Author|Huajuan Dance King

Editor|Jingyu

As the hottest field in recent years, AI has attracted countless entrepreneurs aiming to achieve the "AGI" dream. However, in this crowded arena, the concentration of investment and revenue is even more pronounced than during the internet boom.

According to the latest analysis by The Information, the combined annualized revenue of 34 leading AI startups has reached approximately $80 billion, representing a 112% increase compared to six months ago.

This figure sounds vibrant, suggesting the entire sector is racing forward. But a closer look reveals a chilling statistic:

OpenAI and Anthropic alone have captured 89% of that $80 billion.

The remaining 32 companies are left to share the remaining 11%.

Let's first examine the real scale behind these two numbers.

Anthropic's annualized revenue now exceeds $30 billion. OpenAI's own disclosed figures are between $24 and $25 billion. Together, they represent an annualized scale of roughly $55 billion.

These are two "startups" founded less than a decade ago, and this is "annualized revenue," not valuation bubbles, but the actual rate of real money flowing into their accounts.

More noteworthy are the respective growth logics of these two companies.

OpenAI's revenue engine primarily consists of ChatGPT's C-end subscription users. It progresses step by step from free to Plus, Team, and Enterprise. This path yields fast growth but also encounters a ceiling—consumer willingness to subscribe and their payment capacity have limits. Moreover, this market heavily relies on product-level user experience; if competitors launch a more user-friendly product, the switching cost for users is almost zero.

Anthropic has taken a different path. From day one, Dario Amodei defined enterprise clients and API integration as the core battleground. Claude is not meant to be a chatbot users merely like, but to become an infrastructure component within enterprise software stacks. This strategy offers much stronger stickiness—once a company deeply integrates Claude's API into its own products and workflows, the switching cost becomes prohibitively high.

In April of this year, a figure confirmed the effectiveness of this strategy: Anthropic's market share in the US enterprise market surpassed OpenAI's for the first time, reaching 34.4%. In mid-2023, this figure was less than 1%.

From less than 1% to 34%, Anthropic achieved it in less than two years.

01 Other AI Companies Survive in the Cracks

Of course, the AI startup market isn't limited to OpenAI and Anthropic. Mistral, Cohere, AI21 Labs, Perplexity, Character.AI... there is a large group of well-funded companies that have recruited top talent, each with its own story and approach.

But an 11% market share divided among 32 companies means each holds, on average, only about 0.34% of the total pie.

This isn't to say these companies lack value. Perplexity has built a genuine user base in the niche of AI search; Mistral has established a unique moat in the European market through its open-source strategy; Cohere focuses on enterprise-level private deployments, serving financial and medical institutions with extremely high data security requirements. These are all real businesses generating real revenue.

However, a harsh reality is emerging: As resources, talent, and bargaining power for compute procurement increasingly concentrate towards the top, the survival space for mid-tier companies is being systematically compressed.

Top engineers will prioritize joining OpenAI or Anthropic; cloud computing giants will offer more favorable compute agreements to leading companies; corporate procurement departments are making decisions where "using ChatGPT" or "using Claude" has become the default option, and other choices require more time to explain and justify.

This is a self-reinforcing flywheel: higher revenue → larger compute investments → stronger models → higher revenue.

An AI entrepreneur in Silicon Valley once said something to the effect that "building foundational large models is essentially a war of capital attrition. You need enough money to survive until the next round of funding, and then survive until the one after that, until the market landscape stabilizes." Looking at today's data, this war of attrition is nearing its end.

02 "Oligopolies" Aren't Comfortable Either

Of course, an 89% share of ARR doesn't mean OpenAI and Anthropic can rest easy.

Just in the past two weeks, OpenAI has simultaneously found itself entangled in several dizzying situations.

Sam Altman testified in court, stating under oath that Elon Musk once demanded 90% ownership of OpenAI. The outcome of this lawsuit will directly impact OpenAI's corporate governance structure and its transition from a non-profit to a for-profit entity.

At the same time, negotiations between OpenAI and Apple regarding a partnership for Siri have reportedly reached a serious impasse, with OpenAI reportedly preparing legal action. This is a delicate signal—the partnership with Apple was once a crucial channel for OpenAI to reach hundreds of millions of iPhone users; if it breaks down, the impact cannot be underestimated.

On the product front, OpenAI maintains a rapid pace. On May 11th, it launched OpenAI Deployment Company to help enterprises build around AI; on May 15th, it released a limited preview of GPT-5.5-Cyber for cybersecurity professionals; free users can now also see inline images within conversations.

The density of product releases and commercial disputes is soaring almost in sync.

This is a typical characteristic of a company entering a phase of "ruler's anxiety." When you are already the market leader, you must simultaneously handle technological pressure from competitors, commercial friction from partners, commercialization expectations from investors, and scrutiny from regulators and the judiciary. Each direction consumes attention.

In contrast, Anthropic currently presents a much "quieter" external image. No dramatic lawsuits, no CEO courtroom testimonies. The team led by Dario Amodei and Daniela Amodei focuses on expanding enterprise clients and iterating model capabilities, steadily chipping away at OpenAI's enterprise market share.

Of course, "quiet" does not equate to a lack of pressure. Behind Anthropic is an investment bet from Amazon reaching tens of billions of dollars. Support on this scale comes with expectations of commercial returns of equal magnitude.

03 Where Does the Industry Go After 89%?

An 89% concentration level is not historically unprecedented.

Smartphone operating systems, Android plus iOS, consistently exceed 99%.

Search engines, Google alone captures over 90%.

Cloud computing, AWS, Azure, and GCP combined account for over 65%.

These precedents show that technology infrastructure industries naturally tend towards oligopoly structures. The reasons are simple: economies of scale, network effects, and switching costs—when these three forces combine, they create an almost insurmountable moat.

AI large models, especially general-purpose large models, also possess these three characteristics. Therefore, today's 89% concentration may not be the endpoint, but an intermediate state—the final landscape could be even more concentrated than today.

But there is one variable not present in historical precedents—the speed of advancement in AI capabilities is much faster than the technological iteration of operating systems, search engines, or cloud computing.

Anthropic's growth from 1% in 2023 to 34% today is essentially due to a qualitative leap in the capabilities of the Claude model series. If a team that is obscure today trains a model tomorrow that significantly surpasses GPT-5 and Claude in a key dimension, the balance of market share could tilt again at any moment.

For those 32 companies surviving within the 11%, perhaps the clearest-eyed strategy is not to confront the giants head-on but to find those vertical scenarios where "general-purpose large models are insufficient, and specialized models are necessary," and dig deep. Legal documents, medical imaging, code security audits, industrial quality inspection—these fields have strong professional barriers that cannot be solved by simply fine-tuning GPT-5.

Industry concentration does not equate to the disappearance of opportunity. It simply means that the form of opportunity has shifted from "building a better general AI" to "building a specialized AI that is irreplaceable in a specific domain."

Two mountains are already standing there. The smart ones won't think about how to move them but will focus on finding the fertile land at their feet that others haven't yet discovered.