You have probably used Claude in the last week. You have probably not thought once about the company that builds it. That company is Anthropic, and on 1 June 2026 it confidentially filed a draft S-1 with the U.S. Securities and Exchange Commission at a private valuation of USD 965 billion.

Anthropic was founded in 2021 in San Francisco by seven former OpenAI researchers, with siblings Dario and Daniela Amodei at the head. The group left OpenAI before the launch of ChatGPT, citing differences over the parent company’s direction on AI safety. Dario, OpenAI’s former VP of Research, is CEO; Daniela, previously OpenAI’s VP of Safety and Policy, is President. The remaining co-founders (Chris Olah, Jack Clark, Jared Kaplan, Sam McCandlish and Tom Brown) anchor the technical leadership. Each of the seven is now worth approximately USD 8 billion on paper following the latest funding round. The first Claude model launched in March 2023, the family has since expanded across the Opus, Sonnet and Haiku tiers, and Anthropic topped CNBC’s 2026 Disruptor 50 list as the most disruptive private company in the United States.

What Anthropic actually sells is access to Claude. The model family is split into three tiers by capability and cost: Opus is the top-tier model used for complex reasoning, long-horizon tasks and agentic workflows; Sonnet is the workhorse that carries the bulk of production traffic, balancing capability against speed and cost; Haiku is the fast, cheap tier built for high-volume use cases like customer-support routing or summary generation. All three are distributed through three channels. The largest, by revenue, is the API: developers and enterprises pay per token consumed, with no minimum commitment. The second channel is direct enterprise contracts, fixed-rate reserved-capacity deals for guaranteed throughput, structurally the highest-margin layer of the business. The third channel is consumer subscriptions: Claude Pro at USD 20 per month and Claude Max at USD 100 to 200 per month, which give individuals priority access to the same models through Claude.ai. The most important productised layer on top of the raw API is Claude Code, a command-line and IDE agent that lets developers delegate coding tasks directly to Claude; it crossed USD 1 billion in annualised revenue in November 2025 and USD 2.5 billion by February 2026, more than doubling in roughly three months, and is the single fastest-growing product line in the company.

The selling point that sets Claude apart is concentrated in three places. The first is coding. By independent enterprise survey data, Claude holds between 42% and 54% of enterprise coding workloads against roughly 21% for OpenAI, and Claude Code is the most-used agentic coding tool in production. The second is reliability and integration for regulated industries: the combination of Public Benefit Corporation status, the Long-Term Benefit Trust, and a steady cadence of published safety research has made Anthropic the default choice for customers in financial services, legal, defence and healthcare, where governance and explainability are screened at procurement. The third is long-context and reasoning performance: extended-thinking mode and long context windows (recent generations exceed one million tokens) are the two capabilities most cited by enterprise customers when they displace alternative models for document analysis and agentic workflows. The combined effect is visible in the customer base. Over 1,000 business customers spend more than USD 1 million annually on Claude, eight of the Fortune 10 deploy it, and total business customers exceed 300,000. The single largest workload across the customer base is software engineering, with financial-services document analysis, legal research, customer-support automation and government deployments (Anthropic is available on AWS GovCloud) rounding out the next tiers.

The governance architecture is the part of this S-1 that has no real public-market template. Anthropic is incorporated as a Delaware Public Benefit Corporation (for-profit corporation legally required to pursue a specific public benefit alongside traditional business objectives). Above the standard equity layer sits the Long-Term Benefit Trust (legal structure designed to ensure sustained benefits over time, either for public purposes in corporate governance or for personal financial protection), an independent five-member body holding a special class of voting stock (Class T common). The Trust’s board-election rights grow over time, eventually allowing it to appoint a majority of the company’s directors. The structure is closer in spirit to Alphabet’s dual-class governance than to a founder-controlled company, with one important difference: the controlling counterparty is a mission-focused independent body rather than an individual. For public-market investors that is an explicit, codified subordination of voting power, and one of the unresolved questions ahead of pricing.

The revenue trajectory is the cleanest expression of how quickly enterprise demand for frontier AI has materialised. The annualised run-rate moved from roughly USD 87 million in January 2024 to USD 1 billion at the end of 2024, USD 9 billion at the end of 2025, USD 30 billion in early Q2 2026, and USD 47 billion by late May 2026. Dario Amodei told CNBC the business grew “80-fold” in the first quarter alone. Approximately 85% of revenue comes from business customers, the inverse of OpenAI’s consumer-dominated mix. In revenue terms it is a B2B infrastructure business rather than a consumer brand.

Anthropic Annualised Revenue Run-Rate (Jan 2024 – May 2026, USDbn)

Source: Bloomberg, Anthropic, CNBC, InterCapital Research

The cost side of the business runs almost entirely through three compute relationships, each disclosed by the company or its counterparties. The largest is with Amazon (April 2026): Amazon committed an additional USD 25 billion of equity in Anthropic (USD 5 billion upfront, up to USD 20 billion tied to milestones), on top of USD 8 billion previously invested, and Anthropic in return committed to spending more than USD 100 billion on AWS over 10 years, secured by up to 5 GW of AWS Trainium2 and Trainium3 capacity. The second, also from April 2026, is with Google and Broadcom: Broadcom will design and supply approximately 3.5 GW of Google TPU capacity to Anthropic from 2027 through 2031. The third, disclosed in SpaceX’s May 2026 S-1, is a USD 1.25 billion per month contract through May 2029 for access to SpaceX’s Colossus and Colossus II clusters, USD 15 billion per year and up to USD 45 billion over the term, with a 90-day exit clause on either side. Together these three agreements represent approximately USD 145 billion of disclosed compute commitments across three chip architectures (AWS Trainium, Google TPU and Nvidia-based COLOSSUS). The diversification is genuine supply optionality, but it also turns the income statement into a straightforward race between pay-per-token revenue growth and a contractually fixed compute bill.

 Anthropic Disclosed Compute Commitments by Counterparty

Anthropic Disclosed Compute Commitments by Counterparty_copy

Source: Anthropic, Amazon, Bloomberg, SpaceX EDGAR S-1, InterCapital Research

Anthropic has progressed through one of the steepest funding sequences in private-market history. Cumulative private capital raised stands at approximately USD 133 billion per Forge Global. The most relevant recent waypoints are the Series F in September 2025 (USD 13 billion raised at USD 183 billion post-money), the Series G in February 2026 (USD 29.2 billion across two tranches at USD 380 billion), two corporate rounds in April 2026 (USD 5 billion from Amazon and USD 10 billion from Google, both at USD 380 billion), and the Series H closed on 28 May 2026 (USD 65 billion across four tranches at USD 965 billion). The Series H lead tranche priced at USD 589.01 per share against a Series G price of USD 259.14, a 2.27-times step-up in roughly three months. The Series H was led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, with a long co-lead list including Capital Group, Coatue, GIC, ICONIQ, Fidelity, General Catalyst, Blackstone, Brookfield, T. Rowe Price and Temasek. This was the round at which Anthropic eclipsed OpenAI on private valuation for the first time.

Anthropic Private Funding History (Sep 2025 – May 2026, USDbn)

Source: Forge Global, Anthropic, Bloomberg, InterCapital Research

The mechanics behind the USD 965 billion mark are worth a footnote. The headline valuation translates to roughly 20 times forward revenue on the USD 47 billion run-rate, around half the implied multiple OpenAI would carry at a USD 1 trillion IPO mark on its lower run-rate. Investor demand in the run-up to the round was extraordinary: secondary-market trades in Anthropic shares were reportedly clearing at an implied USD 1 trillion valuation, with USD 960 billion lots being snapped up within hours in the weeks before the primary round priced. The company took the unusual step of publicly warning investors against certain secondary platforms offering access to its shares. The valuation, in other words, is anchored less in a textbook discounted-cash-flow exercise than in a clearing-price observation from the secondary market: at the current growth rate, that is where institutional investors who could not get into the primary round were willing to transact.

Anthropic’s listing is the second of three frontier AI-economy IPOs converging on the market in 2026. SpaceX filed its S-1 on 20 May 2026 at a press-rumoured USD 1.75 trillion valuation. Anthropic filed on 1 June 2026. OpenAI is widely reported to be preparing its own filing, after closing a USD 122 billion funding round at USD 852 billion in March 2026 (the largest single private financing on record), with Amazon contributing USD 50 billion and Nvidia and SoftBank USD 30 billion each, and may target a valuation closer to USD 1 trillion in the listing itself per Reuters reporting. On the underlying businesses, the comparison is sharper. OpenAI is generating approximately USD 2 billion per month in revenue (USD 24 billion annualised) and projecting roughly USD 30 billion run-rate by year-end, while internal projections place it on track for around USD 14 billion of losses in 2026. Anthropic, by contrast, has been guiding toward break-even by 2028, a path management can credibly underwrite because the gross-margin profile on Claude API usage inflects sharply once contracted compute capacity is in place. Whether the public market prices that earlier path to profitability into Anthropic’s multiple is one of the open questions ahead of pricing.

2026 AI-Economy IPO Snapshot-Valuation, Run-Rate, Revenue Multiple

2026 AI-Economy IPO Snapshot-Valuation, Run-Rate, Revenue Multiple_copy

Source: Bloomberg, CNBC, Yahoo Finance, SpaceX EDGAR S-1, InterCapital Research

Anthropic is filing into a market that is, for the first time in a decade, being asked to absorb three multi-hundred-billion-dollar listings inside a single year. The business arriving on the public market is, in many ways, the cleanest enterprise-AI story available: roughly 85% business revenue, more than 1,000 customers spending over USD 1 million annually, three structural compute partnerships across three chip architectures, a clear product moat in coding, and a guided path to profitability ahead of its largest competitor. The complications, namely the Long-Term Benefit Trust governance layer, the magnitude of the contractually fixed compute bill, and the absence of historical comparables for software businesses growing at this rate, will be the questions the prospectus has to answer once the SEC review is complete.