Author｜Hualin Dancing King

Editor｜Jingyu

AI, as the hottest arena in the past two years, has attracted countless entrepreneurs to realize the dream of "AGI." However, in such a crowded arena, the concentration of investment and income is even greater than that of the internet back in its heyday.

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

This figure sounds prosperous, the entire sector is booming. But upon closer examination, you will find a chilling statistic:

OpenAI and Anthropic together took 89% of this $80 billion.

The remaining 32 companies share the remaining 11%.

Let's first look at the actual scale behind these two figures.

Anthropic's annualized revenue has exceeded $30 billion. OpenAI's own disclosed figure is between $24 billion and $25 billion. Combined, the two amount to nearly $55 billion in annualized scale.

These are two "startups" founded less than ten years ago, and this is "annualized revenue," not a valuation bubble; it is the speed at which real money flows into accounts.

What's more worth noting is each company's growth logic.

OpenAI's revenue engine mainly comes from C-end subscription users of ChatGPT. Starting from free, then Plus, Team, and Enterprise, each tier upward. This path is fast, but there are also ceilings—consumer subscription willingness and payment ability have limits, and this market is extremely dependent on product-level perceived experience. Once competitors launch more user-friendly products, the cost of user migration is almost zero.

Anthropic, however, has taken another path. From day one, Dario Amodei set enterprise customers and API integration as the core battlefield. Claude is not meant to be a chatbot that users like, but rather to serve as an infrastructure component in enterprise software stacks. This strategy is much stickier—once a company deeply integrates Claude's API into its own products and workflows, the migration cost becomes exceedingly high.

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

From 1% to 34%, Anthropic took less than two years.

01 Other AI Companies Live in the Gaps

Of course, the AI startup market is not just limited to OpenAI and Anthropic. Mistral, Cohere, AI21 Labs, Perplexity, Character.AI... there are many other companies that have received substantial funding and recruited top talent, each with its own story and strategy.

However, the 11% market share has to be divided among 32 companies, averaging only about 0.34% of the total pie per company.

This isn’t to say these companies lack value. Perplexity has built a real user base in the AI search niche; Mistral has created a unique moat in the European market through its open-source strategy; Cohere focuses on enterprise-level private deployment, serving financial and healthcare organizations with extremely high data security requirements. These are all real businesses, generating real income.

But a cruel reality is emerging: as resources, talent, and purchasing power for computing capabilities in the industry increasingly concentrate at the top, the survival space for mid-tier companies will be systematically compressed.

Top engineers will prioritize going to OpenAI or Anthropic; cloud computing giants will offer better computing agreements to leading companies; purchasing departments in enterprises will default to "using ChatGPT" or "using Claude," requiring more time to explain and persuade for alternatives.

This creates a self-reinforcing flywheel: the higher the revenue → the greater the investment in computing power → the stronger the model → the higher the revenue.

An AI entrepreneur in Silicon Valley once said that "working on foundational large models is essentially a war of capital consumption; you need enough money to survive to the next round of financing, and then to the round after that, until the market structure stabilizes." According to today’s data, this war of attrition is nearing its end.

02 The "Oligopolists" Are Not Relaxed Either

Of course, an 89% share of ARR does not mean OpenAI and Anthropic are resting easy.

In the past two weeks, OpenAI has found itself in several dizzying situations.

Sam Altman testified in court and stated that Musk had previously demanded 90% equity in OpenAI. The outcome of this lawsuit will directly impact OpenAI's governance structure and its shift from a nonprofit to a for-profit entity.

At the same time, serious disagreements arose in negotiations between OpenAI and Apple regarding their Siri partnership, with reports indicating that OpenAI is preparing for legal action. This is a subtle signal—collaboration with Apple has been a crucial channel for OpenAI to reach hundreds of millions of iPhone users, and a rupture in this partnership could have significant repercussions.

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

The density of product releases and the density of commercial disputes are rising almost in sync.

This is actually a typical feature of a company entering the "ruler anxiety" stage. When you are already the market leader, you must simultaneously contend with technological pressures from competitors, business friction from partners, commercialization expectations from investors, and scrutiny from regulators and the judiciary. Each direction consumes attention.

In contrast, Anthropic's current external image is much "quieter." There are no loud lawsuits or dramatic court appearances by the CEO. The team led by Dario Amodei and Daniela Amodei focuses on expanding enterprise customers and iterating model capabilities, gradually eating into OpenAI's enterprise market share.

Of course, "quiet" does not mean without pressure. Behind Anthropic is Amazon's multi-billion-dollar investment bet, and such a scale of capital support brings an equal level of expectation for commercialization returns.

03 Where Will the Industry Go After 89%?

A concentration of 89% is historically not uncommon.

In smartphone operating systems, Android and iOS combined have consistently exceeded 99%.

In search engines, Google alone holds over 90%.

In cloud computing, AWS, Azure, and GCP together account for over 65%.

These precedents illustrate that the technological infrastructure industry naturally tends towards an oligopoly. The reasons are simple: scale effects, network effects, and switching costs; these three forces combine to create virtually insurmountable moats.

Large AI models, especially general large models, similarly possess these three characteristics. Therefore, today’s 89% concentration may not be the end but a midpoint—the final structure may be even more concentrated than today.

However, there is a variable here that has not been seen in historical precedents—the speed of AI capability advancement is much faster than the technological iterations of operating systems, search engines, or cloud computing.

Anthropic's growth from 1% in 2023 to 34% today is fundamentally due to qualitative leaps in the capabilities of the Claude series models. If a team that is still relatively unknown today were to train a model that significantly surpasses GPT-5 and Claude in a key dimension tomorrow, the balance of market share could tilt again at any moment.

For those 32 companies living within the 11%, the most lucid strategy may not be to confront directly but to find those vertical scenarios where "general large models are inadequate, and specialized models perform better," and drill down into those. Legal documents, medical imaging, code security audits, industrial quality inspections—these fields have strong professional barriers that cannot simply be solved by fine-tuning GPT-5.

Concentration in the industry does not mean the disappearance of opportunity. It simply means that the form of opportunity has shifted from "creating a better general AI" to "creating an irreplaceable specialized AI in a certain field."

Two mountains have already established themselves. The clever ones do not think about how to move them but instead seek that piece of fertile land at the foot of the mountain that others have yet to discover.