Anthropic

Anthropic is an AI safety company founded in 2021 by Dario Amodei, Daniela Amodei, and several former OpenAI researchers, and is headquartered in San Francisco. Its founding thesis — that capability and safety have to advance together — shapes its research agenda and is the clearest line of difference from the other frontier US labs. The company is backed by a mix of financial and strategic investors, the largest being Amazon, which has committed over 13 billion dollars across rounds, and Google; semiconductor partners Samsung, SK Hynix and Micron joined the most recent round. Anthropic reached its first profitable quarter in 2026, and in October 2026 it made its IPO filing public, seeking a Nasdaq listing under the ticker ANT.

Anthropic is controlled by its Long-Term Benefit Trust, a small body of external trustees who hold no equity in the company yet elect a majority of its board through their own class of shares. The public filing keeps that structure and adds a single founder share that gives the seven co-founders 50.1 percent of shareholder votes with no economic rights, plus a separate share class for employees. Corporate-governance scholars argue the design hard-wires a conflict between profit-seeking investors and self-appointed trustees, and because the trust has so far acted in an advisory capacity, how much constraint it can actually apply remains untested. Anthropic has added outside checks too: independent evaluators embedded inside the company since September 2026, and since October six researchers from the nonprofit METR with the right to publish findings Anthropic cannot veto. Also in October, it announced a Public Benefit Fund to hold 5 percent of its shares, subject to stockholder approval.

The company develops the Claude family of large language models, which runs from a public flagship through an economical near-frontier tier to a balanced workhorse and a fast, inexpensive tier — alongside Claude Mythos, an unrestricted frontier-cybersecurity line held back from general release and deployed instead through a defensive-security partnership with roughly 50 organizations. Beyond the models, Anthropic ships agentic products including Claude Code, Claude Cowork and managed agents, plus vertical lines for legal and small-business customers. It also runs its own wet biology lab in the Bay Area, focused on fundamental biology rather than drug discovery, whose first published result, in September 2026, was an enzyme system with CRISPR-like DNA repeats that Claude agents found and whose function is still unknown. It also offers vetted life-science teams access to models with loosened biology safeguards. It originated constitutional AI, the Model Context Protocol now adopted across the industry, and the Claude Agent SDK. Claude is distributed through a direct API and through Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry.

Anthropic runs on leased capacity rather than data centers it owns, and the financing is as distinctive as the scale. It has committed to a multi-year purchase of Google Cloud capacity worth roughly 200 billion dollars, most of it Google's own tensor processing units, and has signed a run of multi-billion-dollar leases with specialist AI cloud providers on top of that, including its first site in Canada, a Nova Scotia campus due to come online in 2027. The chips are largely leased rather than bought: a private-credit vehicle buys the hardware and leases it back, a structure borrowed from aircraft leasing, so a very large book of chip contracts depends on Anthropic meeting its lease payments. The company has also begun recruiting a custom silicon team to co-design hardware and models, though nothing has been taped out.

Two disputes sit on the company permanently enough to belong in its profile. Anthropic chose contract terms covering autonomous weapons and mass surveillance over US defense access — the first frontier lab to make that trade — and litigated the resulting procurement standoff, winning at the district level while an appeal continues; adoption has meanwhile kept deepening at the intelligence-agency level. Separately, it carries the industry's heaviest copyright-litigation load, having settled Bartz v. Anthropic for 1.5 billion dollars, the largest copyright settlement in US history, on a holding that training on lawfully acquired books is fair use while storing pirated copies is not. Because Anthropic settled rather than appealing, that split never became binding precedent and the settlement granted no forward license, so training-data provenance remains a standing financial risk rather than a settled question.