BTC $63,230.31 -2.79%
ETH $1,764.09 -3.03%
BNB $584.22 -1.43%
XRP $0.9807 +0.86%
SOL $73.64 -0.26%
TRX $0.3091 +0.43%
DOGE $0.0589 -0.88%
ADA $0.1523 -1.91%
BCH $206.77 -2.59%
LINK $9.05 -2.87%
HYPE $66.61 -4.22%
AAVE $109.05 -3.95%
SUI $0.7175 -2.56%
XLM $0.1399 -0.02%
ZEC $925.87 -2.82%
AAPL $354.25 +0.12%
AMZN $262.08 +1.23%
GOOGL $282.55 +0.21%
MSFT $502.50 -2.01%
META $615.22 -1.12%
NVDA $154.60 -0.82%
TSLA $382.23 -0.09%
SNDK $764.00 -5.49%
INTC $53.80 -1.28%
SPCX $142.36 +1.25%
MU $698.63 -2.01%
AMD $500.01 -0.72%
BTC $63,230.31 -2.79%
ETH $1,764.09 -3.03%
BNB $584.22 -1.43%
XRP $0.9807 +0.86%
SOL $73.64 -0.26%
TRX $0.3091 +0.43%
DOGE $0.0589 -0.88%
ADA $0.1523 -1.91%
BCH $206.77 -2.59%
LINK $9.05 -2.87%
HYPE $66.61 -4.22%
AAVE $109.05 -3.95%
SUI $0.7175 -2.56%
XLM $0.1399 -0.02%
ZEC $925.87 -2.82%
AAPL $354.25 +0.12%
AMZN $262.08 +1.23%
GOOGL $282.55 +0.21%
MSFT $502.50 -2.01%
META $615.22 -1.12%
NVDA $154.60 -0.82%
TSLA $382.23 -0.09%
SNDK $764.00 -5.49%
INTC $53.80 -1.28%
SPCX $142.36 +1.25%
MU $698.63 -2.01%
AMD $500.01 -0.72%

Should we give Anthropic 1.4 trillion dollars?

Core Viewpoint
Summary: AI has indeed struck gold. But after all the costs of chips, electricity, data centers, R&D investment, and financing are settled, how much gold can those who mine it actually keep?
Tencent Technology
2026-11-18 22:08:12
AI has indeed struck gold. But after all the costs of chips, electricity, data centers, R&D investment, and financing are settled, how much gold can those who mine it actually keep?

Author: Hu Yong, Professor at Peking University's School of Journalism and Communication

Editor: Su Yang, Tencent Technology

The artificial intelligence industry is ushering in a moment of significant importance.

In November 2026, Anthropic, which was established just over five years ago, launched its initial public offering (IPO), attempting to enter the capital market with a valuation exceeding one trillion dollars.

According to the information provided in the prospectus, the basic circulating total share capital of Anthropic after the IPO will be approximately 13.07 billion shares. Based on an issue price range of $70 to $78, this corresponds to a basic market value of approximately $916 billion to $1.02 trillion. If we include about 4.92 billion potential shares from employee options, the fully diluted valuation is approximately $1.26 trillion to $1.4 trillion.

A company whose core product is a large language model achieving such a massive market valuation in such a short time is a rare phenomenon in the history of the technology industry.

However, Anthropic's listing not only signifies the capital market's recognition of the prospects of artificial intelligence but also highlights a long-standing issue that has been obscured by technological breakthroughs and financing booms, which must finally undergo public scrutiny: Artificial intelligence has created enormous market demand, but can it generate profits that match its valuation?

We hope to explore several topics through this article:

  1. $70 billion annualized revenue, adjusted operating profit of $4.1 billion, and a net loss of $104 billion all appear in the prospectus; what is the basis for judging the business model?

  2. Each inference of the model consumes computing power, and the most active users may not necessarily contribute the most revenue.

  3. Model companies organize capital in the manner of energy, telecommunications, and infrastructure, shifting the valuation logic from "how strong the model is" to capital costs, investment return cycles, and capital return rates.

  4. Companies that master user relationships, workflows, and ecosystems may find it easier to convert intelligence into revenue than model companies.

  5. There is now consensus on pricing intelligence based on "getting the work done," but there is still no consensus on who bears the economic losses caused by task failures or errors.

  6. The value created by artificial intelligence does not necessarily equal the revenue and profits of model developers.

  7. After settling costs such as chips, electricity, R&D, and financing, how much can those "mining for gold" actually retain is what Anthropic's listing truly needs to test.

01 $70 Billion vs. $13 Billion: Revenue Miracle and Profit Doubts

In the previous gold rush of artificial intelligence, the most certain winners seemed to be those selling "shovels." NVIDIA provides GPUs, TSMC handles advanced chip manufacturing, cloud computing companies build data centers, and power and energy companies provide the foundational conditions for increasingly large computing facilities.

Regardless of which large model company ultimately prevails, these infrastructure suppliers can benefit from the expansion of the entire industry.

In contrast, companies like OpenAI and Anthropic that directly develop large models have seen astonishing user and revenue growth but face a more complex economic problem: can the costs of developing intelligence, the prices of selling intelligence, and the capital investments required to maintain technological leadership form a sustainable business cycle?

Anthropic's prospectus provides important insights into understanding this issue.

According to the updated listing documents from November 2026, the company's revenue for 2025 is approximately $4.585 billion, and revenue for the first nine months of 2026 has reached $33.242 billion, with about $17 billion in revenue achieved in the third quarter alone. This means that the revenue from just one quarter has already surpassed the total for the entire first half of the year. If we simply annualize the third quarter's revenue, Anthropic's revenue run-rate is approaching $68 billion to $70 billion.

This growth rate far exceeds the pace of development in the traditional software industry, indicating that generative artificial intelligence is no longer an experimental technology primarily relying on capital subsidies to maintain user scale, but is instead forming a commercially viable market with actual payment capabilities.

Notably, Anthropic's revenue primarily comes from enterprise clients and services billed based on intelligence usage, rather than solely relying on fixed monthly fees from individual consumers. This indicates that the commercialization focus of large models is shifting from selling access to chat tools to selling intelligent capabilities that can participate in production activities.

Large models have not only struck gold but have done so in tonnage. However, the explosive growth in revenue has not automatically resolved the profitability issue.

In the first nine months of 2026, Anthropic's spending on computing power and infrastructure reached approximately $18.8 billion, accounting for about 57% of its total revenue of $33.242 billion.

Operationally, the company still reported losses, with GAAP (Generally Accepted Accounting Principles) operating losses for the first nine months amounting to approximately $1.26 billion to $1.3 billion. The net loss for the first nine months reached as high as $104 billion, but a significant portion of this came from the fair value re-evaluation of convertible financing instruments and related liabilities, rather than cash bleeding from daily operations. Meanwhile, the company disclosed that adjusted operating profit has turned positive to approximately $4.1 billion (with an adjusted operating profit margin of about 12%), and the efficiency of revenue relative to computing costs has significantly improved.

Therefore, it is clearly overly simplistic to assert that Anthropic's business model cannot be established based solely on its massive net loss; however, it is equally lacking in sufficient basis to conclude that large models have solved the profitability problem based solely on the positive adjusted operating profit.

The key lies in whether the company can generate sufficient free cash flow (FCF) in the long term after bearing ongoing R&D investments, computing power procurement, infrastructure construction, and financing costs.

An increasing number of companies are beginning to emphasize the quantifiable productivity improvements brought about by artificial intelligence, including accelerating workflows, automating routine tasks, expanding service capabilities, and improving cost control. Clearly, there is strong demand for artificial intelligence; however, Wall Street is losing patience with the costs required to meet this demand.

The most striking example is Meta. The company's third-quarter financial report shows that although artificial intelligence drove a 26% increase in Meta's advertising revenue, the company's quarterly free cash flow turned negative, marking the first instance of net cash outflow in over a decade. Meanwhile, Meta has also raised its capital expenditure expectations to support its aggressive artificial intelligence infrastructure expansion plans.

The entire Wall Street's anxiety over the ROI (Return on Investment) of AI infrastructure investments has been amplified.

In the past two years, the market would give blind premiums whenever it heard "AI capability enhancement, increased user stickiness"; however, at the current stage, as capital expenditures approach hundreds of billions or even trillions across the industry, Wall Street has begun to calculate like actuaries: When will these computing power investments actually translate into real net increases in free cash flow?

If the investment return cycle is stretched too long, or if the revenue increments brought by AI cannot offset the hundreds of billions in annual depreciation and electricity bills, the market will impose valuation penalties.

According to market research firm Gartner's forecast, if companies continue to maintain their current spending pace, by the end of 2030, over $3.5 trillion in artificial intelligence infrastructure investments may face asset impairment risks. It is expected that AI service providers will invest nearly $1 trillion in AI infrastructure in 2026, and an additional $6 trillion from 2027 to 2030.

Gartner points out that during this period, the AI revenue brought by enterprise clients and consumers may not be sufficient to support these investments for profitable returns.

02 Diseconomies of Scale

The huge mismatch between supply and demand and finances reveals a fundamental difference between large models and the traditional software industry.

For decades, the reason Software as a Service (SaaS) has become a favored business model in the capital market is that after software development is completed, the marginal cost of providing services to more users is relatively low. As customer scale expands, companies can typically increase profit margins through economies of scale. However, large models do not entirely follow this logic.

Each inference of the model, code generation, long text analysis, and intelligent agent task execution requires actual computing resources. Especially as artificial intelligence evolves from answering questions to autonomously completing complex tasks, a single task may require repeated calls to the model, retrieving external information, executing programs, and verifying results, leading to increased costs. For companies providing fixed monthly subscription services, the most active users may not necessarily be the highest profit contributors; they may instead be the users consuming the most computing power.

This is why major model vendors have to "screen users" through context length limits, call limits, or tiered billing.

In a situation where heavy users could turn from "assets" into "burdens," assessing whether a large model company's business model is mature cannot rely solely on the number of paying users, subscription revenue growth, or user activity; it must also observe the unit economic benefits of different user groups: how much revenue each user contributes, how much inference resources they consume, how profit margins change with increased usage intensity, and whether the company can improve this relationship through more reasonable pricing and technological optimization.

Large models are not destined to fail in forming economies of scale. Model compression, inference optimization, chip upgrades, caching technologies, and task allocation among different models can continuously reduce unit computing costs. The operational efficiency improvements disclosed by Anthropic also indicate that technological advancements can change cost structures.

However, artificial intelligence faces a unique demand expansion mechanism: as intelligent services become cheaper and more powerful, users may not only complete existing tasks at lower costs but may also begin to demand machines to undertake more work that was previously impossible.

The Jevons Paradox in economics provides an enlightening reference here: improving resource utilization efficiency does not necessarily reduce the total resource consumption; rather, it may stimulate demand growth. For large model companies, a decrease in unit inference costs does not necessarily mean a decrease in overall computing power expenditures. Cheaper intelligence may lead to broader applications and potentially larger computing demands.

The disclosures in Anthropic's prospectus regarding future computing power procurement commitments highlight this contradiction. The long-term computing power and infrastructure arrangements that the company has signed and disclosed total nearly $900 billion, with a significant portion expected to occur before 2030.

These figures do not mean that Anthropic has paid the corresponding amounts, nor can they be simply viewed as current debt, but they reflect an important change: cutting-edge artificial intelligence companies are transitioning from traditional light asset software companies to heavy asset technology enterprises that heavily rely on chip manufacturing, grid capacity, data centers, and long-term capital financing. Although large model companies are still betting on algorithm innovation and research talent, they are increasingly constrained by the supply of physical infrastructure.

If a company must continuously invest hundreds of billions to maintain technological leadership, then even if its revenue growth is astonishing, it cannot guarantee that its capital return rates are sufficient to support its current valuation.

Artificial intelligence is presenting a unique industrial form: it creates revenue at the speed of the software industry but may require capital organization in the manner of energy, telecommunications, and industrial infrastructure industries. Investors used to focus solely on which company has the most advanced models, the most users, and the fastest revenue growth; however, in the future, capital costs, computing power procurement prices, infrastructure utilization rates, and investment recovery cycles may equally determine a company's long-term competitiveness.

03 Possible Paths for Model Commercialization

Compared to Anthropic, OpenAI has a broader commercial layout, attempting to become a general intelligent assistant for individual users while also aiming to be a foundational platform for enterprises and developers to use artificial intelligence. However, OpenAI also faces similar contradictions.

Although OpenAI's annual recurring revenue (ARR) and quarterly revenue have reached the tens of billions of dollars, multiple public financial disclosures and industry forecasts indicate that due to the exorbitant training costs of next-generation cutting-edge models, massive inference expenses, and hundreds of billions in computing power leasing agreements, its cumulative operating losses in recent years and the expected cash flow gap in the coming years still amount to hundreds of billions or even over a trillion dollars.

Therefore, attracting users is not the key; the biggest challenge lies in how to convert widespread usage demands into revenue sufficient to cover ongoing capital investments.

Meta, on the other hand, showcases another path.

Unlike OpenAI and Anthropic, Meta does not primarily rely on selling large models for revenue but instead integrates artificial intelligence into its existing advertising, recommendation, and social platform systems. The enhancement of model capabilities can improve content distribution, ad matching, and marketing automation efficiency, thereby directly boosting the profitability of its mature business.

For Meta, artificial intelligence is primarily a technology to improve the efficiency of existing business systems, rather than necessarily a new product that requires separate charges. This model has clear advantages: it can monetize by leveraging existing customer relationships, user scale, and advertising transaction systems without having to rely entirely on consumers to establish new payment habits for intelligent services.

However, Meta also faces the issue of continuously expanding data center investments, with massive capital expenditures putting pressure on free cash flow. This indicates that having a mature business model does not eliminate the capital-intensive nature of artificial intelligence; it merely provides a more stable revenue foundation to bear these costs.

The case of Meta also reveals a deeper contradiction in the commercialization of large models: the economic value created by models may not necessarily be directly obtained by the model providers.

As artificial intelligence becomes increasingly embedded in various products and services, companies that control user relationships, application scenarios, and commercial channels may find it easier to realize value than those that solely provide model capabilities. This also raises the key question of whether foundational models will gradually commoditize, which is crucial for understanding the profit prospects of large model companies.

If the models developed by different companies gradually converge in performance, reliability, and user experience, customers will increasingly focus on price and service costs. Even if model providers continue to achieve technological advancements, they may struggle to maintain high profits in the long term due to competition.

The diffusion of technological capabilities benefits consumers but does not necessarily favor the original technology developers. Historically, many important technologies ultimately became widely used infrastructures, but the proliferation of technology itself does not guarantee that its inventors will reap the maximum economic benefits.

However, the commoditization of models does not mean that all artificial intelligence services will be commoditized.

What companies truly purchase is often not a model that can answer questions, but an intelligent system capable of understanding internal business, accessing proprietary data, executing complex processes, and meeting security and compliance requirements. When artificial intelligence deeply enters the financial, research and development, legal, and customer service processes of enterprises, switching suppliers is no longer just about changing a chat interface; it may involve data migration, system integration, risk control, and organizational training.

Therefore, in the future, a more solid business moat may not lie in the model parameters themselves, but in the integration of models with enterprise workflows, as well as the developer networks, data environments, and service ecosystems formed as a result.

From this perspective, Anthropic and OpenAI may not necessarily remain in the position of foundational model providers forever. They are also developing complete services aimed at consumers and enterprises, attempting to establish differentiated advantages through product experience, workflows, and customer relationships.

Whether it is Anthropic lowering the threshold for enterprises to deploy intelligent agents through Claude Managed Agents, leveraging Claude Code to delve into software development processes, or OpenAI promoting the integration of artificial intelligence into the daily operations of enterprises, both companies are exploring the same direction: shifting from providing model capabilities and conversational tools to offering intelligent services that can be embedded in business processes and continuously execute tasks.

At the same time, the development of intelligent agents may further change the pricing logic of artificial intelligence. Traditional subscription models sell the right to use software for a certain period, while token-based billing sells the consumption of computing resources, whereas intelligent agents may actually sell the capability to complete a task.

If an AI system can independently complete legal document reviews, software testing, or corporate financial analysis, what customers truly care about may not be how many tokens it consumed, but whether the task was completed, whether the results were reliable, and how much cost was saved as a result.

Charging based on tasks, workload, or even actual outcomes may become an important component of future business models. However, this model will also bring new responsibility issues: when artificial intelligence fails to complete a task, or its errors cause economic losses, who should bear the costs and risks? The closer intelligent services are to real labor, the harder it becomes to fully apply the traditional software industry's rules to their business models.

04 Who Can Keep the Gold They Dig Up?

From a broader perspective, the artificial intelligence industry is undergoing a restructuring of its value distribution structure.

The most obvious early beneficiaries were chip, manufacturing, and infrastructure suppliers, as the expansion of the entire industry translates into demand for computing resources. However, as model capabilities diffuse, profits may gradually shift to companies that can control enterprise workflows, user entry points, and application ecosystems.

If computing power remains scarce for a long time, infrastructure suppliers will still have strong bargaining power; if model supply becomes sufficient, application enterprises may gain greater value distribution space; if a few platforms can integrate models, data, payments, and workflows, they may form a new platform economy.

Ultimately, who will obtain the largest profits depends on which link in the industrial chain can form sustained scarcity.

Historically, Microsoft's long-term competitive advantage came not only from licensing revenue of the Windows operating system but also from the developer ecosystem, software compatibility, and enterprise customer networks formed around the operating system; Google did not build its commercial empire by charging users for search but by transforming the user entry points created by search into advertising revenue.

Today, Anthropic and OpenAI also face an important choice: whether foundational models are merely technology products that require continuous massive capital investment to maintain a leading edge, or whether they can become the platform foundation that connects developers, enterprise customers, and application ecosystems.

If they can establish a stable developer network, deeply integrated enterprise workflows, and a service system that is difficult to replace on top of the models, they may gain competitive advantages and pricing power that surpass the performance of the models themselves. However, whether this platform transformation can generate profits sufficient to cover ongoing computing power investments remains to be seen in long-term operational performance.

According to recent regulatory filings submitted by Anthropic, this is the scale that the company's upcoming initial public offering (IPO) could potentially reach in the best-case scenario: a valuation of $1.4 trillion (diluted) and $90 billion in fundraising.

This ambitious IPO is essentially a referendum by investors on the artificial intelligence industry. It provides a chance for broader investors to choose whether to support the company in continuing to burn cash or to avoid this company, which is expected to become an important barometer in the artificial intelligence industry, and refuse to give a historic high valuation of $1.4 trillion.

Regardless, Anthropic's IPO filing reveals a significant issue that cannot be ignored: to maintain revenue growth, the company must bear the continuously rising computing costs. This clearly makes it more difficult to persuade investors. If a highly sought-after artificial intelligence star company continues to perform poorly after going public, can the market still bear it? This needs time to verify.

One question is easily overlooked: the economic value created by artificial intelligence does not necessarily translate into revenue and profits for model developers.

Companies use artificial intelligence to improve productivity, and the cost savings may remain in the hands of the users; consumers may obtain convenience through free or low-cost intelligent services, which may manifest as consumer surplus rather than revenue for model suppliers. Even if artificial intelligence ultimately significantly enhances the productivity of the entire society, model developers may not be able to receive corresponding economic returns based on the value they create.

There is no inherent correspondence between the social value of technology and the financial returns of developers.

AI has indeed dug up gold. But after settling all costs for chips, electricity, data centers, research and development investments, and financing, how much can those who dig for gold actually keep? This is the core question that the business model of large models has yet to answer.

Anthropic's listing is not the end of this question but rather brings it into a more rigorous phase. Previously, the market could imagine the future based on technological breakthroughs, user growth, and financing valuations; after going public, companies must continuously prove their commercial value through revenue, profits, cash flow, and capital returns.

Whether the models are smarter is certainly important, but more importantly, whether companies can produce, distribute, and sell intelligence in an economically sustainable manner.

warnning Risk warning
app_icon
ChainCatcher Building the Web3 world with innovations.