Anthropic explained: history, products and how it makes money

Update, October 22, 2026: For Anthropic's active copyright cases, including the Canadian publishers' Ontario suit and the October 21, 2026 Concord fair-use hearing, see our copyright roundup.

Update, October 28, 2026: The Pentagon dispute entry below, which says we found no later ruling as of October 9, 2026, is out of date. On August 27, 2026, Judge Lin ruled largely for Anthropic on summary judgment, Taft Law reported; on September 25, 2026, the DC Circuit, in a 2-1 decision, left the Pentagon's supply-chain-risk designation in place under a different law, Nextgov/FCW and Bloomberg Law reported. Anthropic asked the full DC Circuit to rehear the case on October 13, and the Justice Department appealed Judge Lin's ruling to the 9th Circuit on October 26, Bloomberg Law reported. See Anthropic's 40% government discount and where the dispute stands.

Anthropic makes the Claude models and apps. In five years, it has gone from a small group of researchers who left OpenAI to a company valued at $965 billion in its May 28, 2026 funding round. This explainer covers how the company is structured and governed, who has invested and on what terms, what it sells, how fast its revenue is growing according to its own statements, and the criticism it faces. Disclosure: 9to5AI uses Claude to help draft stories (see the AI-use note). We have kept to Anthropic's own filings and posts, plus independent reporting, and attributed all company claims.

Origins and structure

Anthropic was founded in 2021 by former OpenAI employees, including siblings Dario Amodei, now CEO, and Daniela Amodei. It describes itself as "an AI safety and research company" whose goal is to build "reliable, interpretable, and steerable AI systems."

Two structural features set it apart from a typical startup:

• Public benefit corporation (PBC). Anthropic is a Delaware PBC. Its stated public benefit is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity." Under PBC law, directors may weigh that purpose alongside shareholder returns.
• The Long-Term Benefit Trust (LTBT). Announced on September 19, 2023, the trust is "an independent body of five financially disinterested members." It holds a special class of shares, Class T stock, which lets it elect a growing number of directors. Anthropic said the trust "will elect a majority of the board within 4 years." (Update, October 20, 2026: on April 14, 2026, Anthropic said trust-appointed directors already hold a majority, and its trust page now lists three trustees; see Anthropic's Long-Term Benefit Trust explained.) It also has a significant limit: "failsafe" provisions allow stockholders to change the trust without the trustees' consent if "sufficiently large supermajorities" agree.

As of October 8, 2026, Anthropic's company page lists the trustees as Neil Buddy Shah, Richard Fontaine and Ben Bernanke. The board members it lists are Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell and Vas Narasimhan.

Funding and valuations

Last verified: October 8, 2026. All figures come from Anthropic's own announcements.

The Series H includes $15 billion of investment that cloud companies had already committed, $5 billion of it from Amazon. That makes the headline figure larger than the amount of new money raised in that round. The round's co-leads included Capital Group, Coatue, D1, GIC, ICONIQ and XN. Memory makers Micron, Samsung and SK hynix joined as strategic partners.

Four days after the Series H, on June 1, 2026, Anthropic said it had confidentially submitted a draft registration statement on Form S-1 to the US Securities and Exchange Commission for a proposed IPO. The company said the number of shares and the price had not been set and that any offering depends on market conditions. A confidential filing gives Anthropic the option to list once the SEC review is complete; it is not a commitment to go public, and no listing date has been announced as of October 8, 2026.

The cloud partners: investors and suppliers

Anthropic depends on a few very large companies that are both investors and its main suppliers of computing power:

• Amazon. On April 20, 2026, Amazon said it would invest $5 billion immediately and up to $20 billion more later, on top of the $8 billion it had already invested. In return, Anthropic committed to spend more than $100 billion on AWS over ten years. The deal covers up to 5 gigawatts of capacity and Amazon's Trainium chips. Anthropic calls AWS its "primary training and cloud provider for mission-critical workloads."
• Google. On April 6, 2026, Anthropic announced a deal with Google and Broadcom for multiple gigawatts of Google's TPU chips, starting in 2027. It called this its "largest compute commitment to date" and did not disclose a dollar value. The Series H announcement describes this as 5 gigawatts of TPU capacity. Our gigawatt datacenter analysis puts commitments of this size in context.
• Microsoft and NVIDIA. On November 18, 2025, NVIDIA committed up to $10 billion and Microsoft up to $5 billion. Anthropic agreed to buy $30 billion of Azure computing capacity.

The result is that Claude is available on AWS, Google Cloud and Microsoft Azure. Anthropic says it is the first frontier model on all three. The arrangement is also circular. A large share of the money Anthropic raises from these partners is committed to buying their cloud services. That deserves scrutiny when you read headline valuations.

What Anthropic sells

• Claude apps. Free, Pro ($20/month), Max ($100 or $200), Team and Enterprise plans, covering chat, the Cowork agent, Claude in Chrome and Claude Code. Our Claude app explainer breaks down each plan.
• The Claude API. Developers pay per token, either directly or through the three cloud platforms. The flagship Claude Opus 5.5, released on September 22, 2026, costs $4 per million input tokens and $20 per million output tokens. Sonnet 5.5 (September 28) and Haiku 5.5 (October 7) are cheaper. Our guide on which Claude model to use covers the full lineup and prices.
• Claude Code. Anthropic's coding agent. In February 2026, Anthropic said Claude Code alone had passed $2.5 billion in run-rate revenue, with more than half from enterprises. Developers can build their own agents on the same foundations with the Claude Agent SDK (see our Agent SDK guide) and read about the newest Claude Code feature.
• Enterprise. Seat-based and usage-based contracts. In October 2026, Anthropic's newsroom highlighted customers such as Barclays and a $100 million program to train 10,000 engineers.

How fast revenue is growing (by Anthropic's own account)

Anthropic is private and does not publish audited accounts. What it does publish is run-rate revenue: current revenue annualized, typically by multiplying a recent month by 12. Run-rate is not annual revenue, and it says nothing about profit. It is still the figure the company reports at each funding round.

Update, October 13, 2026: Reuters reported in August, citing a source, that Anthropic's run-rate passed $65 billion at the end of July. Anthropic hasn't confirmed it, and Bloomberg reported on October 9 that Anthropic counts some cloud-partner sales gross, so its figures aren't directly comparable with OpenAI's; see our gross vs net explainer.

By Anthropic's account, the number of business customers spending more than $1 million a year rose from about a dozen two years before February 2026, to more than 500 that February, to more than 1,000 by April. We could not find credible figures on profitability or compute costs, so we don't report any.

The Responsible Scaling Policy

Anthropic's main safety commitment is its Responsible Scaling Policy (RSP), first published in September 2023. Version 3.0 followed on February 24, 2026. It ties increasingly strict safeguards ("AI Safety Levels") to model capabilities. Version 3 separates what Anthropic commits to do on its own from what it recommends the whole industry should do. It adds a public, non-binding "Frontier Safety Roadmap" and promises Risk Reports every three to six months, with external review in some cases.

The update was unusually candid. Anthropic admitted "substantial uncertainty about whether [its models] have definitively passed" its own capability thresholds. It said federal safety discussions had not gained "meaningful traction." It also said that meeting the highest safety levels may require collective action across the industry. One reading is that Anthropic has softened its earlier commitments. Anthropic describes the new ones as "difficult but still achievable."

Controversies and criticism

• Copyright. In a class action brought by authors, Bartz v. Anthropic, Judge William Alsup ruled in June 2025 that training on books was fair use, calling it "exceedingly transformative." He ruled that using pirated copies to build Anthropic's central library was not fair use. In September 2025, Anthropic agreed to pay $1.5 billion, about $3,000 per book for an estimated 500,000 books, as NPR reported. On July 20, 2026, Judge Araceli Martínez-Olguín of the US District Court for the Northern District of California granted the settlement final approval, according to the Authors Guild and other authors' organizations. Some rights-holders opted out, and other lawsuits, including one from music publishers, remain open.
• The Pentagon dispute. On February 26, 2026, Dario Amodei said Anthropic "cannot in good conscience" drop two restrictions demanded by the Department of War: on mass domestic surveillance and on fully autonomous weapons. The next day, according to Reason, President Trump ordered federal agencies to stop using Anthropic's technology, with a six-month phase-out. Defense Secretary Pete Hegseth directed that Anthropic be designated a supply-chain risk. Anthropic said it would challenge any such designation in court. On March 26, 2026, US District Judge Rita F. Lin in San Francisco granted Anthropic a preliminary injunction that paused both the supply-chain-risk designation and the directive telling agencies to stop using its technology until the case is decided on the merits, NPR reported. We found no later ruling as of October 9, 2026.
• Export controls and the Fable 5 pause. On June 12, 2026, the US government applied export controls to Claude Fable 5 and Mythos 5. Anthropic had no reliable way to check users' nationality, so it suspended access for everyone. The controls were lifted on June 30, and Anthropic restored access on July 1. Anthropic says the trigger was a safeguard bypass reported by Amazon researchers, and that its fix, a new classifier, now flags more harmless coding requests.
• Distillation. On February 23, 2026, Anthropic accused DeepSeek, Moonshot AI and MiniMax of running campaigns that generated more than 16 million exchanges with Claude through about 24,000 fraudulent accounts, to train their own models. These are Anthropic's allegations.

Open questions

1. Can the trust really constrain the company? The LTBT's power grows over time, but stockholder supermajorities can still amend it. At a $965 billion valuation, investors have a great deal at stake.
2. How durable is the revenue? Run-rate figures show momentum, not profit. Much of the funding comes back to cloud partners as compute spending.
3. Who controls access to the most capable models? Anthropic has limited Mythos to selected partners (Project Glasswing) and, since October 6, 2026, to customers in its Cyber Verification Program, and Fable 5 was paused under export controls. Governments are now directly involved in who can use frontier models.
4. What happens with the government? The Pentagon dispute tests whether a PBC can refuse a major customer on ethical grounds and stay commercially viable.

For this month's funding and enterprise news, see our funding and enterprise news roundup. For the company Anthropic's founders left, read OpenAI explained.