The annualized revenue of the top 34 AI startups has reached approximately $80 billion, a 112% year-over-year increase. OpenAI and Anthropic together account for 89% of the market share, generating about $55 billion in annualized revenue between them. Anthropic grew its share of the U.S. enterprise market from less than 1% in mid-2023 to 34% in under two years. Industry resources are systematically consolidating among leading companies, squeezing the survival space for mid- and lower-tier firms. This highly concentrated landscape reflects a common trend in technology infrastructure industries, but the pace of AI capability iteration far exceeds that of traditional tech sectors, leaving market dynamics still fluid. For smaller companies, focusing deeply on vertical and specialized applications may be a more pragmatic survival strategy.
Author and source: Hualin Wuwang
As the hottest industry over the past two years, AI has attracted countless entrepreneurs striving to realize the dream of AGI. Yet, in this overcrowded field, the concentration of investment and revenue is even greater than it was during the early days of the internet.
According to the latest analysis by The Information, the combined annualized revenue of 34 leading AI startups has reached approximately $80 billion, representing an 112% increase since six months ago.
This number sounds promising, and the entire sector is booming. But upon closer inspection, you'll find a chilling statistic:
OpenAI and Anthropic took 89% of these $80 billion.
The remaining 32 companies account for the remaining 11%.
Let’s first look at the true scale behind these two numbers.
Anthropic's annualized revenue has exceeded $30 billion. OpenAI's self-reported figures range between $24 billion and $25 billion. Together, the two companies amount to a combined annualized scale of approximately $55 billion.
These are two "startups" founded less than ten years ago, and this is "annualized revenue"—not a valuation bubble—but the real speed at which money is flowing into their accounts.
More importantly, consider the growth logic of each company.
OpenAI’s revenue engine primarily consists of end-user subscriptions to ChatGPT, gradually upselling from free to Plus, Team, and Enterprise tiers. This path has progressed quickly, but it faces limitations—consumers’ willingness and ability to subscribe have caps, and this market is heavily dependent on product-level perceived experience; if a competitor releases a more user-friendly product, the cost of user migration is nearly zero.
Anthropic has taken a different path. From day one, Dario Amodei made enterprise customers and API integration the core focus. Claude isn’t designed to be a chatbot users love—it’s meant to become an infrastructure component within enterprise software stacks. This strategy creates much stronger lock-in: once a company deeply integrates Claude’s API into its products and workflows, the cost of switching becomes extremely high.
In April this year, a figures 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%.
Anthropic reached 1% to 34% in less than two years.
01 Other AI companies live in the gaps
Of course, the AI startup market isn't just about OpenAI and Anthropic. There are also Mistral, Cohere, AI21 Labs, Perplexity, Character.AI—and many other well-funded companies that have recruited top talent, each with their own unique story and strategy.
But with an 11% market share divided among 32 companies, each averages only about 0.34% of the total market.
This is not to say these companies lack value. Perplexity has built a genuine user base in the AI search niche; Mistral has established a unique moat in the European market through its open-source strategy; Cohere focuses on enterprise-grade private deployments, serving financial and healthcare institutions with extremely high data security requirements. These are real businesses generating real revenue.
But a harsh reality is emerging: as resources, talent, and purchasing power for computing power become increasingly concentrated among top players, the survival space for mid- and lower-tier companies will be systematically compressed.
Top engineers prioritize joining OpenAI or Anthropic; cloud giants offer preferential computing agreements to leading companies; and when enterprise procurement teams make decisions, “using ChatGPT” or “using Claude” have become default options—other choices require additional time to explain and justify.
This is a self-reinforcing flywheel: higher revenue → greater computational investment → 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 consumption—you need enough funding to survive until the next round of financing, then again until the next, until the market landscape stabilizes.” Based on today’s data, this war of consumption is nearly over.
02 Even "Oligarchs" Have a Hard Time
Of course, an 89% ARR share does not mean OpenAI and Anthropic can rest easy.
Over the past two weeks, OpenAI has simultaneously found itself caught in several dizzying situations.
Sam Altman testified in court, stating directly that Musk had requested 90% ownership of OpenAI. The outcome of this lawsuit will directly impact OpenAI’s corporate governance structure and its transition from a nonprofit to a for-profit entity.
Meanwhile, negotiations between OpenAI and Apple over a potential Siri partnership have broken down, with reports indicating that OpenAI is preparing to take legal action. This is a subtle but significant signal—partnering with Apple was once a crucial channel for OpenAI to reach hundreds of millions of iPhone users, and the collapse of this collaboration could have substantial consequences.
At the product level, OpenAI continues to move quickly. On May 11, it launched OpenAI Deployment Company to help businesses build around AI; on May 15, it rolled out a limited preview of GPT-5.5-Cyber for cybersecurity professionals; and free users can now see inline images within conversations.
The density of product releases and the density of commercial disputes are rising almost in tandem.
This is a classic sign of a company entering the "ruler's anxiety" phase. When you’re already the market leader, you must simultaneously manage technological pressure from competitors, business friction with partners, commercial expectations from investors, and scrutiny from regulators and the judiciary. Each direction consumes your attention.
In contrast, Anthropic currently has a much quieter public presence. There are no high-profile lawsuits or dramatic courtroom appearances by the CEO. Led by Dario Amodei and Daniela Amodei, the team focuses steadily on expanding enterprise clients and iteratively improving model capabilities, gradually eating into OpenAI’s enterprise market share.
Of course, "quiet" does not mean without pressure. Behind Anthropic is Amazon's investment of tens of billions of dollars, and such massive capital backing comes with equally substantial expectations for commercial returns.
Where is the industry headed after 03 89%?
An 89% concentration is not uncommon historically.
Smartphone operating systems, Android and iOS, have consistently accounted for over 99%.
Search engines: Google takes over 90%.
Cloud computing, with AWS, Azure, and GCP together accounting for over 65%.
These precedents illustrate that the technology infrastructure industry naturally tends toward an oligopoly. The reason is simple: the combined forces of economies of scale, network effects, and switching costs create a moat that is nearly insurmountable.
AI large models, especially general-purpose large models, also possess these three characteristics. Therefore, the current 89% concentration may not be the end point, but rather an intermediate state—the final landscape could be even more concentrated than it is today.
But there is a variable here that has no historical precedent—the rate of advancement in AI capabilities is far faster than the technological iterations of operating systems, search engines, or cloud computing.
Anthropic has grown from 1% in 2023 to 34% today primarily because the Claude series models achieved a qualitative leap in capability. If a currently obscure team were to train a model tomorrow that significantly outperforms both GPT-5 and Claude on a critical dimension, the market share balance could shift again at any moment.
For the 32 companies operating within the 11%, the most prudent strategy may not be direct confrontation, but rather identifying and deeply exploring vertical use cases where general-purpose models fall short and specialized models excel—such as legal documents, medical imaging, code security audits, and industrial quality inspection. These fields have strong professional barriers that cannot be overcome simply by fine-tuning GPT-5.
Industry consolidation does not mean the disappearance of opportunities; it simply means that opportunities have shifted from "building a better general AI" to "building a specialized AI that is indispensable in a specific domain."
Two great mountains are already there. The wise don’t think about how to move them, but rather look for fertile land at their base that others haven’t yet discovered.