What Does Anthropic (ANT) Do? The Company Behind Claude, Now Public
Anthropic builds the Claude AI models and sells them by usage and by subscription. What ANT does, how it makes money, why a $104.0B net loss is mostly paper, and who controls the company after its Nasdaq IPO.
Anthropic builds Claude. It trains large AI models in its own research lab, then sells access to them: businesses and developers pay for every token they use, and individuals and teams pay a monthly subscription. That is what the company does. Since 2026-11-19 it also has a ticker, $ANT on the Nasdaq Global Select Market, after an IPO priced at $83.00 a share, above the $70 to $78 range it first marketed.
The standard story is growth: $41M of revenue in 2023, $33.2B in the first nine months of 2026. The other half of the same filing is the bill: about $900.9B of committed spending on compute and infrastructure, and a $104.0B net loss for the nine months. Both are true, and both come from the registration statement. This piece walks through what Anthropic sells, how it makes money, why the loss is mostly paper, who actually controls the company, and what to watch. Figures are from the S-1 and its amendment unless noted; the first trade had not printed when this was written.
The TL;DR. Anthropic sells AI by the token. Revenue grew more than twelve times in a year, operations now bring in cash, and almost every dollar raised is headed for computers. The frame that matters: compute and infrastructure cost about 57 cents of each revenue dollar over nine months (QA's arithmetic), and that ratio decides whether the company makes money.

What does Anthropic do?
Anthropic was founded in 2021 by seven co-founders, former OpenAI researchers led by Dario Amodei (chief executive) and Daniela Amodei (president), and it was incorporated from the start as a public benefit corporation. Its filing calls it first a research lab, with a mission it describes as the responsible development of advanced AI for the long-term benefit of humanity. Before the IPO it had raised $141.1B from investors and partners.
What a customer actually buys is Claude, in three ways:
- The API. Developers and companies call Claude models from their own software and pay per token, the unit of text a model reads and writes.
- Claude Code and Claude Cowork. Claude Code is a coding agent that works inside a developer's tools; the filing puts it at the center of the growth story. Claude Cowork takes the same idea to office work: an agent that reads files and uses applications to finish a task.
- Subscriptions. Monthly plans for individuals and teams on the Claude apps.
The same models are also sold through the marketplaces of the big cloud providers, so a company already buying computing from Amazon, Google or Microsoft can add Claude to the same bill.
How Anthropic makes money
Two lines, very unequal. In the first nine months of 2026, 79% of revenue was consumption (pay for what you use) and 21% subscriptions. The growth sits in consumption:
| Period | Revenue |
|---|---|
| 2023 | $41M |
| 2024 | $386M |
| 2025 | $4.6B |
| First nine months of 2026 | $33.2B |
| Third quarter of 2026 | $17.0B |
The same nine months of 2025 brought in $2.5B. By customer, more than 11,900 spent at least $100K over the last twelve months, against about 950 a year earlier, and net revenue retention was about 900% on the company's own definition (business customers only; consumer subscribers are excluded). No single customer was above 10% of revenue, but the top five were about 25%, and 44% of revenue came through cloud marketplaces, run by companies that are also suppliers and, in Google's case, a named competitor.
The filing also quotes a run-rate revenue of about $81B at 2026-10-31, the last 28 days of usage scaled to a year plus one month of subscriptions times twelve. The company itself says it is not a forecast.
Where it sits
On QuantAbundance, $ANT sits in the AI Applications and Agentic AI themes: it is the model and the agent, where most listed AI names are the chips, the power or the data centers underneath. That makes it the customer of much of the AI infrastructure trade. Its compute comes from Google (tensor processing units), Amazon, Microsoft and SpaceXAI, and the concurrent $26.0B private placement at the IPO price came from Google, $NVDA Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC. For the years before the listing, when the only way in was through those partners, see How to invest in Anthropic: the proxy basket.
The numbers
- Operating result. A $1.3B operating loss on $33.2B of revenue over nine months. The second quarter showed an operating profit of $974M; the third slipped back to a $567M loss, which the filing ties to about $5B of expensive short-term compute and $2B of non-cash charges for an employee donation matching program.
- The net loss. $104.0B over nine months, of which $94.3B is the change in fair value of convertible notes, derivatives and SAFEs. As the valuation rose, what Anthropic owed those holders in shares rose too. It shows up later as dilution, not as cash leaving the bank.
- Cash. Operating cash flow of $6.6B in the first nine months of 2026, after outflows in every earlier year. Cash of $95.6B and short-term investments of $36.3B at 2026-09-30, before the IPO money.
- The compute bill. $18.8B of compute and infrastructure cost over nine months, about 57 cents per revenue dollar (QA's arithmetic). Committed spending of about $900.9B, mostly from 2027 to 2045, with about $70.7B due in the next twelve months.
- The deal. 980 million Class A shares at $83.00, about $81B before the underwriters' option (QA's arithmetic), plus the $26.0B private placement.
The bull case
- Revenue up more than twelve times year on year, with compute cost growing much more slowly in the third quarter (revenue up 1,127%, compute and infrastructure up 382%).
- Operations that now generate cash, and a balance sheet of $95.6B in cash plus $36.3B of short-term investments before the IPO.
- A customer base that spends more every year: about 900% net revenue retention and more than 11,900 customers above $100K.
- Claude Code and agents give a product people pay for by the hour of work saved, the kind of usage that scales with consumption pricing.
The bear case
- The commitments. About $900.9B of spending is due whatever revenue does. A slowdown does not shrink the bill.
- Control. One Class F share held by the founders' company carries 50.1% of the vote, and the Long-Term Benefit Trust elects a majority of the board. Public Class A holders own economics, not control.
- Growth will slow, by the company's own words. The filing says growth rates are not indicative of the future and expects them to decline; price cuts already offset about $600M of revenue from the second to the third quarter.
- Partners who are rivals. Amazon holds 14.7% and Google 13.7% of capital stock in non-voting shares; both sell compute to Anthropic, and Google is named next to OpenAI as a main competitor.
- Supply and law. Options, restricted shares, note conversions and public benefit share issuances add stock over time, lock-ups expire from 180 to 365 days, and the company paid a $1.5B copyright settlement in 2025 while the law on model training stays unsettled.
How to access
$ANT trades on the Nasdaq Global Select Market. A newly listed stock is usually not in index funds at first, so in QA's data no ETF holds it yet; until it qualifies, the direct listing is the only clean exposure. To trade it from a US-retail account alongside the other names on the site, see /stack/ibkr. The full teardown video and the live page are on /stocks/ant; rule-based alerts on the name are part of /pro.
What to watch
- The first quarterly report as a public company. Revenue against the October run rate, and whether the third quarter's operating loss turns back into a profit.
- Compute per revenue dollar. About 48 cents in the second quarter, when Anthropic made money, about 58 cents in the third, when it did not (QA's arithmetic). It is the profit story in one ratio.
- Lock-up calendar. The first insider release comes after 180 days, or with the first quarterly results of 2027 if earlier; the co-founders and largest holders wait up to 365 days.
- Observable reference levels. With no trading history yet, the deal sets them: the $83.00 IPO price, the $78 top and the $70 bottom of the first marketed range, and the first opening print. These are observable reference levels, not trade levels or targets. The live level map is on /stocks/ant.
Live data on this ticker: /stocks/ant. Price, filings, the level map, bot positions.
QuantAbundance is educational research. Nothing here is investment advice. See /disclosures.
Frequently asked questions
- What does Anthropic do?
- Anthropic is an AI research company and public benefit corporation that builds Claude, a family of large AI models, and sells access to them. Businesses and developers pay for what they use (metered in tokens) through its API, Claude Code and the big cloud marketplaces; individuals and teams pay monthly subscriptions. Its S-1 says 79% of revenue in the first nine months of 2026 was consumption and 21% subscriptions. It listed on the Nasdaq Global Select Market as ANT on 2026-11-19.
- How does Anthropic make money?
- Mostly from usage. Customers pay per token for Claude models through the API, Claude Code and cloud partners' marketplaces (44% of revenue in the first nine months of 2026 came through those marketplaces), plus subscription plans. Revenue was $41M in 2023, $386M in 2024, $4.6B in 2025 and $33.2B in the first nine months of 2026, with $17.0B in the third quarter alone.
- Is Anthropic profitable?
- Not on a GAAP basis yet, but closer than the headline suggests. The first nine months of 2026 show a $104.0B net loss, of which $94.3B is the rising fair value of convertible notes, derivatives and SAFEs as the valuation climbed: an accounting charge, not cash spent. The operating loss was $1.3B on $33.2B of revenue, and operating cash flow was positive at $6.6B.
- Who controls Anthropic after the IPO?
- Not the public shareholders. Class A shares carry one vote each; Amazon's and Google's Class N shares carry none; one Class F share held by a company owned by the co-founders carries 50.1% of the vote; and the Long-Term Benefit Trust's single Class T share elects a majority of the board. Anthropic is a controlled company and a public benefit corporation.
The desk behind this piece
Some of these are affiliate relationships. See disclosures.
Related research
- 8 min readAfter SpaceX: the 2026 IPO pipeline, ranked by what's actually filed
- 7 min readHow to invest in Anthropic - the proxy basket (AMZN, GOOGL, VCX, ARKV)
- 8 min readTeraWulf (WULF): what it does, how it makes money, and why a bitcoin miner became an AI landlord
- 7 min readHow to invest in Databricks - the public investors are real but the stakes are too small to matter
Go deeper
Want more of these in your Google results? Add QuantAbundance as a preferred source.
Add to Preferred SourcesGet the daily digest.
One email a day: the AI-trade watchlist, every new ticker teardown and the bubble shifts. Free, leave in one click.
Free when you join: the $HOOD Teardown Dossier (Robinhood Markets), the full desk file we sell in the shop. One per new reader.
No spam. One email per day max, unsubscribe in one click. Pro ($49/mo) includes every dossier and report in the shop.