Produced by | Huxiu Tech Team
Author | Liang Karl
Editor | Miao Zhengqing
Cover image | Visual China
These 7 people had the last laugh
On November 19, Eastern Time, Anthropic closed at $97.17 per share on its first day of trading, up 17.07% from its offering price. Its closing market capitalization was about $1.28 trillion, after briefly reaching $2 trillion intraday. (Huxiu note: Calculated using the total number of shares after the offering and concurrent private placement, as stated in the final prospectus; excludes the over-allotment option.)
At last, Anthropic, the AI giant founded less than six years ago, has become a publicly traded company after raising about $81.3 billion in the largest IPO in history. Its shares are listed on the Nasdaq under the ticker symbol “ANT.”
At 10:30 p.m. Beijing time on November 19, after regular trading began, the indicative opening price was revised upward several times during the delayed opening auction. The first trade came just over two hours later at $111.50, up 34.3% from the $83 offering price.
The IPO was seen as a major test for the AI industry. Anthropic expanded the offering, and its final price came in above expectations, giving investors a sense of relief. Based on fully diluted shares, the opening price pushed Anthropic’s market capitalization to just over $2 trillion, about $500 billion more than its valuation at the offering price.
However, since Anthropic began trading, the Nasdaq Composite has continued to fall, and tech stocks broadly gave back their early gains. The seven tech giants participating in the concurrent private placement also saw volatility. Anthropic closed its first trading day at $97.17, up 17.07%. Its opening price was the high for the day, and its low was $92.72.
Rewind to November 13, at JPMorgan Chase’s Manhattan headquarters. Anthropic CEO Dario Amodei sat across from JPMorgan Chase CEO Jamie Dimon. A livestreamed fireside chat with current and prospective investors was one of the final stops on his pre-IPO roadshow.
Asked why the company was going public now, Dario Amodei replied, “I think the biggest reason comes back to these bets on computing capacity.” He said there were tremendous opportunities in biology, robotics, cybersecurity, and knowledge work more broadly, and Anthropic wanted to ensure it invested enough to seize them.
As for the timing, he said he was not in a particular hurry to go public in any specific month or on any specific day, or under any particular market conditions. Before starting the process, he said, he had asked many people for advice. The one piece of advice that resonated most was, “Never let your company be worth even a dollar more than it really is.”
He added that this was much like one of Anthropic’s internal principles: “Underpromise and overdeliver.”
Before ringing the opening bell, Dario Amodei gave a speech of about five minutes. His tone was more forceful than it had been on the roadshow. He said, “Anthropic has pioneered several approaches to understanding and guiding AI behavior so that it remains under human control. But there is still a lot of work to do, and the hardest safety work is still ahead of us.”
“We will control the pace at which we train and deploy AI systems, taking as long as necessary until we are confident they are safe and will remain under human control.”
A higher price, more shares, and a greenshoe: how did the stock open like this?
Controversy over AI valuations has not gone away, yet Anthropic still dared to go against the grain by raising its price and increasing the size of its offering, while institutions were willing to compete for shares.
An experienced secondary-market analyst told Huxiu that, in this unusual situation, external regulatory pressure had not held Anthropic back. It has responded proactively to safety concerns. It had also locked in additional computing capacity through long-term agreements in anticipation of approval requirements and grid constraints. At the same time, institutions continue to see growth potential in models and applications, backed by a code market worth hundreds of billions of dollars and a collaborative-work market worth trillions.
An analyst at another brokerage shared a similar view: “The short-term bottleneck to AI commercialization is currently on the supply side. Demand has not actually contracted.” As computing capacity becomes more available, frontier models still have room to accelerate further.
Anthropic’s offering structure also supported first-day trading. The offering increased to 980 million shares, or about 7.4% of the approximately 13.2 billion total shares outstanding. The offering price was also above the top of the original price range, leaving room for trading in the early days after listing. In addition, the underwriters hold an over-allotment option for 147 million shares, which may be used alongside their price-stabilization activities.
The brokerage analyst cited above told Huxiu, “Increasing the size of the offering rather than simply raising the offering price helps stabilize the share base.” Compared with continuing to push up the offering price, issuing more shares provides a larger public float and can help ease the supply-demand imbalance in the early days of trading.
Of course, the strength of these measures will ultimately depend on how high a valuation the market is willing to give Anthropic. The key is whether Anthropic can build a lasting competitive moat.
On the enterprise market, Dai Kun, vice president and chief analyst at research firm Forrester, is less optimistic. He told Huxiu that in enterprise API spending, Anthropic is already in the top tier, if not in the lead, with OpenAI close behind. Google is catching up quickly through platform innovation, while Microsoft has Azure as an entry point for enterprise deployment and distribution.
But being ahead does not mean having a moat. Dai Kun believes Claude’s advantages in fields such as finance, law, and healthcare mainly come from model capabilities, safety mechanisms, and enterprise trust. “But competitors can catch up with these capabilities.” Large enterprises also need far more than access to a single model. They need platform capabilities such as data management, knowledge infrastructure, context engineering, and agentic workflow orchestration. For now, Anthropic has a first-mover advantage.
Sun Zhenya, a research manager at third-party firm IDC, also noted that large language model applications in different vertical industries are still concentrated mainly in coding and have yet to penetrate core enterprise workflows. To serve these business scenarios, models need access to internal company data, business rules, and logic—things that general-purpose model providers or agents themselves cannot solve. What they need is context.
The price set by the capital markets reflects investors’ assessment of Anthropic’s current capabilities and leading position. The question is how long that lead can last.
980 million shares, $83, and those who got on board at the last minute
With the IPO complete, thousands of Anthropic employees and investors are about to become enormously wealthy. To understand this wealth-creation story, we first need to look at where the company came from.
In January 2021, Dario Amodei and his sister Daniela Amodei founded Anthropic with five other people. All seven shared one background: nearly all of them had worked at OpenAI.
The prospectus details each of their backgrounds. Dario Amodei was formerly OpenAI’s vice president of research, while Daniela Amodei held roles including vice president of safety and policy and vice president of people. In other words, Anthropic’s original team came largely from OpenAI’s core research and management ranks. The prospectus describes them as “a group of researchers responsible for defining many of the breakthroughs of this generation.”
As for why they left OpenAI to start a new company, Dario Amodei shared a personal story at the opening bell ceremony.
“Daniela and I lost our father to a disease that became curable just a few years later,” he said. He began his career in biological research. Working in a lab, he realized that scientists face a vast, dense jungle of complexity, and that humanity’s continual expansion of the boundaries of knowledge depends on a range of tools and ideas—“they come from thought, from intelligence.”
From its founding through its IPO, Anthropic has consistently emphasized the long-term value AI could bring. But Dario Amodei has not shied away from how far the technology still is from truly transforming science and medicine. “I don’t want to get too far ahead of ourselves at this point. What we have so far are only early results.”
The company’s fundraising and valuation, however, tell a more straightforward story of growth. From a $124 million Series A in May 2021 to a $65 billion Series H in May 2026, its fundraising has grown exponentially and its valuation has soared. In just five months, it even rose from a $965 billion valuation at the time of its Series H to a peak of $2 trillion.
According to the prospectus, Anthropic issued 980 million shares at $83 each, for a total offering of approximately $81.34 billion. The company issued about 956.4 million new shares; the remaining approximately 23.65 million were sold by seven founders, who will receive the proceeds. Underwriters also received a greenshoe option for 147 million shares, equivalent to 15% of the 980 million shares in the base offering.
In addition to the public offering, another private placement worth $26 billion merits attention. Subscribers included Google, Nvidia, AMD, Broadcom, Micron, Microsoft, and TSMC. Google and Nvidia each subscribed for $10 billion. What is interesting about these seven companies is that they are all chip suppliers or cloud service providers to Anthropic.
For these companies, investing in Anthropic is both a financial investment and a stake in future computing capacity and cloud services business.
The risk factors section of the prospectus also discloses this relationship. Amazon, Google, and Microsoft are simultaneously Anthropic investors, customers, distribution partners, and competitors; Broadcom is both a supplier and a lessor. Meanwhile, the share of Anthropic’s revenue generated through cloud-provider channels has risen from 11% to 44%, and those providers account for 52% of its related accounts receivable.
This means Anthropic’s business growth is now closely intertwined with the interests of these big tech companies. They benefit from Anthropic’s growth while also providing the infrastructure it needs to develop.
Before the IPO was priced, two subscription announcements drew attention. The board and related parties indicated non-binding interest in subscribing for up to $305 million of shares. TSMC, meanwhile, held a board meeting on November 10 and approved a resolution to subscribe for up to $500 million. The former was a non-binding indication of interest, while the latter was approved by the board; they are not the same in nature, but both took place before final pricing.
For investors buying new shares at $83, the prospectus also disclosed a less flattering figure: pro forma tangible book value was just $9.70 per share, implying immediate dilution of $73.30 per share. This is the gap between book value and the offering price after the new shares are issued. It is also a reminder that buyers are paying primarily for expectations of the company’s future growth, not the book value of its existing assets.
For Anthropic employees, the IPO presents a very different picture. According to estimates by Hill, after the listing, 598 employees without founder backgrounds will hold shares worth at least $100 million, while about 2,800 employees may hold shares worth at least $10 million. By comparison, SpaceX’s IPO this June created about 4,400 millionaires.
Employees who joined the company early in junior roles often end up with greater financial returns than those who joined in senior roles just before the IPO. The same applies to Anthropic. Early employees’ options had an average exercise price of just 52 cents, while public-market investors paid $83—a net increase of about 159 times.
One company, two worlds.
One stock, repricing the entire AI trade
Anthropic’s IPO could affect more than just the company itself. The market is likely to use its trading price as a valuation benchmark for OpenAI and other AI companies.
“OpenAI’s fundraising and eventual IPO valuation will very likely be anchored to Anthropic,” a senior secondary-market analyst cited above told Huxiu. “This gives the entire industry a new pricing benchmark.” She expects the early days of trading may see a capital-siphoning effect similar to the one during SpaceX’s IPO.
However, Anthropic chose to go public at a difficult moment. The market is sharply divided in its views on AI. On November 12, news that Nvidia’s board had accepted Jensen Huang’s resignation sent shockwaves through the market. Nvidia later reported earnings that beat expectations and forecast quarterly revenue of $130 billion, yet its stock still fell 3% that day, down about 30% from its October high.
The market is now worried not only about whether AI demand can last, but also about when massive capital spending will translate into actual returns. Anthropic’s $83 offering price was set against this market sentiment.
The analyst believes the buyers are mainly institutions and retail investors who had not received enough shares in the earlier allocation, and that the 15% greenshoe option will provide some support. Still, the stock may not keep rising. “There is a real possibility it will gradually drift lower or even fall below the IPO price before year-end.” ARR (annual recurring revenue) figures will be repeatedly checked against disclosures from third parties and the company. If growth falls short of expectations, that could drag down the share price.
In her view, management has already guided the market’s expectations for Anthropic’s year-end revenue. “It’s unlikely ARR will be much higher than $81 billion by year-end. It might reach $90 billion, but the valuation boost driven by ARR will be limited.”
The market reaction to Anthropic’s IPO was soon reflected in the share prices of other AI companies.
After news broke that the offering price was above the top of the previous range, Nasdaq 100 futures rose 523 points before the market opened, up about 1.9%; S&P 500 futures gained 1.12%, and Dow futures rose 0.71%. The Euro Stoxx 600, which had fallen for three consecutive days, also turned higher.
Publicly traded companies with investment ties to Anthropic also rose. Micron gained more than 4%, AMD rose 3.4%, Google gained 2.9%, Broadcom rose 2.8%, Amazon gained 2%, Nvidia rose 1.7%, and Microsoft gained 1.6%. Anthropic had yet to begin trading, but the share prices of related companies had already had a run-up.
But the market’s direction quickly changed after U.S. stocks opened on IPO day. Although all three major indexes initially rose, the Nasdaq’s gain later narrowed to 0.2%, the S&P turned down 0.3%, and the Dow fell as much as 537 points. Nvidia’s gain narrowed to about 1%, while Micron, which had been up more than 4% premarket, fell back to around 1%.
It is worth noting that when the broader market started falling, Anthropic had not yet recorded its first trade. This reflected ongoing investor concerns about interest rates and returns on AI capital spending. The optimism generated by the IPO simply did not change the market’s underlying divisions.
The share prices of the seven private-placement investors did not move in tandem throughout the day, with AMD posting the steepest decline. AMD subscribed for $2.5 billion worth of shares. Its stock rose as much as 4.7% early in the day, but quickly retreated after short seller Hunterbrook published a report, falling more than 11% at its worst intraday point. Though not directly related to the IPO, this coincidence amplified concerns about sentiment across the “AI computing chain.”
Another easily overlooked change is that, starting this quarter, fluctuations in Anthropic’s share price will also affect some of the book gains reported by Amazon and Google. A Huatai Securities research report published on November 18 estimated that, based on the midpoint of the offering range, every $1 rise or fall in Anthropic’s share price would affect the two companies’ earnings per share by approximately $0.12 and $0.11, respectively.
Therefore, a first-day gain only shows that the market is willing to buy at that price. The real question this IPO must answer is whether Anthropic can maintain its current valuation as index funds gradually enter the market and lock-up shares begin trading. For the post-IPO share price, absorbing the supply of newly unlocked shares matters more than the first-day excitement.
The truth behind going public despite $104 billion in losses
Anthropic’s financials look alarming because of its net loss, but its computing commitments deserve even more attention.
The prospectus reports a net loss of $104.039 billion for the first nine months of 2026. Of this, $94.324 billion came from “changes in the fair value of convertible notes, derivative liabilities, and SAFEs,” and does not mean the company actually spent an equivalent amount of cash. Because the valuation of these liabilities is linked to the company’s valuation, a rise in valuation can actually increase this portion of its book losses. The operating loss for the same period was $1.26 billion.
The real issue is that long-term commitments related to computing capacity and infrastructure total $900.9 billion, of which $70.7 billion is due within the next 12 months. Meanwhile, Anthropic’s ARR stood at $81 billion as of October 31. This figure is calculated by annualizing revenue from a given period; it is not actual revenue.
Revenue is growing rapidly, but the pressure from spending on computing contracts is also substantial. Whether future revenue can keep pace with investment in computing capacity will directly affect Anthropic’s business.
From an industry perspective, Dai Kun told Huxiu, “Over the next few years, only a handful of leading companies may gradually be able to achieve coverage.” The key is not just whether revenue can keep growing, but whether that growth can outpace the costs of training, inference, and model iteration over the long term. Improvements in inference efficiency and chip value for money can lower unit costs, but competition among model providers will also continue to push model prices down.
However, an optimistic IT-industry analyst told Huxiu that, looking only at Anthropic and using the figures disclosed in its prospectus, it can maintain a gross margin above 50% as long as model revenue grows by more than 20%. “It’s already proven that the model works.”
In the fireside chat mentioned above, Dario Amodei also addressed this issue. He called computing capacity “the deepest business challenge facing the company.” Buy too little, and the company may be unable to meet customer demand or train the next generation of models. Buy too much, and it could overextend itself financially. Anthropic’s current strategy is to “lean toward buying a little less,” then add more capacity if demand exceeds expectations.
This approach can reduce the burden of idle computing capacity, but it depends on the company being able to secure additional resources quickly when demand rises. The funds raised in the IPO give Anthropic more flexibility in its procurement strategy. The optimistic analyst cited above told Huxiu that without this financing, Anthropic would definitely face a funding gap in the next four or five years.
In his view, the cost of equity is far lower than the cost of debt. “Whether it uses the money to repay debt or for something else, it’s actually beneficial.” However, this assessment is based more on an industry perspective than a financial one.
But whether it is to broaden investor participation or to prepare for the next few years, Anthropic needs more external capital. For a company investing heavily in computing capacity while continuing to train new models, current revenue alone is unlikely to cover its expansion plans in full.
Some clues about how the funds will be spent can be found in Anthropic’s recent product moves.
On November 10, Anthropic launched Claude Sheets, Whiteboards, Meetings, and Data, and added more than 9,000 connectors, starting to bring Claude into the workflows of more non-technical employees. In the past, Anthropic’s enterprise revenue was more closely tied to developers and IT departments. Now, the company wants finance, operations, marketing, HR, and other departments to become paying users as well.
However, analysts generally believe office software itself may not become the biggest driver of token consumption in the future. Sun Zhenya said office workflows will be an important entry point for everyday workplace interactions with AI, encouraging frequent use. But for now, coding still accounts for more than 70% of enterprise token spending.
On large language model companies building office suites, Dai Kun believes the key is not necessarily that the business model has been validated, but that office workflows combine frequent use, enterprise data, business processes, and payment channels, making them one of the right environments for agents to do real work. Going forward, the boundaries between applications are more likely to fade. Models will call on different data sources and tools to complete tasks, while traditional formats such as documents and spreadsheets will serve collaboration, auditing, and final delivery.
As for the next use case capable of generating revenue at scale, Dai Kun is more optimistic about professional knowledge work and business process transformation. Demand for AI in industries such as finance, law, and healthcare is not limited to generating a report or answering a question; it also involves embedding models into end-to-end processes such as research, review, compliance, and decision-making.
He has observed that enterprise IT budgets are also changing. Large companies are beginning to make strategic investments in Agentic AI, including spending on models, Copilots, and agent development platforms in their IT budgets, and assessing returns based on business use cases. However, enterprises are still at an early stage of adopting Agentic AI, and widespread commercialization is still some way off.
From this perspective, a substantial portion of the approximately $107.3 billion Anthropic raised in total will need to cover computing costs over the next few years, while the company continues to expand its enterprise customer base and use cases. Before its computing contracts come due and competitors drive model prices lower, Anthropic needs to persuade more businesses to pay for Claude.
Shareholders become rivals: price wars on two fronts
Beyond fundraising, Anthropic faces a practical challenge: the companies providing it with capital and infrastructure are also its most direct competitors.
On November 9, Anthropic released Claude Fable 5.2, reclaiming the top spot at a price of $8 per million input tokens. The next day, Google made Gemini 4 Carbon widely available, with an introductory price of $2 per million input tokens, set to rise to $4 in December. Google says its costs are about one-quarter those of Fable.
In the MKA evaluation, Gemini 4 Carbon scored 71.89%, slightly below Fable’s 73.26%, but the cost per task was only $5.63, compared with $35.28 for Fable.
Notably, Google itself is an Anthropic shareholder. According to disclosures on holdings following the IPO and concurrent private placement, Google owns about 13.7% of Anthropic, primarily in non-voting Class N shares. It can benefit from Anthropic’s business growth while also using its own models to compete for customers.
But this may already be the norm. On November 17, Microsoft unveiled seven MAI models at its Ignite conference. Microsoft said that, in internal evaluations, MAI Code 2 achieved 96% of Claude Opus 5.5’s coding performance at just 5% of the price. According to Microsoft, that price includes a 50% promotional discount and will increase in May 2027. MAI Code 2 is available through GitHub Copilot.
The relationship between Microsoft and Anthropic is also quite complicated: Microsoft is an investor, customer, and cloud provider, as well as a competitor. Anthropic has acknowledged this multifaceted relationship in its prospectus.
Dai Kun explained that as multicloud architectures become a common choice for enterprise AI deployments, the relationship between the two sides will grow more complex. Large cloud providers supply capital, chips, and cloud services, control distribution channels to enterprise customers, and develop their own models. “In practice, they are both partners and competitors.” Anthropic’s decision to work with multiple cloud platforms is also intended to reduce its dependence on any single provider.
In the short term, Microsoft’s own capabilities mean it will not yet put pressure on Anthropic’s gross margin, the IT-industry analyst cited above told Huxiu. Model capabilities depend not only on computing power, but also on factors such as training methods, model architecture, data, and talent. “It’s not just computing power.”
An industry insider familiar with Microsoft told Huxiu that Microsoft had previously been cautious about developing its own large models because it had already invested in projects such as OpenAI, making in-house model development a lower priority. But that has now changed.
Dai Kun also cautioned that if different models’ capabilities gradually converge, cloud providers with control over enterprise channels and an infrastructure cost advantage will become more competitive. Pure-play model companies will then face challenges to their gross margins and pricing power. Only leading providers that maintain a clear technological edge will be more likely to sustain higher prices.
As for whether Anthropic’s new office suite will replace Microsoft’s office software, Dai Kun believes it is too early to read too much into it. In the short term, it “will put more competitive pressure on Microsoft Copilot than replace Microsoft 365.”
Google and Microsoft’s moves are both reminders to Anthropic that leading in model capability does not mean it can maintain high prices indefinitely. As more cloud providers and tech companies launch their own models, the cost to customers of switching between products may fall further, making price and performance more direct competitive factors.
Whether Anthropic can maintain its lead depends not only on whether its models can continue to pull ahead, but also on whether customers are willing to keep paying over the long term.
“Underpromise and overdeliver”
Completing its IPO marks the beginning of a new stage for Anthropic.
The first-day gain reflected market confidence in Anthropic, but it was also influenced by the size of the offering, the allocation structure, and the available trading float. The first-day performance alone is not enough to judge the company’s long-term value.
Going forward, the market’s focus will gradually shift to operating data. Investors will continue to compare Anthropic’s ARR with the company’s long-term targets. According to the company’s disclosed plans, its revenue target for 2028 is $200 billion, and its target operating margin at maturity is 35%. Whether it can achieve that growth over the next two years will be an important indicator of its business performance.
At the opening bell ceremony, Dario Amodei spoke about a concept he called “the race to the top.”
He said that if an AI company can demonstrate that it is possible to make the right choices on safety while still building an extraordinarily successful business, other companies may follow suit. To him, the industry’s goal should not simply be to compete for the top spot in model capabilities. “Whether Anthropic wins or loses at any given moment,” the industry should be made better as a whole.
He also said, “The hardest safety work is still ahead of us,” and that the company would take as long as necessary.
For a company that has just completed the largest IPO in history, these remarks can easily be interpreted as an expression of corporate values. But whether investment in safety ultimately becomes a competitive advantage will depend on the results. Anthropic must continue investing in safety research while proving that this does not impede product iteration, revenue growth, or commercial returns.
Looking at capital flows in the AI industry, the market’s focus may also be shifting. In recent years, large amounts of capital flowed to GPU, data center, cloud service, and chip companies. As models increasingly enter enterprise workflows, investors are starting to look for companies that can turn AI capabilities into actual revenue.
The brokerage analyst cited above called Anthropic “a hedge against AI disruption of traditional industries.” This means some investors are buying Anthropic not only because they think it could become a bigger model company, but also because AI technology could transform traditional industries.
Whether Anthropic can deliver on these expectations will depend on its operating results over the coming quarters—a pressure its founders and executives must confront head-on. The IPO proceeds are now in the bank, but spending on computing and competition among models will not stop. For investors, every earnings report from here on will be more persuasive than promises made at the opening bell.
