HomeArticle

Anthropic goes public, but its values may determine how far it can go.

竞合人工智能2026-11-19 11:28
True values are never just a single sentence written on a company's official website.

From departing OpenAI to a $1.5 trillion IPO; from an idealist focused on AI safety to a commercial giant swept up by capital, computing power and the U.S. national security system. What Anthropic truly needs to answer may not be how much more powerful its models can become, but whether it can still adhere to what it once believed in.

On November 19, 2026, the artificial intelligence industry ushered in a landmark moment.

Anthropic officially debuted on the NASDAQ under the ticker symbol ANT. Previously, the developer of the Claude large language model set its offering price at $83, exceeding the originally planned range of $70 to $78. Calculated based on the share capital after issuance, the company's valuation reached approximately $1.5 trillion, and the scale of IPO fundraising also set a historical record in the U.S. capital market.

The market's enthusiasm is quite straightforward. Earlier, institutional subscription orders once reached about six times the offering size, and some reports even indicated the figure was close to seven times.

If we only look at the figures given by the capital market, Anthropic is undoubtedly one of the most dazzling technology companies of this era. It not only represents a peak in the commercialization of generative AI, but also means that AI companies are beginning to redefine the valuation boundary of the capital market at a speed that even internet giants and traditional energy giants could hardly imagine.

But behind this grand listing ceremony, there is a question worthy of investors' careful consideration:

Can an enterprise that has written AI safety and the long-term interests of humanity into its corporate mission always adhere to its values in the face of capital, power and commercial interests?

This is not an abstract philosophical question. Anthropic's development experience over the past few years has already provided a rather complex answer to this question.

01 From Baidu to OpenAI: What Dario Amodei truly believes in is the Scaling Law

To understand Anthropic, you first need to understand its founder Dario Amodei.

Many people interpret Anthropic's story as a group of AI safety idealists leaving OpenAI to start their own new venture. But if you trace the timeline back, you will find that the company's technical genes actually come from a very pragmatic engineering logic.

In 2014, Amodei joined Baidu's U.S. research team to participate in deep learning related research at that time. In the technical environment led by Andrew Ng and other researchers, he gradually recognized a rule that later changed the entire AI industry: when data, computing power and model scale continue to expand, the capabilities of artificial intelligence systems may see leaps that far exceed traditional expectations.

This is the Scaling Law that has been widely discussed later.

Stacking data, stacking parameters, stacking computing power, model capabilities improve as the scale grows.

This logic later became an important technical foundation for the development of the GPT series of models. Amodei joined OpenAI in 2016 and became one of its core researchers, participating in the research and development of models such as GPT-2 and GPT-3, and also participated in promoting reinforcement learning technology based on human feedback.

Judging from his technical resume, he is not a person who only talks about safety and knows nothing about business and engineering. On the contrary, he has a deep understanding of the technical route of large models, and knows what kind of resource investment is needed to win this competition.

The problem is that he gradually diverged from the OpenAI management on who should master this technology, at what speed it should be commercialized, and how much decision-making power safety principles should have.

In December 2020, Dario, his sister Daniela Amodei, and five co-founders left OpenAI and later founded Anthropic.

These seven people formed Anthropic's initial founding team.

This departure certainly cannot be simply interpreted as a personal power struggle. In a later interview with Stanford Graduate School of Business, Daniela explained it as the team's desire to build a company that truly puts AI safety and responsibility at the core of the organization.

But we cannot ignore another reality: At that time, OpenAI had gradually moved towards commercialization, the strategic value of GPT technology was getting higher and higher, and the conflicts between research directions, organizational power and commercial interests were becoming more and more acute.

From the day of its birth, Anthropic has carried a special identity: it is not only an important successor of OpenAI's technical route, but also a rebellion against OpenAI's governance model.

This dual identity determines the fundamental contradictions of its subsequent development.

02 The Other Side of Idealism: Capital, Connections and Power Networks

For an AI company to carry out so-called safety research, it must first survive.

In the era of large models, surviving means huge amounts of capital, computing power resources, chip supply and cloud computing contracts.

During Anthropic's early financing process, there are several stories worthy of attention.

The first one is FTX.

In 2021, Alameda Research under Sam Bankman-Fried invested about 500 million US dollars in Anthropic, when the company's valuation was only about 40 billion US dollars. This investment later became a highly dramatic case in the history of technology investment.

With Anthropic going public, if this equity had been retained all the time, its theoretical value could have reached tens of billions of dollars. But in fact, after FTX went bankrupt, this part of the equity was sold for about 1.3 billion US dollars in 2024 to repay creditors.

There is another connection behind this investment worthy of attention: Bankman-Fried has long participated in the effective altruism movement, and Anthropic's concept of emphasizing AI safety and the long-term interests of humanity intersects with this ideological system at that time.

The second is Eric Schmidt.

The former Google CEO later became an important early investor in Anthropic. According to media reports, the connection between Schmidt and Amodei was first introduced by Cami Clark, who later became Amodei's wife.

Schmidt participated in the investment during Anthropic's early financing round in 2021. As the company's valuation climbed rapidly, the paper return on this early investment also became staggering.

These stories show that Silicon Valley's capital network is not just cold institutional investment. Technical talents, personal relationships, trust between investors, and common judgments on future technology directions may all determine the early fate of a startup.

But we must distinguish between normal relationship networks and corporate governance issues.

Dario and Daniela are siblings, and Daniela serves as President of Anthropic and concurrently Chairman of the Board, which is the public management structure. We cannot directly conclude that there is improper benefit transfer in the company just because the two are relatives. As for the online rumor that the so-called brother-in-law serves as the head of safety, there is currently no sufficiently reliable public evidence to support it, so it should not be regarded as a fact.

What is really worth discussing is how Anthropic arranges the power of capital, strategic partners and the founding team in the same governance system.

Giants such as Amazon, Google, and Microsoft are not only investors, but also have formed deep cloud computing and infrastructure cooperation relationships with Anthropic.

According to the prospectus documents, the company has signed extremely large-scale computing power and infrastructure commitments. It needs giants to provide computing power, and giants obtain an important AI ecosystem entry point through investments and long-term procurement contracts.

This is an interdependent commercial relationship.

But it also raises a question: when investors are also suppliers, cloud computing platforms and strategic partners, how much independence does Anthropic actually have?

Can a company that claims to independently judge the direction of AI development truly be commercially independent of the giants that provide it with survival resources?

03 50.1% Voting Rights: Who Supervises the Guardians of Values?

If the early financing stories are only common phenomena in the Silicon Valley startup ecosystem, then Anthropic's IPO governance structure is the part most worthy of careful study in this listing.

According to the prospectus documents submitted to the U.S. Securities and Exchange Commission, Anthropic adopts a multi-level share and governance arrangement.

The most notable of these is Class F shares.

This is a special founder share held by Founder LLC established by the founders. Dario, Daniela and five other co-founders are the initial members of this entity.

This share has no ordinary economic rights and interests, but has voting rights equivalent to 50.1% of the company's total voting rights.

More importantly, the prospectus clearly stipulates that the voting rights of Class F will remain at 50.1% of the total voting rights. Even if the company continues to issue Class A or Class N shares in the future, this part of the voting rights will not be diluted.

At the same time, Anthropic has also established the Long-Term Benefit Trust. The trust appoints special directors on the board of directors through Class T shares, and these directors will account for the majority of the board.

In other words, Anthropic is not a traditional listed company where public shareholders determine corporate control through capital proportion.

It has actually established two intertwined power systems: the founding team retains absolute voting advantages at the shareholder level through Class F, and the Long-Term Benefit Trust holds important power at the board level.

This design has its rationality.

The development of AI technology may bring extremely complex long-term risks. If the company has to report short-term profits to the capital market every quarter, the management may be forced to sacrifice long-term safety investment. Through a special governance structure, the founding team can avoid excessive short-term capital pressure from affecting the technical route.

But the problem lies precisely here.

If the founders have control that will not be diluted for a long time, who will supervise the founders?

If public shareholders cannot change the company's strategy through normal shareholder voting, then what they buy is the economic benefit of the company, or a kind of trust in the values of the founding team?

Anthropic's Long-Term Benefit Trust tries to solve this problem, but the trust itself also needs to face tests of independence, accountability mechanism and actual decision-making ability.

An idealistic governance structure does not mean that governance problems disappear automatically.

Especially when a company's valuation reaches 1.5 trillion US dollars, every major decision of the founding team may affect trillions of dollars in capital interests.

04 The Pentagon Incident: Stick to Principles, or Lose the Market?

If corporate governance is Anthropic's internal contradiction, then its conflict with the U.S. defense department has pushed this contradiction into the real world.

In 2026, Anthropic had a serious disagreement with the U.S. defense department over the military use of Claude.

The core of the controversy is not whether the company is willing to provide AI services for the U.S. government, but whether it is willing to lift two clear restrictions: fully autonomous lethal weapon systems, and large-scale surveillance of the American public.

The Pentagon hopes to obtain broader authorization for legal use. Anthropic insists on retaining these two red lines.

From the company's point of view, this is not difficult to understand.

Anthropic has long emphasized that advanced AI cannot be allowed to be used in all scenarios just because it is technically feasible. Especially for autonomous lethal weapons, once the system makes life-and-death decisions without effective human control, the attribution of responsibility and risk control will become major issues.

But the logic of the U.S. government is equally realistic.

The military believes that since it is a legal military operation, the government should not let a private technology company decide what tools the country can use through product restrictions.

The two sides ultimately failed to reach an agreement.

According to the judgment announced by the U.S. Court of Appeals for the District of Columbia Circuit on September 25, 2026, the court supported the defense department's decision to exclude Anthropic from the relevant supply chain. The court held that the restrictions Anthropic placed on the use of AI may constitute a supply chain security risk, and dismissed the company's relevant challenge.

The most dramatic part of this incident is:

Anthropic originally left OpenAI precisely because it did not want commercial interests to override AI safety. Now, when it actually implements this principle in the U.S. government's military procurement contracts, it finds that it is facing another set of larger power systems.

It can formulate its own AI ethical rules, but it cannot guarantee that the government will definitely accept these rules.

It can reject certain commercial contracts, but it cannot require the national security system to operate according to the values of a private company.

This is not a simple question of who is right and who is wrong, but an increasingly important real problem in the AI era:

When artificial intelligence becomes national security infrastructure, who on earth should determine the ethical boundaries of technology companies?

05 The Real Test Is Not Valuation

Anthropic's listing means that it has grown from an AI laboratory founded by researchers to one of the most important technology and commercial organizations in the world.

Its commercial achievements are truly amazing.

Claude has built competitive advantages in programming, enterprise applications and complex knowledge work fields. The company's revenue is growing rapidly, and the market has extremely high expectations for its future commercialization capabilities.

But behind the figures from the capital market, there is another set of figures.

According to disclosures in the prospectus documents, Anthropic's revenue scale is expanding rapidly, but its commitments for computing power, cloud computing and infrastructure have also reached an astonishing scale. The updated prospectus shows that the company's future commitments related to computing power and infrastructure are close to 900 billion US dollars.

This means that it can no longer run only on idealism.

It must continue to expand its customer base, continuously acquire computing power, continuously launch more powerful models, and also prove to investors that these inputs can eventually be converted into profits.

In this process, will safety principles become the cost of commercial competition?

When competitors are willing to relax restrictions on model use, can Anthropic still adhere to its own boundaries?

When large cloud service providers request larger-scale commercial cooperation, can the company still maintain independent judgment?

When the U.S. government puts forward new requirements, is the founding team willing to bear the cost of rejection?

These are the questions that Anthropic really needs to answer.

There is another detail that cannot be ignored: around Nvidia and its founder Jensen Huang, there are some speculations in the market about Anthropic providing information to the media to influence public opinion. But so far, there is no reliable public evidence sufficient to confirm that Anthropic has systematically manipulated the media and specifically launched public opinion attacks targeting Jensen Huang. Such claims cannot be confused with facts supported by court documents and prospectuses. For a company that takes transparency and responsibility as its values, media relations itself should also be subject to public scrutiny.

Anthropic is not a perfect company, and neither is OpenAI.

The initial disagreement between the two companies once represented two different understandings of the future development direction of the AI industry. But today, they have both entered an era where huge capital, commercial competition and national strategies are intertwined.

In this sense, Anthropic's listing is not only a capital event for an AI company, but also a commercial sample of how technological ideals face real interests.

An enterprise can announce its values at the beginning of its establishment, write safety into its articles of association, design a complex trust system, and reject certain seemingly highly profitable contracts.

But real values are never just a sentence written on the company's official website.

It is something that you are still willing to pay the price to adhere to when conflicts of interest occur.

Anthropic has already proven with its technology that it has the ability to become an important player in the global AI industry.

Next, it needs to prove another thing to the market:

When it has a valuation of 1.5 trillion US dollars, huge commercial interests and increasingly important political influence, can it still maintain the self it believed in when it left OpenAI back then?

This may be the real variable that determines how far Anthropic can go.

This article is from the WeChat official account "Coopetition AI", author: Coopetition, published with authorization from 36Kr.