Anthropic

Anthropic is an AI-safety research company founded in the US in 2021, led by the Amodei siblings — Dario and Daniela — both formerly of OpenAI. Headquartered in San Francisco, it builds the conversational LLM Claude on two foundations: Constitutional AI, in which a written constitution governs an LLM's behaviour, and mechanistic-interpretability research. As of April 2026 Anthropic was reporting an annualised revenue of US$43 billion (close to a fivefold increase since the start of the year); in May 2026 a new round was reported at a US$900 billion post-money valuation, putting it in contention with OpenAI for the title of the most valuable private AI company.

History

Dario Amodei led research at OpenAI as VP of Research, having driven GPT-2 and GPT-3 development, before leaving with COO sister Daniela and five others in late 2020. Anthropic was founded in early 2021 with the explicit mission of "advancing safety research faster than frontier capabilities". A US$124 m Series A followed, attracting OpenAI researchers Jared Kaplan, Tom Brown, and Sam McCandlish — three of the principal authors of the GPT-3 paper.

The Constitutional AI paper appeared in May 2022. Claude 1.0 launched in limited release in March 2023, with Claude 2 in July and Claude 2.1 (200,000-token context) in November. Google invested about US$550 m in 2023; Amazon announced up to US$4 bn in September 2023, completing in March 2024. Claude 3 (Opus / Sonnet / Haiku) shipped in March 2024, Claude 3.5 Sonnet in June, "Computer Use" (a model that can read the screen and operate mouse and keyboard) in October. Claude Opus 4 and Sonnet 4 followed in May 2025, Claude Opus 4.5 in November 2025, and Claude Opus 5 / Sonnet 4.7 across 2026. Claude Code, the developer CLI, went generally available in May 2025, crossed US$1 bn ARR by November 2025, and more than US$2.5 bn ARR by February 2026, on a doubling cadence into 2026.

In November 2024 Anthropic, Palantir, and AWS announced a partnership to provide Claude to the US intelligence and defense community — the central case of the 2024 AI military-use debate. In April 2026 Amazon committed up to a further US$25 bn (on top of US$8 bn already invested, so roughly US$33 bn total), with Anthropic committing over US$100 bn in spend on AWS Trainium and securing up to 5 GW of training capacity. By May 2026 Goldman Sachs, JPMorgan, and Morgan Stanley were reported as candidates to lead an IPO.

Products and Services

Key Events

• January 2021: Dario and Daniela Amodei depart OpenAI; Anthropic founded shortly after.
• April 2022: Series B raises US$580 m, including around US$500 m linked to Sam Bankman-Fried's FTX (the related shares passed to the FTX bankruptcy estate after FTX's collapse).
• March 2023: Claude 1.0 in limited release.
• July 2023: Claude 2 general availability; in September Amazon announces up to US$4 bn.
• March 2024: Claude 3 family (Opus / Sonnet / Haiku).
• June 2024: Claude 3.5 Sonnet; widely reported to surpass GPT-4o on major benchmarks.
• October 2024: Claude gains Computer Use; the Sonnet family is demonstrated driving a desktop.
• November 2024: Palantir-AWS partnership to provide Claude to US intelligence and defense. The updated RSP introduces ASL-3 controls.
• May 2025: Claude Opus 4 / Sonnet 4 launched; Claude Code made generally available.
• November 2025: Claude Opus 4.5.
• February 2026: Series G, US$30 bn at US$380 bn post-money.
• April 2026: Amazon commits up to a further US$25 bn; Anthropic commits over US$100 bn in AWS compute spend.
• May 2026: Reported talks for a round at US$900 bn pre-money; IPO preparations underway.

Key Metrics

• ARR: US$43 bn as of April 2026 (~5x growth in four months)
• Valuation: US$380 bn post-money at Series G (Feb 2026); discussions at US$900 bn-plus by May 2026
• Cumulative investment: about US$3 bn from Google, around US$33 bn (committed) from Amazon, plus Spark Capital, General Catalyst, Lightspeed, and others
• Employees: estimated 2,000-3,000 in 2026
• Claude Code ARR: over US$2.5 bn (February 2026)

Influence and Criticism

Anthropic is unusual among large AI labs in placing safety at the centre of its public identity. Its 2023 Responsible Scaling Policy (RSP) introduced graded AI Safety Levels (ASL) that explicitly tie capability progress to safety mitigations, and that framework has influenced industry and policy discourse. The lab's mechanistic-interpretability work — including sparse-autoencoder feature extraction in frontier models — keeps safety research at academic frontier quality. Yet the founding logic ("safety-focused labs need to be at the frontier") has been criticised as a contributor to overall acceleration of capabilities (the "acceleration paradox").

The November 2024 Palantir-AWS partnership provoked particular debate over whether an avowed AI-safety company should be supplying LLMs to military and intelligence agencies. Anthropic's usage policies prohibit direct use of Claude for nuclear or biological weapons design or for surveillance, while permitting defence analytics and cybersecurity; the guardrails are publicly documented. In 2025-2026 Anthropic has been simultaneously pushing commercial growth and engaging on the EU AI Act GPAI obligations, US AI rule-making (including around Stargate), and the rapid rise of Chinese frontier labs led by DeepSeek.

Litigation has expanded in step with revenue. Universal Music and other rights-holders sued over lyrics in training data (from 2023), publishers have filed copyright cases, and a 2024-2025 lawsuit from Reddit targeted scraping of its corpus — Anthropic now carries the same litigation surface as OpenAI. The depth of the Amazon relationship — equity in exchange for committed AWS compute spend (over US$100 bn, up to 5 GW) — exemplifies the "compute-for-capital" model that has emerged at the frontier; balancing that dependency against research independence is an open problem.