Author｜Hualin Dance King

Editor｜Jingyu

AI—the hottest sector over the past two years—has drawn countless entrepreneurs aiming to realize the “AGI” dream. Yet despite this crowded field, investment and revenue concentration is even higher than during the early days of the internet.

According to a recent analysis by The Information, the annualized revenue of 34 leading AI startups has reached approximately $80 billion—a 112% increase over just six months ago.

That figure sounds robust—suggesting the entire sector is surging forward. But look closer, and you’ll find a chilling statistic:

OpenAI and Anthropic together account for 89% of that $80 billion.

The remaining 32 companies split the other 11%.

First, let’s examine the real scale behind these numbers.

Anthropic’s annualized revenue has already surpassed $30 billion. OpenAI’s self-reported figure falls between $24 billion and $25 billion. Combined, their annualized revenue totals roughly $55 billion.

These are two “startups,” each founded less than ten years ago—and this is annualized revenue, not valuation hype. It’s hard cash flowing into their bank accounts at this pace.

Even more noteworthy is each company’s distinct growth logic.

OpenAI’s revenue engine centers on ChatGPT’s consumer-facing subscription tiers—from free access to Plus, Team, and Enterprise plans—scaling upward step-by-step. This path delivers rapid growth but faces a ceiling: consumer willingness and ability to pay are finite, and the market hinges critically on perceived product quality. If a competitor launches a superior alternative, user migration costs are virtually zero.

Anthropic pursued a different strategy. From day one, Dario Amodei positioned enterprise clients and API integration as its core battleground. Claude isn’t designed to be a likable chatbot—it’s engineered to become an infrastructure component embedded within enterprise software stacks. This approach yields far stronger stickiness: once a company deeply integrates Claude’s API into its products and workflows, switching becomes prohibitively expensive.

In April, a key metric confirmed this strategy’s success: Anthropic’s share of the U.S. enterprise market officially surpassed OpenAI’s, reaching 34.4%. Just mid-2023, that figure stood below 1%.

From 1% to 34%—Anthropic achieved this in under two years.

01 Other AI Companies Operate in the Cracks

Of course, the AI startup landscape includes far more than just OpenAI and Anthropic. Mistral, Cohere, AI21 Labs, Perplexity, Character.AI—and many others—have raised substantial funding and assembled world-class talent, each pursuing its own narrative and strategy.

But with only 11% of total market share to divide among 32 companies, the average share per firm amounts to just ~0.34%.

This doesn’t mean these companies lack value. Perplexity has built genuine user traction in the AI search niche; Mistral carved out a distinctive moat in Europe through its open-source strategy; Cohere focuses on enterprise-grade private deployments serving financial and healthcare institutions with stringent data-security requirements. These are real businesses generating real revenue.

Yet a harsh reality is emerging: as industry resources, top-tier talent, and bargaining power over compute procurement increasingly consolidate at the top, mid-tier companies face systemic pressure on their operating space.

Top engineers prefer OpenAI or Anthropic; cloud hyperscalers offer better compute terms to leaders; and enterprise procurement teams now treat “ChatGPT” or “Claude” as default choices—evaluating alternatives demands extra time and justification.

This forms a self-reinforcing flywheel: higher revenue → greater compute investment → stronger models → higher revenue.

A Silicon Valley AI founder once put it bluntly: “Building foundational large language models is, at its core, a capital burn race—you need enough money to survive until your next funding round, then the next, until the market stabilizes.” Judging from today’s data, that race is nearing its end.

02 Even the ‘Oligarchs’ Aren’t at Ease

Naturally, commanding 89% of ARR doesn’t mean OpenAI and Anthropic can rest easy.

Over the past two weeks alone, OpenAI found itself entangled in multiple dizzying developments.

Sam Altman testified in court, personally confirming Elon Musk’s prior demand for a 90% equity stake in OpenAI. The lawsuit’s outcome could directly reshape OpenAI’s governance structure—and influence its transition from nonprofit to for-profit entity.

Simultaneously, negotiations between OpenAI and Apple over their Siri partnership have hit serious friction, with reports indicating OpenAI is preparing legal action. This is a subtle yet significant signal: Apple collaboration had been OpenAI’s vital conduit to hundreds of millions of iPhone users—its collapse would carry substantial repercussions.

On the product front, OpenAI remains highly active. On May 11, it launched OpenAI Deployment Company, helping enterprises build around AI; on May 15, it rolled out GPT-5.5-Cyber—a limited preview version tailored for cybersecurity professionals; and free users now see inline images within conversations.

Product launch density and commercial dispute frequency are rising in tandem.

This is the hallmark of a “ruler’s anxiety” phase. When you’re already market leader, you must simultaneously contend with technical pressure from challengers, commercial friction from partners, monetization expectations from investors, and scrutiny from regulators and courts—each demanding attention and bandwidth.

By contrast, Anthropic projects a markedly “quieter” external profile—no headline-grabbing lawsuits, no CEO courtroom appearances. Under Dario and Daniela Amodei’s leadership, the team steadily advances enterprise client acquisition and model capability iteration—incrementally eroding OpenAI’s enterprise market share.

That said, “quiet” doesn’t mean stress-free. Anthropic sits atop Amazon’s multi-billion-dollar investment bet—a level of capital backing matched only by equally ambitious commercial return expectations.

03 Where Does the Industry Go After 89%?

An 89% concentration ratio isn’t unprecedented in history.

Smartphone operating systems: Android plus iOS consistently command over 99%.

Search engines: Google alone captures over 90%.

Cloud computing: AWS, Azure, and GCP collectively hold over 65%.

These precedents show that technology infrastructure sectors naturally gravitate toward oligopolistic structures. The reason is straightforward: economies of scale, network effects, and switching costs combine to form nearly insurmountable moats.

Large language models—especially general-purpose ones—exhibit all three traits. Thus, today’s 89% concentration may not mark the endpoint, but rather an intermediate stage—final consolidation could exceed current levels.

Yet one variable distinguishes AI from historical parallels: the pace of capability advancement far outstrips that of operating systems, search engines, or cloud infrastructure.

Anthropic’s leap from 1% to 34% since 2023 stems fundamentally from qualitative leaps in Claude-series model capabilities. Should an obscure team tomorrow train a model that decisively surpasses GPT-5 and Claude across critical dimensions, market share could tilt again overnight.

For the 32 companies operating within that 11%, the clearest strategic path may not be head-on competition—but instead targeting verticals where “general-purpose LLMs fall short, and domain-specific models excel,” then digging deep. Legal document analysis, medical imaging, code security auditing, industrial quality inspection—these domains feature formidable professional barriers; fine-tuning GPT-5 won’t cut it.

Industry consolidation doesn’t erase opportunity—it merely reshapes it. Opportunity has shifted from “building a better general AI” to “building an indispensable domain-specific AI.”

Two towering mountains already stand. The smartest players aren’t trying to move them—but instead seeking fertile ground at their base, undiscovered by others.