Anthropic Debuts on Nasdaq as Capital Markets Question Commercialization and Returns on Computing Investment
Anthropic officially debuted on Nasdaq with its IPO priced at $83 per share. Its shares closed up 17.07% on the first day of trading, giving it a fully diluted valuation of approximately $1.5 trillion. The company disclosed around $900.9 billion in purchase and lease commitments and is pursuing a diversified computing strategy involving Trainium, TPUs, and GPUs from Nvidia and AMD. Capital markets are shifting their focus from model performance toward commercialization capabilities, profitability prospects, cash flow, and returns on computing investment.
Why read this
SelectedAI score 90
Anthropic's post-listing share price volatility and prospectus data show that capital markets are shifting their assessment from model capabilities toward commercialization, cash flow, and returns on computing investment.
Topics
Body · AI translation
Article Body EM_StockImg_Start
EM_StockImg_End
Anthropic, the AI giant closely watched by capital markets and the technology industry, has officially debuted on Nasdaq.
At the opening of U.S. trading on the 19th, Anthropic’s IPO briefly lifted market sentiment. When Anthropic began trading later in the session, its shares climbed as much as about 39% to $115.10 before quickly retreating, ultimately closing up 17.07%. Meanwhile, all three major U.S. stock indexes ended lower. Some investors even said that the AI company, founded less than six years ago and responsible for the largest IPO on record, had “single-handedly dragged down the entire U.S. stock market.”
Anthropic reportedly priced its IPO at $83 per share and issued 980 million shares, raising approximately $81.3 billion. Together with $26 billion in private placements from the “seven giants”—Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC—the total amount raised exceeded $107 billion, breaking the record set by SpaceX’s IPO in June this year. Anthropic’s fully diluted valuation was approximately $1.5 trillion.
Anthropic’s listing marks a pivotal point in the AI industry’s transition from a technology race to comprehensive competition spanning capital, infrastructure and commercialization. Industry insiders interviewed by Cailian Press said that the investment rationale for the AI sector has undergone a phased shift. Behind this “mega-IPO,” capital markets are paying closer attention to companies’ commercialization capabilities, earnings prospects and returns on capital. Key questions now include how enormous computing-power commitments will translate into actual returns and whether leading positions in model performance can be sustained.
Li Qing, executive director at Frost & Sullivan China, said Anthropic’s listing represents more than just a major IPO; it is also a public-market test of global AI business models and their ability to generate returns on capital. For capital markets, Anthropic’s listing provides a public-market pricing benchmark for foundation-model companies, encourages AI companies to establish more transparent commercialization and financial evaluation systems, and advances the AI industry from venture-capital-driven development toward a more mature capital-market financing system. The market will no longer focus solely on model benchmark scores.
Nvidia Is Not the “Only Answer”
Nvidia President, CEO and director Jensen Huang recently resigned from his positions, prompting extensive discussion across the industry. One industry practitioner told Cailian Press that Nvidia’s corporate plans remain relatively clear in the near term. Assuming Huang ultimately loses all control over Nvidia, the real impact is not expected to emerge until two or three product generations later.
In Anthropic’s plans, however, Nvidia is not the “only answer” for supplying computing power. Under purchase commitments and lease agreements entered into by Anthropic as of and after September 30, 2026, its total committed spending amounts to approximately $900.9 billion. Most of this relates to computing and infrastructure, but not all of it will go toward GPUs.
In its prospectus, Anthropic explicitly identified product lines from four major computing-power suppliers: Amazon Web Services’ Trainium AI accelerator chips, Google TPUs, Nvidia GPUs and AMD GPUs. The computing power supplied by Nvidia will primarily come from the Grace Blackwell and Vera Rubin series, while Anthropic will also have opportunities to work directly with Nvidia on design and engineering.
Anthropic emphasized that this approach, which does not depend on any single supplier, expands the computing resources available to the company and enables it to optimize workloads for price and performance. The resulting flexibility requires years and substantial investment to build and would be difficult to replicate at scale.
Even so, during a recent conversation with JPMorgan Chase chief Jamie Dimon, Anthropic co-founder and CEO Dario Amodei acknowledged that computing power is the most fundamental business challenge facing the company. He said one important reason the company chose to go public was to increase its cash reserves and provide a larger buffer against highly uncertain demand for computing power.
Anthropic’s prospectus shows that the AI industry’s demand for computing infrastructure has evolved from short-term procurement into the long-term strategic allocation of resources. However, Anthropic explicitly warned that some computing-power suppliers rely on third-party financing and credit support. If financing costs rise or financing becomes unavailable, signed computing-power contracts could be delayed, repriced, scaled back or even left unfulfilled.
This week also brought a sign that the AI infrastructure investment boom may be cooling: semiconductor research firm SemiAnalysis sharply reduced its forecasts for AI accelerator shipments, capital expenditures, HBM demand and electricity demand over the coming years. The revisions for 2028 through 2030 were particularly pronounced. Its forecast for global AI accelerator capital expenditures in 2028 was cut by 28%, while its forecasts for 2029 and 2030 were reduced by 41% and 40%, respectively.
In Li Qing’s view, computing-power providers must focus not only on market demand but also on project financing, asset utilization and cash flow. Competition across the industry chain will gradually shift from who has the most capacity to who achieves higher capacity utilization, lower unit computing costs and stronger returns on capital.
Can Claude Maintain Its Lead?
Just ahead of its listing, Anthropic released its next-generation flagship models, Claude Fable 5.2 and Claude Mythos 5.2, and announced expanded workplace capabilities for Claude with the launch of an AI productivity suite. Continued advances in model capabilities, particularly its commanding lead in areas such as AI coding, have given Anthropic a strong competitive starting point.
Financial data show that Anthropic generated cumulative revenue of $33.242 billion in the first three quarters of this year, far above the $2.545 billion recorded in the same period last year. Third-quarter revenue surged 1,127% year over year, while computing and infrastructure spending grew by only 382% over the same period. As of the end of September, 11,900 customers had contributed more than $100,000 in revenue over the preceding 12 months, compared with just over 1,500 at the end of last year.
However, Anthropic’s third-quarter revenue grew by about 47% quarter over quarter, versus roughly 147% in the second quarter. Revenue continued to grow rapidly, but the pace had already slowed. The ratio of revenue to computing expenditure also fell from 2.08 in the second quarter to 1.72 in the third. Meanwhile, Anthropic remained under pressure on the profit front, posting an operating loss of about $567 million and a net loss of approximately $23.3 billion in the third quarter.
(Image source: Anthropic prospectus)
At present, around 80% of Anthropic’s revenue comes from business customers, demonstrating enormous corporate demand for paid AI services. But as its revenue base expands, future growth will increasingly depend on customer expansion, greater usage frequency, product pricing and the development of new businesses. Anthropic must still further prove that its business model can deliver stable, long-term profitability.
In Li Qing’s view, individuals often use AI to improve their experience, obtain information or pursue personal interests, making their willingness to pay and frequency of use uncertain. For businesses, however, the key consideration is economics. As long as AI can generate quantifiable economic value, companies will have an incentive to keep paying for it. This also explains why Anthropic’s positioning in coding, enterprise knowledge processing and professional workflows is so important.
Zhang Liang, AI product director at Guangxi Manong Technology, also told a CLS reporter that, given the accelerating pace at which Chinese models are catching up—particularly with the emergence of the RSI model (note: recursive self-improvement, in which an AI system can modify its own improvement methods and then use the new methods to continue optimizing itself, creating a closed loop of iterative capability enhancement)—the gaps between models will gradually narrow. Anthropic’s future growth potential may therefore remain difficult to assess, and its business model may not be secure. “The foundation of Anthropic’s business model is the capability of the Claude foundation model. If another company surpasses it, customers will choose the stronger foundation model without hesitation. The current timing is relatively favorable for Anthropic’s listing because its capabilities are still ahead. But whether Anthropic’s model capabilities can maintain that lead remains an open question.”
Recent media reports said ByteDance founder Zhang Yiming disclosed during an internal presentation that ByteDance’s large models aim to match the leading international players. He believes ByteDance is capable of competing with OpenAI and Anthropic, but that truly catching up with the technological frontier “will not happen until 2029 or 2030 at the earliest.”
Zhang Liang believes the large-model industry will ultimately be dominated by a small number of foundation models, with open-source and closed-source models coexisting, while cost will actually be the decisive variable. “Competition in the industry will be extremely intense in the short term. Over a sufficiently long time horizon, once AGI is achieved, the technology will spread and the gaps between models will narrow until they reach parity. Competition at that point may no longer be about models, but about other dimensions. For now, everyone is still ‘climbing the mountain’ and remains halfway up.”
At the same time, judging from Anthropic’s current approach, a single company can simultaneously serve as a supplier, investor, partner and competitor within its ecosystem. Behind the accelerating commercialization of AI is a long-term contest spanning models, chips, cloud computing, data centers and even energy: cloud providers are beginning to develop their own chips and models and offer AI services directly to businesses; chip companies are building their own model and software ecosystems; and model companies are expanding upstream by constructing their own infrastructure and even developing enterprise applications and Agent platforms.
“In the future, a company’s competitive advantage within the industry chain will no longer be determined solely by its position in that chain, but by whether it can control key resources, build an ecosystem and acquire end customers,” Li Qing said.
What Does the Capital Market Care About?
An interviewee previously told a CLS reporter that 2026 was expected to become a pivotal year for the capitalization of large-model companies.
At the beginning of this year, Zhipu (02513.HK), billed as “the world’s first publicly traded large-model company,” and MiniMax (00100.HK) successively listed on the Hong Kong Stock Exchange. DeepSeek recently completed a second financing round worth hundreds of billions of yuan, with Tencent, China Mobile (600941.SH), China Telecom (601728.SH), CATL (300750.SZ), GigaDevice (603986.SH) and others participating as strategic industry shareholders. This month, OpenAI said it expected to be ready for a listing in 2027, potentially bringing another “mega-IPO” to the AI industry.
The industry is currently facing multiple challenges. Li Qing told a CLS reporter that the midterm elections were more of a catalyst for the recent correction in technology stocks; what is really affecting the market is that investors have begun to reassess the payback period for AI investments. “The AI industry is moving from expectations-driven valuation expansion into valuation restructuring driven by earnings delivery and returns on capital. This does not mean that the long-term growth rationale for the AI industry has reversed. Rather, the market is becoming more rational and imposing higher requirements on companies’ profitability, cash flow and investment returns.”
Over the previous two or three years, demand for computing power continued to grow, cloud providers increased capital expenditures, large-model capabilities achieved repeated breakthroughs, and AI was set to become the next generation of infrastructure. During this stage, the market was willing to pay a substantial premium for growth expectations and was even relatively tolerant of weak short-term profitability.
However, the market has begun asking three more practical questions:
First, how long can AI capital expenditures continue growing? Global technology giants have already invested vast sums in building data centers, purchasing GPUs and deploying AI infrastructure. Investors are beginning to focus on how much actual revenue these investments can generate and when returns on capital will cover the initial outlays.
Second, can AI commercialization genuinely support current valuations? Improvements in model capabilities do not necessarily translate into proportionate revenue growth. Companies are willing to pay for AI, but whether it can truly replace human labor, improve productivity and reduce operating costs still needs to be validated across more industries.
Third, are there financing and supply risks in the construction of AI infrastructure? Capital expenditures in the AI industry have already reached an enormous scale, requiring not only chips but also electricity, data centers, networking equipment and long-term financing. If financing costs for upstream infrastructure construction rise or project delivery falls short of expectations, the realization of profits across the industry chain will be affected.
(Article source: CLS) Embedded advertisement in the middle of the article
AI Trends