Anthropic’s IPO Investors Want Revenue Per Token. Here’s the Founder-Sized Version of That Question.

Potential investors evaluating Anthropic’s upcoming IPO are reportedly pressing the company for disclosures well beyond standard financial statements, specifically revenue per token and revenue per gigawatt of compute. These are not metrics companies typically publish, and what Anthropic ultimately discloses could set a real benchmark for how OpenAI and other AI companies report their numbers going forward. The specific metrics do not apply to your business. The underlying discipline behind demanding them absolutely does.

Why Investors Are Asking for This Specifically

Revenue is an easy number to report and a surprisingly poor one for understanding whether a business is actually healthy. A company can grow revenue while its underlying unit economics quietly deteriorate, more customers, more infrastructure cost per customer, thinner margins hidden inside a headline growth number that looks impressive on its own. Sophisticated investors know this, which is why they ask for revenue per unit of whatever the business’s core cost driver actually is. For an AI company, that is tokens processed and compute consumed. Revenue per token tells you whether the business is actually making money on the fundamental unit of what it sells, independent of how large the top-line number looks.

This is not a new idea in investing, it is standard practice in any capital-intensive business. Retailers get evaluated on revenue per square foot. Airlines get evaluated on revenue per available seat mile. AI companies are now getting evaluated on revenue per token, because that is the equivalent unit of cost and value in this business.

The Version of This Question Every Founder Should Be Asking

You do not need to disclose anything to public market investors, but you should be asking yourself the exact same category of question about your own business. What is the true unit of cost driving your business, and what is your revenue relative to that unit. For a service business, that might be revenue per billable hour or revenue per client served. For a product business, it might be revenue per unit produced or revenue per support ticket resolved. For a business running meaningful AI infrastructure of its own, it might genuinely be revenue per token, the same metric Anthropic’s investors are asking about, just at a much smaller scale.

Most founders track revenue and track costs, but rarely connect the two into a single ratio tied to their actual core operational unit. That connection is where the real signal lives. A business growing revenue 20 percent a year while its cost per unit of output is quietly climbing is in a very different position than a business growing revenue 20 percent a year with cost per unit holding steady or dropping, even though both look identical on a simple revenue chart.

Why This Matters More as AI Becomes Part of Your Cost Structure

If AI tools are now part of how you deliver your product or service, whether that is content generation, customer support, internal operations, or something more central to your offering, you have a genuine version of Anthropic’s question sitting inside your own business. What is your revenue relative to your AI spend, specifically, not lumped into a general overhead category but isolated as its own line. If a client or project relies heavily on AI-assisted delivery, do you know whether that specific relationship is actually profitable once AI costs are properly attributed to it, or are you assuming profitability because the top-line number for that client looks fine.

This is exactly the blind spot we have written about before with AI token costs generally, spend that feels small in the moment and invisible in aggregate reporting, until someone asks the specific unit-economics question and the answer is genuinely unclear.

A Simple Exercise Worth Running This Week

Pick your business’s actual core unit, the thing that scales when you grow and costs money when you produce it. Calculate your revenue against that unit specifically, not against your total revenue divided by headcount or some other rough proxy, the actual unit tied to your real cost driver. If AI tooling is a meaningful part of delivering that unit, isolate that cost specifically rather than burying it in general overhead.

If the resulting number looks healthy and stable, that is useful confirmation. If it looks murky, or you realize you cannot actually calculate it cleanly because your cost tracking is not granular enough, that gap itself is the finding worth acting on. Investors asking Anthropic this question are not being unusually demanding. They are asking the question that any founder, at any scale, should already be able to answer about their own business.

---

If you want to think through where AI costs specifically fit into that calculation, this connects directly: Your AI Bill Is About to Surprise You