Key Takeaways
- The top 1 % of customers generate roughly 80 % of revenue for both OpenAI and Anthropic, a concentration level unmatched in other software categories tracked by Ramp.
- At Anthropic, two AI‑coding tools—Cursor and GitHub Copilot—accounted for about $1.2 billion of the company’s $5 billion revenue in the last fiscal year, close to a quarter of total sales from just two customers.
- OpenAI reports a broad user base (over 9,000 organizations processing >10 billion tokens and nearly 200 exceeding 1 trillion tokens each) but does not disclose the revenue share of any single client.
- Anthropic’s enterprise footprint is expanding rapidly: roughly 6,000 customers now spend ≥ $100,000 annually (up seven‑fold year‑over‑year) and more than 1,000 exceed $1 million annually, with API‑driven usage constituting about 80 % of its revenue.
- Unlike traditional seat‑based software, AI costs are usage‑driven; spending scales with how much the model actually does, not merely with headcount.
- Market‑share data from Menlo Ventures shows Anthropic now holds ~40 % of enterprise LLM API spend, OpenAI ~27 %, and Google ~21 %, together accounting for roughly 88 % of the market.
- Financial services exhibit deep AI adoption (95 % report broad or embedded use) and anticipate rising spend, suggesting the next wave of heavy users may emerge outside pure‑play tech firms.
- The persistence of this top‑heavy revenue pattern remains an open question; while non‑tech sectors could diffuse spending, reliance on a few high‑impact applications (e.g., coding tools) continues to concentrate risk.
Concentration of Revenue Among Top Customers
Recent data from Ramp highlight a striking imbalance in the enterprise AI market: the top 1 % of customers account for about 80 % of revenue at both OpenAI and Anthropic. As Ramp lead economist Ara Kharazian observed on X, “This is a level of concentration risk unseen in any other software category we track.” He reiterated the point on LinkedIn, emphasizing that such dominance is unusual compared with traditional SaaS models where revenue is more evenly distributed across the client base. The figure has remained stable even as a growing number of businesses begin paying for generative AI services, indicating that the skew is not a transient artifact of early adoption but a structural feature of the current market dynamics.
Anthropic’s Revenue Drivers: Cursor and GitHub Copilot
Within Anthropic’s revenue stream, two specific products have emerged as outsized contributors. VentureBeat reported that Cursor and GitHub Copilot together drove roughly $1.2 billion of the company’s $5 billion revenue milestone last year—close to a quarter of total sales derived from just two customers. One of those customers, GitHub, is owned by Microsoft, which has also poured $13 billion into Anthropic’s chief rival, OpenAI. This tight coupling between a major cloud provider and a leading AI startup underscores how strategic partnerships can quickly translate into disproportionate revenue concentrations, especially when the underlying tools address high‑value developer workflows.
OpenAI’s Broader Usage Patterns
OpenAI paints a picture of a more dispersed user base. In its State of Enterprise AI report, the company noted that more than 9,000 organizations have processed over 10 billion tokens through its API, and nearly 200 have exceeded 1 trillion tokens each. While these numbers illustrate widespread adoption, OpenAI has not disclosed what share of total revenue any single customer represents, leaving analysts to infer the extent of concentration from indirect signals. The disparity between the sheer volume of users and the revenue concentration reported by Ramp suggests that a relatively small subset of heavy‑usage accounts may be driving the bulk of OpenAI’s income, even if the firm chooses not to publish those details.
Anthropic’s Growing Enterprise Base
Anthropic’s own disclosures reinforce the top‑heavy narrative while showing rapid expansion of its enterprise footprint. The company told investors it now serves roughly 6,000 customers spending at least $100,000 annually—a seven‑fold increase over the past year. Moreover, the number of clients allocating ≥ $1 million per year more than doubled in just two months to surpass 1,000, according to both 24/7 Wall St. and PYMNTS. Research firm Sacra estimates that enterprise and startup API calls, priced by usage rather than a flat seat fee, drive about 80 % of Anthropic’s total revenue, a figure that aligns closely with the concentration measured by Ramp’s external data. This dual picture—broad customer growth paired with outsized spend from a minority—highlights the scalability of usage‑based pricing models in AI.
Why AI Spending Differs From Seat‑Based Pricing
The divergence between AI expenditures and conventional software licensing stems from how costs are incurred. A traditional contract usually charges per seat or per user, making expenses predictable based on headcount. By contrast, generative AI pricing is largely usage‑driven: organizations pay for the number of tokens processed, the complexity of prompts, or the volume of generated output. As one industry observer put it, “AI costs aren’t set simply by headcount.” This variable cost structure allows a single high‑impact application—such as an AI‑powered coding assistant—to consume a large share of a provider’s compute budget, translating quickly into outsized revenue streams. Consequently, a small number of intensive users can dominate financial outcomes even as the overall customer base widens.
Market Share Shifts Among LLM Providers
The competitive landscape is also evolving. Menlo Ventures’ 2025 State of Generative AI in the Enterprise report indicates that enterprise API spending has shifted toward Anthropic, which now commands roughly 40 % of the market, up from just 12 % in 2023. OpenAI’s share has fallen to about 27 % (down from 50 % in 2023), while Google has climbed to 21 %. Together, these three providers account for approximately 88 % of enterprise LLM API usage. Notably, Menlo Ventures is itself an investor in Anthropic, which may color its perspective but does not alter the underlying trend of a consolidating market where a few players capture the majority of spend.
Potential Expansion Into Non‑Tech Sectors
Beyond the traditional tech hub, other industries are beginning to embed AI at scale. According to a PYMNTS Intelligence August 2026 Enterprise AI Benchmark Report, 95 % of financial firms report broad or embedded use of newer AI in their data and technology work, and 80 % anticipate increased spending over the next 12 months, with none expecting cuts. The same firms have achieved majority adoption in 27 of 75 tracked tasks—more than the combined adoption seen in healthcare and media. These statistics suggest that sectors handling massive transaction volumes, such as banking, payments, and retail, could become significant sources of AI demand once pilot projects move into full‑scale production.
Open Question: Will Concentration Persist?
The central uncertainty is whether the current top‑heavy revenue pattern will diffuse as AI penetrates non‑technology firms or remain entrenched in a narrow set of high‑impact applications. Anthropic’s reliance on Cursor and GitHub Copilot illustrates how a single use case can generate a quarter of a provider’s revenue virtually overnight, creating concentration risk that mirrors the dynamics seen in earlier platform ecosystems. If banks, retailers, and other data‑intensive industries adopt AI broadly but with moderate per‑firm usage, the overall spend distribution could flatten. Conversely, if a few vertical‑specific AI tools achieve runaway success—much like the coding assistants have for developers—the spending curve may stay skewed, leaving providers exposed to the fortunes of a handful of power users. The coming years will reveal whether the enterprise AI market matures into a more diversified revenue base or continues to be shaped by a small elite of heavy users.
OpenAI and Anthropic Get 80% of Revenue From 1% of Customers