1% of Companies Account for 80% of OpenAI and Anthropic Spending, Growth Already Stalling
nashnova research
Payment platform Ramp tracked 70,000 U.S. companies and found that just 1% of firms drive roughly 80% of commercial spending on OpenAI and Anthropic — and that top-heavy revenue base is now hitting a wall as rising usage collides with falling prices.
80% of revenue from 1% of customers — how fragile is that?
Ramp data shows 1% of companies account for roughly 80% of combined OpenAI and Anthropic commercial spending — more concentrated than any other software category Ramp tracks.
These top spenders are overwhelmingly venture-backed startups, with a median AI spend of $76 per employee per month — more than ten times the $7 for private-equity-backed firms.
This means → the AI labs' revenue hinges on a thin slice of companies burning venture capital. If VC funding tightens, that revenue pipeline narrows fast.
Who is actually propping up Anthropic's revenue?
Anthropic's prospectus reveals that just two undisclosed customers contributed nearly a quarter of its 2025 revenue.
AI coding tool Cursor (parent company Anysphere) and Microsoft's GitHub Copilot together drove roughly $1.2 billion of Anthropic's approximately $5 billion annualized revenue.
YipitData shows Anthropic's top 1% of customers contributed 46% of its revenue in August — nearly double the 25% share a year earlier.
This reflects a concentration problem that is getting worse, not better — the faster revenue grows, the deeper the dependence on a handful of clients.
The same investors fund both the customer and the platform — is the money going in circles?
Anysphere (Cursor's parent) closed a $2.3 billion round last November, led by Accel and Coatue, with Thrive Capital, Andreessen Horowitz, Nvidia, and Google participating.
Three months later, Coatue and Singapore's GIC co-led Anthropic's $30 billion Series G, with Nvidia joining again.
In plain terms = Coatue's money flows into Cursor; Cursor spends it on Anthropic's Claude API; that spending becomes Anthropic revenue, which in turn supports the valuation of Coatue's Anthropic stake. The capital is looping through the same network.
Even Ramp itself sits inside this web: it closed an $1.85 billion round last week at a $60 billion valuation, led by Dragoneer, Thrive, and Founders Fund — all of which also appear on Anthropic's Series G investor list.
Usage is still climbing — so why isn't spending?
Ramp's platform recorded "two consecutive months of near-zero growth in token spending" in August and September, even as raw token volume kept rising.
Total spending in early October hit a new high but was only 1.5% above the August peak.
This means → volume gains are being entirely eaten by price declines. The labs' "trade price for volume" strategy is compressing their own revenue.
How far have prices fallen — and how much leverage do buyers have?
Anthropic's effective token price plunged from over $1.70 per million tokens to $0.83 in roughly one month — a record low since the benchmark index launched in June. OpenAI's corresponding index sits at $0.45.
Ramp chief economist Ara Kharazian attributes this to price deflation: falling frontier-model prices combined with efficiency gains across model tiers are dragging down the blended cost per million tokens.
On the contract side, buyers hold the negotiating leverage — enterprise deals are rarely signed at above-normal spend commitments, and even then, labs are willing to offer below-list-price discounts as concessions.
Has the S-curve already peaked?
Paid AI penetration across Ramp's enterprise network rose just 0.62 percentage points in September, reaching 55.9%.
Kharazian wrote: "It feels like we are at the top of the S-curve, and growth will have to come from deeper usage rather than more new customers."
In plain terms = most companies willing to pay for AI have already adopted it. The remaining growth has to come from getting existing customers to spend more — but prices keep falling. Those two lines are moving in opposite directions.
市场有风险，内容仅供研究参考，不构成投资建议。