For a company that brands itself as the cautious one in frontier AI, Anthropic spent the last twelve months doing some decidedly uncautious things: shipping models into the United States government's classified networks, deploying custom variants for national security customers, and, according to a statement from co-founder Dario Amodei, becoming the first frontier lab to put production AI inside the Department of War for intelligence analysis, modeling, simulation, and cyber operations [Anthropic]. That posture, safety-forward in research, aggressive in deployment, is the bet that has carried the San Francisco company from a 2021 spinout of ex-OpenAI staff to a reported $380 billion post-money valuation [Crunchbase, 2026].
The core product is Claude, the family of large language models the company sells through a consumer app, a Team and Enterprise plan, and an API for developers. The most recent flagship, Claude Opus 4.5, shipped November 24, 2025 and is positioned by the company as its best model for coding, agents, and computer use [Anthropic, 2025]. Claude Sonnet 4 and Opus 4 had landed earlier that year on May 22 [Wikipedia]. Claude Code, the company's developer surface, is now bundled into Team and Enterprise plans with premium seats and admin controls [Anthropic], a packaging move aimed squarely at engineering organizations that have been routing spend toward GitHub Copilot and Cursor.
The traction behind those products is, by the standards of enterprise software, extraordinary. CEO Dario Amodei has confirmed annualized revenue exceeding $19 billion [Yahoo Finance, 2026]. DemandSage reports more than 300,000 business customers and agencies on the Claude Enterprise API and roughly 18.9 million monthly active users on the consumer side [DemandSage, 2026]. Headcount, depending on the source and the snapshot date, sits somewhere between roughly 1,100 in April 2026 [jobsbyculture.com, 2026] and 3,909 in September 2025 [Revelio Labs, 2025], a spread wide enough that the truth likely depends on whether contractors and research affiliates are being counted.
The bet
Anthropic's wedge is a particular kind of trust. The company was founded in 2021 by Dario and Daniela Amodei, siblings who left OpenAI in late 2020 over concerns about that company's commercial direction and AI safety priorities [Observer, 2024]. They incorporated Anthropic as a Public Benefit Corporation, with a stated purpose of responsible development of advanced AI for the long-term benefit of humanity [Anthropic]. That framing, plus a research output that includes interpretability work and the publicly posted Claude's Constitution [Anthropic], is what large regulated buyers, including governments, point to when they justify routing sensitive workloads through Claude rather than a competitor.
The national security push is the clearest expression of the strategy. Anthropic says it was the first frontier lab to deploy in classified US networks and at the National Laboratories, and it is a core partner in the Genesis Mission, a multi-billion-dollar initiative to accelerate American science with AI [Anthropic]. There is also a Claude for Life Sciences program working with pharma, biotech, and research institutions [Anthropic]. The pitch, in plain terms: if you are a buyer who cannot tolerate a model that goes off-script, the safety research is not a cost center, it is the product.
Why it could be big
The capital behind this thesis is now historic in scale. Anthropic closed a $13 billion Series F in September 2025 led by Iconiq [TechCrunch, 2025], then a $30 billion Series G on February 12, 2026 led by GIC and Coatue [GIC, 2026], the latter described by Crunchbase as the second-largest venture funding deal of all time [Crunchbase, 2026]. Earlier rounds included a $450 million Series C in May 2023 led by Spark Capital [TechCrunch, 2023]. The cap table now includes Google, Amazon, Nvidia, Fidelity, Menlo Ventures, and Lightspeed alongside the more recent leads.
Series C May 2023 | 450 | $M
Series F Sep 2025 | 13000 | $M
Series G Feb 2026 | 30000 | $M
Valuation has tracked the same curve, climbing from $61 billion in early 2025 [Bloomberg, 2025] to $183 billion later that year [Crunchbase] to a reported $380 billion post-money on the Series G [Crunchbase, 2026]. At roughly 20 times annualized revenue, the multiple is rich but not anomalous for a frontier lab growing this quickly, and it implies investors are underwriting both continued model leadership and a durable enterprise business on top of it.
The team and traction
Dario Amodei serves as CEO and Daniela Amodei as president. Both came out of OpenAI, where Dario was VP of research, and the founding cohort pulled in a meaningful share of that lab's safety-oriented staff. The culture has been described in reporting as ideologically committed, with The Information noting internal sessions referred to as Dario Vision Quests [The Information]. Whatever one makes of the texture, the output is concrete: a model family that consistently ranks at or near the top on coding benchmarks, a 200K token context window introduced with Claude 2.1 [Anthropic], and an enterprise motion serving more than 300,000 business customers [DemandSage, 2026]. Open roles currently include a Policy Outreach Lead for APAC, a Privacy Operations Specialist, and an Engineering Manager for LLM Scaling [Lever.co, 2026], a hiring mix that reads like a company simultaneously building infrastructure and getting ready for a much larger regulatory surface.
The honest counterfactual
What bears point to is concentration risk, both in compute and in narrative. Amodei used a January 2026 Davos appearance to publicly criticize Nvidia [TechCrunch, 2026], and the company has been in an escalating public dispute with OpenAI over messaging around military AI deployments, with Amodei calling OpenAI's framing of a Pentagon deal straight up lies [TechCrunch, 2026]. Separately, Amodei has been reported to be trying to avoid a deposition in an OpenAI copyright lawsuit [TechCrunch, 2025]. The risk for bears is that a safety-branded company gets pulled into the same public-trust battles as its less safety-branded competitors, eroding the differentiation. What bulls answer, with evidence, is that the deployments themselves keep landing: classified networks, National Labs, Genesis Mission, life sciences partnerships, all of which are harder to win than a press cycle is to lose.
What to watch
The next twelve months will test three things. First, whether Claude Opus 4.5 and its successors hold coding and agent leadership against GPT-class and Gemini-class releases. Second, whether the Pentagon relationship clarifies into a named, sized contract rather than a public argument over messaging [TechCrunch, 2026]. Third, whether the $30 billion raised in February 2026 [Orrick, 2026] funds a path to operating profitability at the current revenue trajectory, or whether another round at a still-higher valuation arrives first. For a company whose founders left OpenAI specifically to slow down, Anthropic is now setting the pace it once worried about. The interesting question is no longer whether the safety brand can scale. It is what the safety brand looks like at $19 billion in revenue and counting.