: $9 billion annualized at the end of 2025. $14 billion in February 2026. $19 billion in March. $30 billion in April. $47 billion in May. Anthropic told investors the run rate will exceed $50 billion by the end of July. CEO Dario Amodei said the company planned for 10x annual growth and instead saw 80x on an annualized basis in early 2026.
Source: Bloomberg, "Anthropic Nears $20 Billion Revenue Run Rate Amid Pentagon Feud," March 2026
Source: Anthropic, "Anthropic raises $65B Series H at $965B valuation," May 2026
About 80% of Anthropic's revenue came from business customers as of October 2025, and roughly 70% to 75% comes from pay-per-token API calls, according to Sacra estimates. Over 1,000 business customers spend more than $1 million a year on Claude, up from more than 500 in February. Eight of the Fortune 10 are customers.
Source: Sacra, "Anthropic revenue, valuation & funding"
The revenue is real. The growth is extraordinary. And the source is your token bill.
The revenue model is per-token billing at frontier prices.
Anthropic's pricing as of June 2026:
| Model | Input ($/M tokens) | Output ($/M tokens) |
|---|---|---|
| Claude Opus 4.8 | $5.00 | $25.00 |
| Claude Opus 4.8 Fast | $10.00 | $50.00 |
| Claude Sonnet 4.6 | $3.00 | $15.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 |
Opus 4.8 launched May 28 with better agentic coding scores (64.3% to 69.2% on SWE-bench Pro) and a 3x cheaper fast mode ($10/$50, down from $30/$150 for Opus 4.7). The standard rate stayed the same: $5/$25 per million tokens.
The output price spread between Haiku and Opus is 5x. Between Haiku and Opus Fast, it is 10x. Every API call that hits Opus when Haiku would suffice contributes 5x more to that $47 billion revenue number than it needs to.
This is not a criticism of Anthropic's pricing. Frontier models cost more to build and run. The price reflects the capability. The problem is that most production API calls do not need frontier capability.
Token bills are showing up in earnings.
In May 2026, CNBC reported that Meta, Shopify, Spotify and Pinterest all flagged rising AI and inference costs as a drag on margins this earnings season.
Source: CNBC, "Cheap AI could derail OpenAI and Anthropic's IPOs," May 20, 2026
The pattern is consistent across industries. Enterprise AI budgets were set based on 2024 token rates and chat-era usage patterns. Agentic workloads, coding agents, and production inference at 2026 adoption levels consume multiples of what the spreadsheets projected.
The data points keep arriving:
- Uber's CTO said the company used up its 2026 AI coding tools budget in four months, after 84% of its developers became agentic coding users and Claude Code use rose from 32% to 63% of engineers between December and February. Forbes put average costs at roughly $150 to $250 per engineer per month, and up to $2,000 for heavy users.
- An AI consultant told Axios that one unnamed client spent half a billion dollars on Claude in a single month after not setting usage limits. The claim is unverified.
- Microsoft canceled most Claude Code licenses in its Experiences and Devices division, with usage expected to end by late June, moving engineers to GitHub Copilot CLI. Axios said cost was part of the reason.
- At Meta, one employee built an internal leaderboard called Claudeonomics that tracked token use across more than 85,000 employees.
Source: CFO Dive, "Uber's finance team overtaken by engineering in AI use," May 29, 2026
Source: Axios, "AI sticker shock hits corporate America," May 28, 2026
Derek Thompson called it "The Great AI Cost Panic of 2026." The AI boom has entered its "wait, is this worth it?" phase. The answer is yes, it is worth it, but not at frontier prices for every call.
Source: Derek Thompson, "The AI Boom Has Entered Its 'Wait, Is This Worth It?' Phase," 2026
The problem is not Claude. It is uniform model selection.
Anthropic built a great model. Opus 4.8 is arguably the best coding model available. The 69.2% SWE-bench Pro score, the improved self-review capability, the dynamic workflow features, these represent genuine advances that justify the price for tasks that need them.
The problem is that most production API calls do not need them.
Datadog's State of AI Engineering report found that 69% of input tokens in customer traces were system prompts, meaning internal instructions, policy definitions and tool guidance.
Source: Datadog, "State of AI Engineering," 2026
AI.cc, an API aggregator, said in a May 2026 press release that customers with full three-tier routing reached a median of $2.31 per million tokens, against $18.40 for all-frontier workloads twelve months earlier. That gap mixes routing with a year of falling prices, so it overstates what routing alone does.
The gap is not about using less AI. It is about using the right model for each call. AI.cc says the frontier tier in such setups typically handles 5% to 15% of calls.
What $47 billion in revenue looks like with routing.
Anthropic's $47 billion run rate works out to about $3.9 billion a month across all customers ($47B divided by 12). If 70% to 75% of that is pay-per-token API usage, as Sacra estimates, API spend is roughly $2.7 billion to $2.9 billion a month.
We cannot know the exact prompt-complexity distribution across all of Anthropic's customers, so the split below is an assumption.
Model the blended spend at three tiers:
Without routing (hypothetical: every call goes to Opus): All calls go to Opus at $5/$25. 100% of tokens at frontier price.
With three-tier routing:
- 60% of calls to Haiku at $1/$5: ~80% cost reduction on those calls
- 30% of calls to Sonnet at $3/$15: ~40% cost reduction
- 10% of calls stay on Opus at $5/$25: no change
Blended cost reduction: about 60% (0.6 x 80% + 0.3 x 40% = 48% + 12%).
On roughly $2.8 billion a month of API spend, a 60% cut in this hypothetical would be about $1.7 billion a month ($2.8B x 0.6), or about $20 billion a year. Not from using less AI. Not from switching providers. From routing each call to the cheapest Claude model that can handle it.
Anthropic still gets paid. Every routed request still hits a Claude model. In this hypothetical the token volume stays the same. But the per-request cost matches the per-request complexity.
The IPO creates a structural tension.
Anthropic's business model scales with token volume at frontier prices. The company's revenue growth depends on more customers sending more tokens to Opus and Sonnet. A $965 billion valuation implies continued growth toward $50 billion and beyond in annualized revenue.
Your business model requires controlling costs. You need AI to be productive, not to maximize your cloud provider's revenue. The tension is structural: what is good for Anthropic's IPO prospectus is not necessarily good for your P&L.
This is not unique to Anthropic. OpenAI faces the same dynamic. Google faces it with Gemini. Every token-priced provider benefits from customers using the most expensive model available. The incentives are aligned on usage but misaligned on efficiency.
CNBC reported in May that cheap AI models could actually threaten OpenAI and Anthropic's IPO valuations. If enterprises route to cheaper models, the blended revenue per token falls. This is the honest version of the market: providers want you on Opus, your CFO wants you on Haiku, and routing is the mechanism that resolves the tension by matching the model to the task.
Source: CNBC, "Cheap AI could derail OpenAI and Anthropic's IPOs," May 2026
Josh Bersin argues that the price enterprises pay for AI tools is set to rise sharply.
Five things to do before the Q3 budget review.
1. Audit your model selection. Pull your API logs from the last 30 days. What percentage of calls go to Opus versus Sonnet versus Haiku? If the answer is "mostly Opus" or "we do not know," you are overpaying.
2. Classify your prompt distribution. Tag a sample of 1,000 requests by complexity. Classification, formatting, simple Q&A, and status checks are Haiku-tier tasks. Multi-step reasoning, long-form generation, and novel problem-solving are Opus-tier.
3. Enable prompt caching. Datadog found that only 28% of LLM call spans on cache-capable models showed any cached input tokens, even though 69% of input tokens were system prompts. Anthropic supports automatic prompt caching. Cache writes cost 1.25x the input price and reads cost 0.1x.
4. Route per request, not per application. Stop hardcoding model selection. A classifier that evaluates each call independently and routes to the cheapest capable model captures the 5x pricing gap between Haiku and Opus on the calls Haiku can handle.
5. Plan for agentic growth, not chat-era budgets. Goldman Sachs projects 24x token consumption growth by 2030. Your AI budget should assume volume growth, not just price changes. Routing is the lever that lets volume grow without the bill growing proportionally.
Source: Goldman Sachs, "AI Agents Forecast to Boost Tech Cash Flow as Usage Soars," May 2026
Where Nadir fits.
Nadir can run beside your application's existing Claude gateway and return a model-and-effort decision receipt without calling a provider. Start in shadow mode and attach outcomes before enforcement. The older 60% / 98% RouterBench result is a reference-assisted research ceiling, not deployed performance.
The integration is two lines: change the base URL, set model="auto". Complete non-streaming responses report the served model and actual cost; a configured, priced benchmark adds the avoided-spend comparison. The dashboard's Savings page shows the current month's totals with a daily chart.
Anthropic built a company valued near $1 trillion on token revenue. That revenue comes from your API bill. You do not have to stop using Claude to control the cost. You have to stop sending every call to the most expensive model.
Sources: [Fortune, "Anthropic confidentially files for IPO"](https://fortune.com/2026/06/01/anthropic-confidentially-files-ipo-965-billion-valuation/) (June 2026). [TechCrunch, "Anthropic files to go public"](https://techcrunch.com/2026/06/01/anthropic-files-to-go-public/) (June 2026). [Anthropic, "Series G"](https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation) (February 2026). [Bloomberg, "Anthropic Nears $20 Billion Revenue Run Rate Amid Pentagon Feud"](https://news.bgov.com/artificial-intelligence/anthropic-nears-20-billion-revenue-run-rate-amid-pentagon-feud) (March 2026). [VentureBeat, "Anthropic says it hit a $30 billion revenue run rate"](https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth) (May 2026). [Anthropic, "Series H"](https://www.anthropic.com/news/series-h) (May 2026). [CNBC, "Anthropic tops OpenAI as most valuable AI startup"](https://www.cnbc.com/2026/05/28/anthropic-open-ai-startup-value.html) (May 2026). [Sacra, "Anthropic revenue, valuation & funding"](https://sacra.com/c/anthropic/). [Anthropic, "Pricing"](https://platform.claude.com/docs/en/about-claude/pricing). [VentureBeat, "Claude Opus 4.8 is here with 3X cheaper fast mode"](https://venturebeat.com/technology/anthropics-claude-opus-4-8-is-here-with-3x-cheaper-fast-mode-and-near-mythos-level-alignment) (May 2026). [CNBC, "Cheap AI could derail OpenAI and Anthropic's IPOs"](https://www.cnbc.com/2026/05/20/cheap-ai-could-derail-openai-and-anthropics-ipos.html) (May 2026). [Fortune, "Microsoft reports expose AI's cost problem"](https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/) (May 2026). [CFO Dive, "Uber's finance team overtaken by engineering in AI use"](https://cfodive.com/news/ubers-finance-team-overtaken-engineering-ai-use/821513/) (May 2026). [Axios, "AI sticker shock hits corporate America"](https://www.axios.com/2026/05/28/ai-spending-roi-enterprise-costs) (May 2026). [The Decoder, "Microsoft pulls Claude Code licenses"](https://the-decoder.com/microsoft-pulls-claude-code-licenses-and-pushes-developers-back-toward-its-own-ai-tool/) (May 2026). [The Decoder, "Meta employees compete for token consumption"](https://the-decoder.com/meta-employees-compete-for-token-consumption-on-an-internal-ai-leaderboard/) (April 2026). [Derek Thompson, "The AI Boom Has Entered Its 'Wait, Is This Worth It?' Phase"](https://www.derekthompson.org/p/the-great-ai-cost-panic-of-2026) (May 2026). [Datadog, "State of AI Engineering"](https://www.datadoghq.com/state-of-ai-engineering/) (2026). [AI.cc press release, "AICC Report: Enterprise Token Costs Drop 67% Year-Over-Year"](https://www.einpresswire.com/article/911544568/aicc-report-enterprise-token-costs-drop-67-year-over-year-as-multi-model-ai-adoption-hits-record-high) (May 2026). [Josh Bersin, "AI Prices Are Going Up, Up, Up"](https://joshbersin.com/2026/05/ai-prices-are-going-up-up-up-and-what-this-means-for-enterprise-ai/) (May 2026). [Goldman Sachs, "AI Agents Forecast to Boost Tech Cash Flow"](https://www.goldmansachs.com/insights/articles/ai-agents-forecast-to-boost-tech-cash-flow-as-usage-soars) (May 2026). Anthropic model pricing as of June 2026.