Anthropic, PBC filed a Form S-1 with the SEC on October 26, 2026 to list Class A shares on the Nasdaq Global Select Market under the ticker ANT. The filing leaves the share count, price range and listing date blank, so there is no IPO date yet. What it does contain is the first audited look at the company’s revenue, losses and compute costs, and those numbers matter to anyone whose agent runs on Claude.
This is not investment advice. We read the S-1 as developers, not as traders.
What did Anthropic disclose?
The headline numbers, taken from the S-1 announcement and coverage by Fortune, Yahoo Finance and Leverage Shares:
| Metric | Figure |
|---|---|
| Revenue, full year 2025 | $4,585M |
| Revenue, H1 2025 | about $1,163M |
| Revenue, H1 2026 | $16,286M |
| Operating loss, H1 2026 | $693M |
| Net loss, H1 2026 | $80,748M |
| Non-cash charge inside that net loss | about $72.4B, tied to valuation increases |
| Q2 2026 compute spend | $5.6B |
| Q2 2026 revenue | $11.6B |
Two cautions. First, the net loss is dominated by an accounting charge, not cash leaving the building. The operating loss of $693M is the number closer to day-to-day economics. Second, outlets disagree on some figures. Reports put the 2025 loss at either $42B or $12.9B depending on how the charge is treated, and cash at either $131.7B (cash plus short-term investments) or $112.4B (cash equivalents only). We are noting the discrepancy rather than picking a side.
The S-1 does not include third-quarter results. The FT and Bloomberg reported on October 14 that Q3 revenue reached $16.8B, which would be more than the entire first half. Treat that as reportedly, since it is outside the filing.
Why does revenue per dollar of compute matter to agent builders?
Divide Q2 revenue by Q2 compute and you get roughly $2.08 of revenue for every $1 spent on compute. That ratio is the single most useful number in the filing for anyone paying per token.
If the ratio were near 1, every price cut would be a loss and prices would have nowhere to go but up. At 2.08, Anthropic has margin to keep cutting token prices while it fights for market share, and public-market investors will want to see that margin hold or widen. The practical reading: expect prices on current-generation models to keep drifting down, but expect the newest and most capable models to launch at a premium and stay there longer.
What this ratio does not tell you is how it splits between the API, Claude Code, and consumer subscriptions. The S-1 reports consolidated figures, so an agent builder cannot tell whether API customers are the high-margin segment or the subsidized one.
What changes when your model vendor goes public?
Going public does not change the API, but it changes the incentives behind it. Four areas to watch.
Pricing stability
Private Anthropic could cut prices as a growth move and absorb the hit. Public Anthropic reports quarterly, and a sudden price cut shows up as a revenue dent that analysts ask about. Cuts will likely still happen, but they may be more predictable and tied to new model launches rather than ad hoc. For builders, predictable is good. Budget planning gets easier even if the floor rises slowly.
Model deprecation cadence
Serving old models costs compute that could serve new ones. A company optimizing the revenue-per-compute ratio has a reason to retire older models faster. Anthropic has published deprecation timelines before, and nothing in the S-1 changes the policy, but the pressure points in one direction. Pin model versions in config, not code, and read deprecation notices the week they arrive.
Enterprise terms
Public companies tend to formalize contracts. Expect clearer SLAs, more standardized data-processing agreements, and more structured volume tiers. Expect less flexibility on one-off deals. If you negotiated custom terms as a startup customer, check whether they survive renewal.
Vendor risk
An S-1 includes a long risk-factors section, and Anthropic’s is reportedly extensive: compute dependence on a small number of cloud providers, regulatory exposure, competition from OpenAI and Google, and the cost of frontier training. Staggered lockups with more than 20 unlock windows over the first year mean a steady drip of insider selling rather than a single cliff. None of that affects an API call today. All of it affects whether the company you depend on looks the same in 2028.
What else happened the same week?
The filing landed alongside a research claim that may matter more in the long run. On October 26, OpenAI said its models had disproved the Exponential Time Hypothesis, a conjecture in computational complexity, and Anthropic posted a similar paper. Outside experts are still reviewing both. If the results hold, they are the strongest evidence yet that frontier models can contribute to open mathematical problems, which is a different kind of capability than writing code or answering support tickets.
Wikipedia lists the IPO as planned for November 2026. Since the S-1 itself gives no date, we would say reportedly targeted.
Keep an abstraction layer
The strongest advice from this filing is the same advice that applied before it: do not hardwire your agent to one model.
A thin layer that routes calls through a single interface, normalizes tool-call formats, and keeps prompts in files rather than code lets you switch providers in days instead of months. Most agent frameworks support this already; the Claude Agent SDK, LangGraph and CrewAI all let you swap the underlying model. Test your eval suite against at least one alternative model every quarter so the switch is a decision rather than a scramble.
This is not a prediction that Anthropic will raise prices or deprecate your model. It is a recognition that a public company answers to shareholders first, and your agent’s uptime is not on the balance sheet.
What to do next
- Read the risk-factors section of the S-1 yourself. It is long, but it is the company’s own list of what could go wrong.
- Audit which model versions your agents call and whether each is pinned in config.
- Price your workloads on at least one competing model and record the delta.
- Watch the Q3 figures when Anthropic files its first quarterly report after listing. The revenue-per-compute ratio is the number to track.
- Subscribe to our newsletter for the follow-up when the price range is set.
If you are new to the space, start with what an AI agent is and our tools directory.
Questions, answered
When is the Anthropic IPO?
No date is set. The S-1 filed on October 26, 2026 leaves the share count, price range and date blank. The listing is reportedly targeted for November 2026, but that is not confirmed in the filing.
What is Anthropic's ticker symbol?
ANT. Anthropic, PBC plans to list Class A shares on the Nasdaq Global Select Market under that ticker, according to the S-1.
Will the Anthropic IPO change Claude API prices?
Nothing in the S-1 announces a price change. The filing shows about $2.08 of revenue per $1 of compute in Q2 2026, which leaves room for prices to keep falling, but a public company also faces pressure to protect margins. Keep an abstraction layer so you can switch models if terms change.
