This is a new development in a story we reported earlier: Anthropic prices record $81.3 billion IPO at $83 per share.
Key facts
- Anthropic completed its Nasdaq listing last Thursday, pricing shares at $83 and reaching about $111 before closing the first day at $97.17; the stock fell another 8% to $89.78 on Friday morning.
- Anthropic must reconcile its safety-first brand with public-market demands for profit and revenue growth now that it is a public company.
- The company's compute costs are projected to reach roughly $900 billion in the years ahead.
- Anthropic is not yet profitable by its financials and has a long way to go before inference revenue covers training costs for new models.
The Claude-maker finished its Nasdaq debut last Thursday after setting shares at $83, climbing to roughly $111, and ending the opening day at $97.17 before an 8% decline to $89.78 on Friday morning.
On the first trading session, the stock began trading shortly before noon at $111.50, up about 34 percent, then moved under $100 for much of the day before closing at $97.17. Buyers who paid the $83 IPO price saw about 17 percent gains on day one, but those who jumped in during the first minutes around $110 to $115 lost money that day.
The offering values the Claude-maker at roughly $1.5 trillion and raised just under $82 billion, exceeding SpaceX’s $75 billion June offering. Anthropic is the initial pure AI laboratory to reach public markets, and the record-breaking debut will create a template for other AI companies planning stock offerings. The scale of the offering and valuation places it among the largest listings.
OpenAI and Anthropic filed draft offering documents around the same time last summer, yet only Anthropic moved ahead. OpenAI postponed its offering, and accounts differ on the reason, pointing to market timing and safety concerns over agentic AI, or systems that can act independently to carry out tasks. The company delayed because of safety fears connected to rapidly advancing agentic AI without proper safeguards, tied to the Hugging Face hacking incident, and Chief Executive Sam Altman said his firm should not go public this year and is targeting a 2027 offering. The coverage gives varying reasons for Anthropic going first while OpenAI held back, and sources disagree on the motive.
The postponement leaves Anthropic as the market bellwether and raises the bar for future listings, according to the coverage. The firm must now balance its safety-first identity with public-market demands for profit and revenue growth.
The firm is not yet profitable according to its financials and has a long way to go before inference revenue covers training costs for new models. Inference revenue comes from running models for users, while training costs fund the development of new models. Its compute costs are projected to reach roughly $900 billion in the years ahead, and it leases computing capacity from hyperscalers while covering power and GPU costs. Those suppliers are now major shareholders in the firm, and the projected sum is part of the massive infrastructure cost burden driving revenue pressure.
Analyst Steve Rattner estimates Anthropic needs to grow revenue by about 17 times over the next decade to match returns from some technology rivals. The estimate serves as an external measure of shareholder returns. Earnings pressure intensifies in upcoming quarters, and the firm previously posted one quarter of huge growth followed by weaker growth. The situation amounts to a near-term public-market test after IPO euphoria.
Anthropic’s sales pitch to large enterprise and government customers stresses safe, trustworthy AI and cybersecurity tools, though those customers also require data sovereignty and the ability to switch providers easily. Data sovereignty means customers want control over where their information is stored. That pitch reflects both the enterprise value proposition and its limits. Some small and mid-sized businesses use internal benchmarks to route roughly 10 percent of tasks to Anthropic’s expensive models, 40 percent to open-source models, and 50 percent to cheaper mid-tier models, and they do not share those benchmarks publicly, a routing behavior tied to cost management that erodes premium model usage.
Microsoft CEO Satya Nadella said Microsoft plans to lower token prices and make AI models interchangeable under Copilot, selecting the cheapest suitable option, which squeezes Anthropic’s margins and adds competitive pressure on pricing. Tokens are the chunks of text that models process. Nadella said every company will build its own evaluations, test models against them, and always pick the most efficient one. Microsoft released seven AI models for coding work, cybersecurity and workplace tasks at the Ignite developers conference on Tuesday.
Anthropic leaders say customers will always want the frontier model because competitors will use it, a view some observers call naive or idealistic. The argument is an internal rationale for frontier-model investment despite customer cost sensitivity. The firm is trying to build a moat so customers find it difficult to port their work away from its tools. Customers pushed back hard after the firm raised prices for the same or worse quality, leading to a public statement and a partial price rollback.
The seven founders co-own Founder LLC, which holds one share with 50.1 percent of the firm’s voting power. Of the seven board seats, the Long-Term Benefit Trust picks three, the founders pick three, and the trust technically selects the seventh tie-breaking seat but needs founder approval, leaving founders with largely unchecked control. Anthropic shareholders cannot vote out executives and have little power to influence firm decisions beyond selling or not buying stock. The board allocation mechanics reinforce founder control.
The trust includes four members—Ben Bernanke, Richard Fontaine, Christopher Stubbs, and Neil Buddy Shah—and none has extensive AI experience; the founders appointed them. Former employees said the members were chosen to avoid rocking the boat. The trust can only appoint or remove certain board members and must give 30 days’ notice to a board committee led by founder Daniela Amodei, and it cannot intervene in daily decisions, so its practical power is constrained. Anthropic’s co-founders say they built the firm on a culture of effective altruism.
Anthropic held back the Mythos model earlier this year because it found the system highly capable of uncovering cybersecurity exploits, which serves as an example of its safety-first stance. The firm has repeatedly warned that its products could pose existential risks to humanity, a position some critics see as hype rather than genuine concern. Expensive safety efforts at Anthropic may fall by the wayside under pressure to make money and buy compute, according to the coverage.
The firm’s president, Daniela Amodei, described the firm as unusual and said it cares as much about people who do not use Claude as those who do. Dario Amodei said highly capable AI working with people could compress a century of scientific and medical progress into a decade. Daniela Amodei said the future of AI remains uncertain and that critics are needed to shape what comes next.
Anthropic agreed in 2025 to pay $1.5 billion to authors to settle a class-action lawsuit claiming it pirated books used to train Claude; the case grew out of a copyright suit Anthropic won on fair use grounds, a major copyright matter hanging over the firm. Fair use is a legal doctrine allowing limited use of copyrighted material without permission. Payments under the settlement are expected to begin the week of Nov. 23, with some notifications arriving earlier, although some authors report difficulty certifying their works. Writers should receive about $3,000 per work under the settlement, though the actual amount may be lower after publishers take their cut, and copyright law in the AI era remains a gray area with many ongoing lawsuits even though Meta and Anthropic won key fair use cases in 2025.
Jack Clark, an Anthropic co-founder, had a reported net worth of $19.8B after the IPO.
Nvidia CEO Jensen Huang stepped down, and Nvidia shares fell 14% in a week, cutting roughly $1 trillion of market value. Nvidia interim Chief Executive Colette Kress told investors that everything was business as usual after Huang’s ouster. The Anthropic stock market debut comes during a tumultuous stretch for the AI industry, including Huang’s ouster. The turmoil provides a governance backdrop for Anthropic’s own public-market transition.
Personal finance expert Daniella Flores said buying Anthropic stock without public earnings history is risky, and that first-day prices are driven by hype. She said capital gains tax could run from 10 percent to 37 percent if the stock is sold in less than a year and 10 percent to 20 percent if sold after a year.

Join the discussion