Hook
Ramp, a US expense management platform, dropped a data point that sent ripples through the AI investment community: Anthropic is leading in US enterprise AI adoption. The claim is stark, binary, and seemingly bullish. But before you adjust your portfolio, let me tell you what the code—and the data—actually says. The code never lies, but the auditors do. And in this case, the auditor is an expense-tracking platform with its own incentives. I've spent years dissecting incentive structures in DeFi protocols, and I recognize the same pattern of selective data presentation here. The signal is not the truth; it's the noise you haven't parsed yet.
Context
Ramp is not an AI research firm. It's a corporate spend management platform that processes invoices, subscriptions, and software bills for thousands of US companies. Its recent report, aggregated by Crypto Briefing (a crypto-native outlet, not a mainstream tech journal), claims that Anthropic's Claude models have surpassed OpenAI's GPT in enterprise adoption based on actual payment data. The report hasn't been published in full, and the methodology—sample size, time range, industry distribution—remains undisclosed. What we have is a single line: "Anthropic leads US enterprise AI adoption." This is a classic low-information-density signal, but in a bear market for AI narratives, every data point becomes a catalyst. The market is desperate for direction, and Ramp provided one.
Core: The Data Audit
Let me apply the same forensic rigor I used when auditing Neo's smart contract architecture in 2017. The first question: what does "leading" mean in Ramp's context? Adoption could be measured by number of paying customers, total spend, or growth rate. Without the raw data, we're forced to infer. Ramp's strength is catching actual invoices—API credits, SaaS subscriptions, cloud marketplace purchases. This is a stronger signal than download counts or GitHub stars. But it's also a narrow window.
The sample bias problem. Ramp's typical customer is a mid-growth tech company—startups, scale-ups, and SMBs. These are the same companies that adopted Claude early because of its developer-friendly API and superior coding performance. OpenAI's enterprise footprint, by contrast, is heavily weighted toward large traditional enterprises (banking, insurance, healthcare) that often use Azure OpenAI Service, which gets billed through Microsoft's unified cloud contracts, not as a standalone line item. If Ramp's data only captures standalone AI subscriptions, it would systematically undercount OpenAI's enterprise share. This is not a minor edge case; it's a structural data gap. I've seen this exact flaw in DeFi analytics—when you measure on-chain TVL without accounting for off-chain custody, you get a distorted picture. The same principle applies here.
The time window problem. Quarterly data vs. monthly data vs. cumulative data produce different leaders. If Anthropic's growth spiked in Q2 2025 due to the Claude 4 launch, but OpenAI's enterprise base is larger in absolute terms over a 12-month window, the "leading" claim is ephemeral. Without a timeframe, the statement is meaningless.
The competitive baseline omission. The report does not provide OpenAI's comparable figures. Any claim of "leading" without a denominator is a sign of selective disclosure. In my 2020 Curve IRV model, I proved that without a full game matrix, any single metric invites arbitrage. Here, the arbitrage is narrative: Ramp's report becomes marketing for Anthropic (and Ramp itself, which uses AI), while the real adoption landscape remains opaque.
The hidden incentive. Ramp is a commercial entity. Its "Ramp Intelligence" product is an AI agent for expense management. Publishing a report that favors a model provider (Anthropic) aligns with Ramp's own positioning as an AI-forward platform. This is not a conspiracy; it's basic incentive alignment. Trust is a vulnerability with a capital T, and the first question any auditor should ask is: who benefits from this narrative? The answer is both Ramp and Anthropic, possibly through shared investors or partnership deals—none of which are disclosed in the article.
Contrarian Angle: What the Bulls Got Right
To be fair, the bulls are not entirely wrong. Anthropic's Claude 3.5 Sonnet and Claude 4 have indeed earned a reputation for reliability, long-context handling, and safety features that appeal to enterprise buyers. Developer surveys on platforms like LMArena and Artificial Analysis show Claude models trading blows with GPT-4o. The shift from "OpenAI is the default" to "Anthropic is a serious contender" is real, especially among engineering teams. Ramp's data, even if skewed, captures a real trend: Anthropic is winning the battle for the mid-market tech stack. If the company can translate this into deals with large enterprises—where compliance and security matter more than developer enthusiasm—the lead could become sustainable. The contrarian angle is not to dismiss the report entirely, but to recognize that the margin of error is wide enough to invalidate a binary investment thesis. The bulls are right that the signal exists; they are wrong that it's sufficient.
Takeaway
Ramp's report is a single data point in a noisy system. It tells us that Anthropic is gaining traction in a specific segment of the US enterprise market, but it does not prove dominance. The rational response is not to revalue Anthropic at a premium, but to demand more data: sample size, time window, industry breakdown, and—most importantly—OpenAI's comparable figures. Until then, treat this as a consensus hallucination, not a floor price. Math doesn't care about your narrative. The ledger never forgets. And the exit liquidity is always someone else's balance sheet.
