AI行业收入高度集中：Anthropic与OpenAI占据近九成市场份额深度解析

The Information's latest update to its generative AI database reveals a staggering figure: the combined annualized revenue from selling AI application or model access rights across 34 leading AI startups, including Anthropic and OpenAI, has approached $800 billion, equating to roughly $66 billion per month.

This aggregate has surged 112% in just six months, a remarkable pace. Yet, the more telling story lies not in the total, but in its distribution. Anthropic and OpenAI now command approximately 89% of this $800 billion annualized revenue, a concentration that has intensified by 4.5 percentage points since six months prior. This leaves the remaining 32 companies competing for a mere 11% slice of the market.

The market's message is unequivocal: this is not a broad-based boom, but an accelerating race where winners take most.

01 A Duopoly Deepens: 89% Is Not the Ceiling

The Information's analysis tracks 34 of the most-watched startups in generative AI, encompassing both model developers and application builders. While this cohort's collective annualized revenue more than doubled in under half a year, the growth engine is almost entirely fueled by the two front-runners.

Latest Annualized Revenue for 34 Top AI Startups

It's critical to note that the revenue figures for Anthropic and OpenAI contain some inflation due to differing accounting practices with their commercial partners.

Anthropic shares a portion of its revenue with cloud providers like Amazon and Google, who resell its models to their own clients. Anthropic records this revenue on a gross basis, without deducting partner payouts.

OpenAI's situation differs. It must remit 20% of its revenue to early backer Microsoft until 2030, a sum that could reach $60 billion this year. However, OpenAI typically excludes sales generated through cloud partner channels in its public disclosures.

Applying a uniform accounting standard would likely narrow the apparent revenue gap between the two giants. Yet, even with adjusted metrics, Anthropic's recent velocity has sent shockwaves through the sector.

The Wall Street Journal reports Anthropic's annualized revenue is on track to hit $50 billion by the end of June. For context, this figure stood at just $1 billion in early 2026 and soared past $30 billion by April. Insiders note the company's original plan was a 10x growth target for the year, but Q1 performance shattered expectations—with annualized revenue and usage skyrocketing 80 times year-over-year.

Anthropic and OpenAI Command a Growing Share of Annualized Revenue from AI-Native Startups

During the same period, OpenAI disclosed a monthly revenue of $2 billion in late March 2026, implying an annualized run rate of approximately $24 billion. While an OpenAI spokesperson later clarified that the March figure was not intended to represent a precise annual run rate, the market perceived a subtle shift in competitive dynamics.

Industry observers summarize the landscape: the AI field is crystallizing into a two-horse race. Anthropic and OpenAI collectively capture 89% of all revenue from top AI startups, forcing the remaining 32 to battle for the residual 11%. The winner-take-most effect is in full force.

02 Four Cross the Line, But Money Flows Back Up

Beneath the shadow of the duopoly, several smaller AI startups are pushing to reach new revenue milestones.

The Information's data shows that since December 2025, three more companies have crossed the $500 million annualized sales threshold: search company Perplexity, voice AI provider ElevenLabs, and coding application Cognition. They join coding app Cursor, which had already entered this tier.

However, a critical and often overlooked issue here is double-counting.

These four $500M+ startups, along with other listed application companies, likely pay billions annually to OpenAI and Anthropic for access to the foundational models that power their own products. This substantial expenditure essentially cycles back to the two companies at the top of the pyramid.

Cursor exemplifies this vulnerability. In the quarter ending this past January, Cursor's gross margin was negative 23%, an unusual position for a startup with significant revenue. While its margins have since turned positive, this volatility underscores the commercial fragility of deep dependence on Anthropic and OpenAI's technology—a risk magnified by Anthropic's recent price increases.

03 The Squeeze on Smaller AI Startups

The rising revenue concentration points to an unsettling reality for many small AI startups: their essential model providers are increasingly becoming direct competitors.

The Information notes that nearly every application company on its list relies partially or entirely on models from Anthropic or OpenAI. Meanwhile, these very model developers continuously launch refined product versions targeting specific industries or white-collar roles, putting them in direct competition with their own downstream customers. This dynamic inevitably further constricts the growth and survival space for smaller ventures.

Investors from firms like Sequoia Capital have long held a thesis: in the current AI era, the vast majority of software value will be captured by the elite companies developing advanced AI models, not by pure-play application developers. The latest market revenue data appears to validate this view.

This dynamic also explains the fervent fundraising environment for "new AI labs" attempting to build novel models to challenge the incumbents. Their goal is to bypass dependence on the existing giants, a pursuit that requires astronomical capital.

Another market realization is gaining clarity: for revenue generation, the key metric is not user count, but token consumption. Tokens are the data units consumed when users task an AI, directly correlating to workload volume. Coding tasks are token-intensive; revenue from a single active developer can far exceed that from a multitude of casual chatbot users.

This logic underpins OpenAI's recent product roadmap reshuffle, shuttering services like its video generation tool Sora to reallocate resources toward its coding tool Codex.

04 Burning $30B Annually: Can the Mega-IPO Wave Sustain It?

While revenue figures shine, profitability tells a different story. Anthropic and OpenAI alone may burn through over $30 billion in cash annually, largely due to the immense costs of training cutting-edge AI models. Most startups on the list are not yet profitable, and even occasional profitable quarters have proven difficult to sustain.

This hasn't slowed their march toward the public markets. According to Reuters, timelines for three blockbuster IPOs are coming into focus: SpaceX is set to lead, potentially launching its roadshow as early as June, with OpenAI and Anthropic expected to follow in the second half of the year. Together, these three could absorb a massive share of investor capital and demand, potentially crowding out other high-profile companies with valuations in the hundreds of billions, like Canva and Databricks, from the near-term IPO window.

PitchBook analyst Kyle Stanford warns that IPO opportunities long awaited by many companies could be pushed into 2027. Executives at pre-IPO software companies report that investment bankers are already urging them to ensure their listing timelines do not clash with SpaceX's proceedings.

The race is far from over. Google's competitive threat looms, Anthropic faces service disruptions and user limits due to compute constraints, and OpenAI's Codex is gaining rapid adoption. Yet one trend is becoming undeniable: AI commercialization is transitioning irreversibly from a phase of broad, exploratory experimentation to a mature, highly concentrated landscape dominated by giants. For the other 32 companies on the list, their window of opportunity is visibly narrowing.