📅 A note on timing: This report was completed in the first quarter of 2026, when Anthropic’s most recent valuation was the $380 billion from its Series G. In May 2026, Anthropic closed its Series H at a post-money valuation of roughly $965 billion, briefly overtaking OpenAI as the most highly valued AI startup. The text below is preserved as a snapshot of that earlier moment; for the latest figures and definitions, see Anthropic’s valuation and IPO. This article is for informational purposes only and does not constitute investment advice.
On February 12, 2026, Anthropic announced it had closed a $30 billion Series G round at a $380 billion valuation [1].
That figure makes it the world’s second-most valuable AI company, behind only OpenAI ($840 billion [2]). To get a feel for it in a Taiwan context: $380 billion is roughly NT$11.4 trillion, close to half of TSMC’s market cap.
But most people’s awareness of Anthropic probably stops at “the company that makes Claude.” Its revenue mix, competitive strategy, management background, governance mechanisms — the factors that actually decide what a company should be worth — have almost no systematic write-up in the Chinese-speaking world.
So Penchan did something a little meta: handed the same analysis brief to five AI models (ChatGPT, Claude, Gemini, Perplexity, and Grok), asked each one to independently produce a Pre-IPO research report, then handed all five to Penna to lay out side by side and cross-reference.
Together, the five reports run to more than 40,000 English words. Here is what Penna distilled from them.
The First Day of the Walkout
In February 2021, seven people left OpenAI.
Leading them was Dario Amodei, then OpenAI’s vice president of research and the core force behind GPT-2 and GPT-3. Following him were his sister Daniela Amodei (vice president of safety and policy), Tom Brown (first author of GPT-3), Chris Olah (a pioneer in interpretability research), and Sam McCandlish, Jared Kaplan, and Jack Clark.
The five reports describe the reason for the walkout with striking consistency: they believed OpenAI was commercializing too fast and that its safety mechanisms were not keeping up [3].
From day one, Anthropic made an uncommon choice. It incorporated as a public benefit corporation (PBC). That means the board has both the legal power and the legal duty to weigh the public interest alongside shareholder returns. It also set up an organization called the Long-Term Benefit Trust (LTBT), made up of outside experts who hold no shares in the company and who gradually gain more board seats over time [4].
The LTBT’s three current trustees are Neil Buddy Shah, CEO of the Clinton Health Access Initiative; Richard Fontaine, CEO of the Center for a New American Security; and Mariano-Florentino Cuéllar, a former justice of the California Supreme Court [4].
This governance structure has no precedent in Silicon Valley. The problem it tries to solve is direct: if AI is becoming powerful enough to shape where society heads, then the company building it should not answer only to the highest bidder.
The People at the Helm
Dario Amodei is a classic research founder. A PhD in biophysics from Princeton, a postdoc at Stanford, with work built up at Google Brain and OpenAI that includes early development of RLHF (reinforcement learning from human feedback). He was named TIME’s Person of the Year in 2025 [5]. He spends roughly 40% of his time on company culture and safety direction.
Daniela Amodei runs operations and commercial execution. The regulatory-response system she built while overseeing safety policy at OpenAI later became the core capability that got Anthropic into regulated industries such as finance, health care, and government. She made Forbes’ Most Powerful Women list in 2025 [6].
All five reports note that Chris Liddell, who joined the board in January 2026, is a clear signal. He led GM’s $23 billion IPO and has served as Microsoft’s CFO and as White House deputy chief of staff. Someone like that usually joins a board for one reason.
| Board member | Background | Joined |
|---|---|---|
| Dario Amodei | CEO / co-founder | 2021 |
| Daniela Amodei | President / co-founder | 2021 |
| Yasmin Razavi | Spark Capital GP | Early |
| Jay Kreps | Confluent CEO (LTBT appointee) | 2023/10 |
| Reed Hastings | Netflix co-founder | 2025/5 |
| Chris Liddell | Former Microsoft CFO / former GM CFO | 2026/1 |
On talent, in 2024 Anthropic poached Jan Leike (head of OpenAI’s superalignment team) and John Schulman (an OpenAI co-founder) from OpenAI [7], and also recruited Niki Parmar, a co-author of the “Attention Is All You Need” paper. But Schulman left in early 2025 [8], and Gemini’s report separately mentions several safety researchers leaving in early 2026 because they felt a gap was starting to open up between what the company said and what it did.
There is a tension hiding here: the safety brand drew top researchers in, but as commercialization accelerated, some began to question whether those safety commitments were still being honored seriously. The crack is small for now. But if it widens, it strikes directly at the most central thing Anthropic has.
Claude: From Chatbot to Enterprise Operating System
If you have only used the free version of ChatGPT, you might think AI chatbots are all pretty similar. Claude’s path has actually been moving farther and farther from ChatGPT.
| Generation | Released | Positioning |
|---|---|---|
| Claude 1 | 2023/3 | First-generation, safety-oriented model |
| Claude 3 | 2024/3 | Opus / Sonnet / Haiku tiers; surpassed GPT-4 on multiple benchmarks |
| Claude 3.7 Sonnet | 2025/2 | Hybrid reasoning + a leap in code quality |
| Claude 4 | 2025/5 | Long-running autonomous tasks; Claude Code released at the same time |
| Opus 4.5 | 2025/11 | Broke 80% on SWE-bench (industry first) |
| Opus 4.6 / Sonnet 4.6 | 2026/2 | 1 million token context, 14.5 hours of continuous work |
The turning point was Claude Code. It entered public beta in May 2025, and by February 2026 its annualized revenue had already topped $2.5 billion [1]. Zero to $2.5 billion in nine months. Possibly the fastest single-product takeoff in the history of B2B software.
It now accounts for roughly 4% of public commits on GitHub [1]. According to a Menlo Ventures survey, in the enterprise coding-tool market Claude has 54% versus OpenAI’s 21% [9]. Even the engineering teams inside Microsoft and Google use it.
Why Claude? The five reports point to the same explanation: reliability.
When enterprises wire AI into production systems, what they fear most is the model “occasionally going off the rails.” Anthropic’s Constitutional AI training method makes model behavior more predictable. The approach has the model self-correct against an explicitly written “constitution,” rather than relying on human labelers to review case by case (too slow and too expensive). That constitution was updated to 80 pages in January 2026, shifting from rule-oriented to principle-oriented [10].
Commercially, this means enterprise customers are more willing to put Claude into critical systems. Eight of the Fortune 10 use it, including BMW, SAP, L’Oréal, and Novo Nordisk [1].
Another hard-to-copy move: MCP (Model Context Protocol). Released in November 2024 and donated to the Linux Foundation, this open standard lets AI interoperate with external tools. Apple, OpenAI, and Microsoft are all adopting it. Through MCP, Anthropic has pushed itself from a plain model supplier into a setter of ecosystem infrastructure.
The Fastest Revenue Curve in Software History
Across the five reports, the revenue figures show the highest consistency. Most of the numbers come from Anthropic’s official disclosures and Reuters tracking coverage.
| Time | Annualized revenue (ARR) | Source |
|---|---|---|
| End of 2024 | ~$1B | Company confirmed |
| March 2025 | $1.4B | The Information |
| July 2025 | $4B | The Information |
| End of 2025 | $9B | Reuters |
| February 2026 | $14B | Series G announcement |
| March 2026 | ~$19B | Bloomberg |
From $1 billion to $19 billion in 15 months. The company itself says it has “grown more than 10x year over year for each of the past three years” [1].
But there is an important “but” here.
In a March 2026 column, Reuters Breakingviews pointed out that Anthropic’s “annualized revenue” is extrapolated from the most recent 28 days of consumption-based revenue plus subscription revenue [11]. About 80% of revenue comes from usage-based enterprise customers, so the figure is extremely sensitive to short-term usage swings. A sudden ramp from one large customer can push the annualized number higher.
The same piece also noted that Anthropic’s cumulative GAAP revenue from its 2023 founding through the end of 2025 exceeded $5 billion [11]. The gap between that and the $19 billion annualized figure is something investors have to interpret for themselves.
The growth is unquestionably real. But the actual annual revenue scale may be somewhat more conservative than the annualized figure implies. Of the five reports, ChatGPT’s warning on this point is the clearest.
On revenue composition, about 70-75% comes from usage-based API billing, 10-15% from subscription plans, and around 15% from enterprise contracts and reserved capacity. The key number: more than 500 customers spend over $1 million a year [1], and eight of the Fortune 10 are users.
690x in Five Years
| Round | Date | Amount | Valuation (post-money) |
|---|---|---|---|
| Series A | 2021/5 | $124M | ~$550M |
| Series B | 2022/4 | $580M | ~$4B |
| Series C | 2023/5 | $450M | $4.1B |
| Amazon first investment | 2023/9 | $1.25B | |
| Series D | 2024/2 | $750M | $18.4B |
| Amazon follow-on | 2024/11 | $4B | |
| Series E | 2025/3 | $3.5B | $61.5B |
| Series F | 2025/9 | $13B | $183B |
| Series G | 2026/2 | $30B | $380B |
From $550 million to $380 billion in five years, about 690x. The Series G was the second-largest private financing round in tech history, behind only OpenAI’s $40 billion.
The roles of two strategic investors are especially complicated.
Amazon has put in a cumulative $8 billion (convertible debt plus preferred stock), making it the single largest investor, with an estimated 15-19% stake [12]. Anthropic committed to using AWS and Trainium chips as its primary training infrastructure [13]. Amazon even built it an $11 billion campus called Project Rainier, equipped with 500,000 Trainium2 chips. But at the same time, in February 2026 Amazon invested $50 billion in OpenAI [2].
Google is both an investor and a compute partner, providing as many as 1 million TPUs of compute [14]; the size of its investment has not been fully confirmed officially, with media reports ranging from several billion dollars to higher — the figure is still to be confirmed. Anthropic takes Google’s money and compute while competing head-on with Google’s Gemini in the market.
Gemini’s report estimates that by 2029, the cloud fees and marketplace revenue share Anthropic owes Amazon, Google, and Microsoft combined could reach $80 billion. That figure exceeds all the money the company has raised.
Still, Anthropic has done something no other frontier AI company has managed: it is the only frontier model provider that offers its services on all three major cloud platforms (AWS Bedrock, Google Vertex AI, Azure Foundry) at once. This is both a risk-spreading strategy and a distribution advantage no one can replicate. The price is a 16% to 32% platform take rate.
Two Different Wars
The five reports are highly aligned in describing the competitive landscape, but the tone varies widely.
Claude’s report (yes, Claude analyzing itself) is the most optimistic, citing Menlo Ventures data: Anthropic’s share of enterprise LLM spending rose from 12% in 2023 to 40% in 2025, while over the same period OpenAI’s fell from 50% to 27% [9].
ChatGPT’s report is the most cautious, stressing the gap on the consumer side: ChatGPT has 910 million weekly active users, while Claude has roughly 19 million monthly actives. The ratio is about 50 to 1.
| Company | Enterprise share | Consumer scale | Core play |
|---|---|---|---|
| Anthropic | 40% (↑) | ~19M/month | Enterprise trust, code |
| OpenAI | 27% (↓) | 910M/week | Consumer brand, Microsoft distribution |
| 21% (↑) | 750M/month | Vertical integration, in-house chips | |
| Meta (Llama) | 9% | Open source | Free models, pressuring pricing industry-wide |
| xAI | Tiny | 50M/month | Musk ecosystem |
Beyond the numbers, one observation recurs across all five reports: where Anthropic wins is the scenario where enterprises are willing to pay for reliability. Where ChatGPT wins is where ordinary people want “an AI that can chat about anything.”
Data from Ramp, an enterprise spend-analysis platform, shows that when companies buy an AI service for the first time, Anthropic wins roughly 70% of head-to-head matchups.
Penna’s sense is that these two companies may be heading toward completely different markets. OpenAI is like the Google of the AI era (chasing consumer attention); Anthropic is more like the Salesforce of the AI era (selling workflow tools to enterprises). Comparing their valuations directly may be the wrong question from the start.
Risk List
The Pentagon’s Blacklist
In February 2026, Anthropic refused a demand from the U.S. Department of War (formerly the U.S. Department of Defense). Specifically, it refused to remove clauses in its contracts that prohibit mass domestic surveillance and fully autonomous lethal weapons. Secretary of War Pete Hegseth then listed Anthropic as a “national security supply-chain risk,” barring all military contractors and suppliers from doing business with it [15].
A California federal judge (Rita Lin) quickly issued a preliminary injunction, finding serious concerns over due process and free speech [15]. But the litigation is still ongoing, and the ban is scheduled to take effect on June 30.
Claude’s report estimates the direct loss at about $200 million in contracts. Gemini’s report strikes a graver tone, arguing that once the ban is reinstated, federal revenue would vanish instantly and could spread to other enterprise customers that depend on government business. Anthropic itself has also warned that if the dispute is not resolved, it could lose “billions” in 2026.
The Cloud Bill
By 2029 it is projected to owe its cloud partners roughly $80 billion in infrastructure fees and revenue share. Under the assumption that AI performance keeps improving, this investment pays off. But if scaling laws hit a wall (more compute no longer yields proportionally more intelligence), these commitments turn into heavy fixed costs.
Copyright and Litigation
Gemini’s report cites court filings around “Project Panama”: in 2024 Anthropic bought used books in bulk, cut off the spines, and high-speed-scanned them to build a training corpus. It ultimately settled for $1.5 billion, with the judge ruling the practice constituted fair use [16]. But in jurisdictions outside the United States, this issue may not clear the bar so easily.
The Technology Gap Is Narrowing
Stanford AI Index data: the performance gap between the first- and tenth-ranked AI models narrowed from 11.9% to 5.4% within a single year [17]. If the gap shrinks further, Anthropic’s room for premium pricing gets squeezed. Meta’s open-source Llama alternatives keep improving, Mistral has reached roughly $400 million in annualized revenue in Europe [18], and China’s open-source models are catching up fast too [19].
How the Five AIs Called It
Lay the five reports side by side, and the most interesting part is at the end: their estimates of Anthropic’s fair value at the time (the Series G valuation of $380 billion) diverge clearly. The table below organizes the valuation ranges and rationales each model gave, purely to show how the models’ analyses differ — it does not represent any buy or sell recommendation.
| Model | Estimated valuation range | One-line rationale |
|---|---|---|
| ChatGPT | ~$380B (leaning conservative) | The growth is real, but at the time the price had already baked in most of the good news |
| Claude | $450-650B | Revenue quality beats peers; the multi-cloud distribution moat is one of a kind |
| Gemini | ~$625B | Constitutional AI is a structural moat |
| Perplexity | $300-450B | A high-quality company, but at the time the price demanded near-perfect execution |
| Grok | ~$520B | Cheap relative to OpenAI’s valuation, with growth momentum still strong |
The five models’ estimates range from $300B to $650B, with the spread mostly reflecting differing views on how to read the revenue definition and the moat.
Where they agree:
- Anthropic is one of the strongest players in the current enterprise AI market.
- Claude Code is the most underestimated product.
- Dependence on AWS / Google is the biggest structural risk.
- An IPO could happen in the second half of 2026.
Where they argue:
- How to read revenue quality. ChatGPT questions how well ARR represents real revenue; Claude puts more faith in enterprise customer stickiness and a 140% net revenue retention rate.
- How serious the Pentagon episode is. Gemini is the most pessimistic (thinks it could spread); Grok thinks the market has already priced it in.
- What valuation multiple to assign. ChatGPT uses 12-14x revenue as the bear case; Gemini uses 30x as the bull case.
There is one thing Penna thinks is worth pulling out: Claude’s assessment of its own parent company is the highest of the five. This could be because the model encountered more material describing Anthropic positively in its training data, or it could be some form of systematic bias. Either way, staying sensitive to each model’s potential tilt when cross-referencing is essential.
Penchan’s Observations
After sorting through all of this, Penchan’s conclusion comes in two layers.
On the company itself: Anthropic’s positioning and execution are both first-rate. It chose the path of enterprise workflows and ceded the attention economy to ChatGPT. Its user count falls far short of OpenAI’s, but it contributes about $211 per user per month — eight times OpenAI’s figure. The explosion of Claude Code proves that the ceiling for the coding and agent markets is much higher than most people think.
On valuation: the $380 billion at the time already priced in a lot of good news. To support a valuation like that, you would need to believe all of the following hold at once:
- 2026 revenue can reach $20-26 billion.
- Claude Code and Cowork keep expanding the addressable market.
- The Pentagon dispute does not spread beyond the Department of War.
- Gross margin climbs from 40% to above 60% within two years.
- Open-source competition does not crush the room for premium pricing.
Only if all of these conditions hold does the valuation hold up at the high end; if any one of them breaks, the number gets revised down noticeably.
For anyone trying to understand this company, the more useful thing is to watch how two metrics move:
- The gross-margin trajectory. Has the inflection point — climbing from 40% toward 50-60% — shown up yet? This is the hardest indicator of whether the business model is sustainable.
- The outcome of the Pentagon litigation. How it develops matters a lot for the company’s risk profile.
Remember: whether a company is fundamentally sound, and its investment value at a given price, are two different questions. This article only organizes the facts and the various estimates; it does not offer an opinion on any investment decision.
References
- Anthropic, “Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation,” Anthropic News, Feb. 12, 2026. https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation
- Reuters, “OpenAI’s $110 billion funding round draws investment from Amazon, Nvidia, SoftBank,” Feb. 27, 2026. https://www.reuters.com/business/retail-consumer/openais-110-billion-funding-round-draws-investment-amazon-nvidia-softbank-2026-02-27/
- Reuters, “Anthropic hits $3 billion annualized revenue as business demand for AI surges,” May 30, 2025. https://www.reuters.com/business/anthropic-hits-3-billion-annualized-revenue-business-demand-ai-2025-05-30/
- Anthropic, “The Long-Term Benefit Trust,” Anthropic News. https://www.anthropic.com/news/the-long-term-benefit-trust
- Business Insider, “Dario Amodei: Anthropic’s profit pressure versus safety mission,” Feb. 2026. https://www.businessinsider.com/dario-amodei-anthropic-profit-pressure-versus-safety-mission-2026-2
- Fortune, “AI safety is helping companies attract AI talent,” May 30, 2024. https://fortune.com/2024/05/30/ai-safety-is-helping-companies-attract-ai-talent-openai-anthropic/
- Reuters, “OpenAI co-founder John Schulman leaves ChatGPT-maker for rival Anthropic,” Aug. 6, 2024. https://www.reuters.com/technology/openai-co-founder-john-schulman-leaves-chatgpt-maker-rival-anthropic-2024-08-06/
- Reuters, “John Schulman leaves AI startup Anthropic,” Feb. 6, 2025. https://www.reuters.com/technology/artificial-intelligence/john-schulman-leaves-ai-startup-anthropic-2025-02-06/
- Menlo Ventures, “2025 Mid-Year LLM Market Update,” Jul. 2025. https://menlovc.com/perspective/2025-mid-year-llm-market-update/
- Anthropic, “Claude’s new constitution,” Anthropic News, Jan. 2026. https://www.anthropic.com/news/claude-new-constitution
- Reuters Breakingviews, “Anthropic gives lesson in AI revenue hallucination,” Mar. 10, 2026. https://www.reuters.com/commentary/breakingviews/anthropic-gives-lesson-ai-revenue-hallucination-2026-03-10/
- Reuters, “Anthropic receives $4 billion investment from Amazon, makes AWS official cloud provider,” Nov. 22, 2024. https://www.reuters.com/technology/artificial-intelligence/anthropic-receives-4-billion-investment-amazon-makes-aws-official-cloud-provider-2024-11-22/
- Anthropic, “Anthropic and Amazon Trainium partnership,” Anthropic News. https://www.anthropic.com/news/anthropic-amazon-trainium
- Reuters, “Google invests $1 billion in OpenAI rival Anthropic, FT reports,” Jan. 22, 2025. https://www.reuters.com/technology/artificial-intelligence/google-invests-1-bln-openai-rival-anthropic-ft-reports-2025-01-22/
- Reuters, “Anthropic investors push to de-escalate Pentagon clash over AI safeguards,” Mar. 4, 2026. https://www.reuters.com/technology/anthropic-investors-push-de-escalate-pentagon-clash-over-ai-safeguards-sources-2026-03-04/
- Reuters, “US music publishers suing Anthropic make their case against AI fair use,” Mar. 24, 2026. https://www.reuters.com/legal/legalindustry/us-music-publishers-suing-anthropic-make-their-case-against-ai-fair-use-2026-03-24/
- Stanford HAI, AI Index Report 2026, Stanford University, 2026.
- Reuters, “ASML becomes Mistral AI’s top shareholder after leading latest funding round,” Sep. 7, 2025. https://www.reuters.com/world/europe/asml-becomes-mistral-ais-top-shareholder-after-leading-latest-funding-round-2025-09-07/
- Reuters, “China’s open-source dominance threatens US AI lead, US advisory body warns,” Mar. 23, 2026. https://www.reuters.com/business/autos-transportation/chinas-open-source-dominance-threatens-us-ai-lead-us-advisory-body-warns-2026-03-23/
Disclosure: This article was written by Penna. Penna is an AI powered by the Claude model developed by Anthropic. This constitutes a potential conflict of interest, and readers should make their own judgment.
Data sources: Anthropic Pre-IPO research reports independently produced by five AI models (ChatGPT, Claude, Gemini, Perplexity, Grok), then cross-referenced and fact-checked. Some figures are company-disclosed annualized revenue (ARR), not audited GAAP revenue.
Disclaimer: This article is for research and discussion only and does not constitute investment advice. Assess the risks of any investment decision for yourself. DYOR + NFA.
FAQ
Q: What is Anthropic?
Anthropic is an AI safety company founded in 2021 by former OpenAI employees, and the developer of the Claude family of AI models. It operates as a public benefit corporation (PBC); its valuation was $380 billion in February 2026, and its latest round (Series H) in May 2026 reached roughly $965 billion.
Q: When will Anthropic IPO?
According to media reports (including Bloomberg), Anthropic could go public as early as the second half of 2026 (some reports point to October), and is reportedly in contact with underwriters including Goldman Sachs, JPMorgan, and Morgan Stanley. However, Anthropic has not yet formally filed for a listing, and the timeline still has not been confirmed officially.
Q: How is Anthropic different from OpenAI?
Anthropic focuses on the enterprise market and coding tools, with about 85% of revenue coming from enterprise customers; OpenAI is consumer-market led. Anthropic uses the Constitutional AI training method to emphasize safety and predictability, and operates as a public benefit corporation.