A $5bn AI valuation should probably make me think about innovation, disruption and the future of technology.
Instead, years of working on corporate finance transactions have trained me to think: lovely. Now show me the data room.
That is not because I assume the valuation is wrong. Some of the growth stories we are seeing are genuinely extraordinary. Instead, it’s because in transactions, the headline number is usually only the beginning. – The more interesting question is what has to sit underneath it for that number to be defensible.
AI is simply the most striking current example.
First, define the number
Every cycle develops a shorthand for scale. The dot-com era had eyeballs, then users, then downloads. Now it is ARR, and it is a genuinely useful metric: investors need a way of describing the scale, momentum and recurring nature of a business.
Whenever I see a significant ARR figure, my instinctive reaction is to ask
ARR of what, exactly?
It may be genuinely contracted recurring revenue, committed revenue that has not yet gone live, usage or consumption revenue sitting alongside subscriptions, or current revenue annualised from a particularly good month. And two businesses reporting the same figure may have entirely different levels of customer concentration beneath it.
These can all be useful metrics butthey are not necessarily economically equivalent. Before putting a multiple against a number, I want to know what the number actually represents — and whether the numbers being compared are genuinely measuring the same thing.
A current legal AI example makes the point. Harvey has recently been discussed by reference to annualised revenue, Legora by reference to ARR. Both may be legitimate measures, but they are not automatically interchangeable, and thus not easily comparable. So the first question is not which business looks cheaper. It is what does this number mean?
Show me the bridge
The bridge is the route from what the business is producing today to the scale, margins, market position and defensibility that today’s valuation assumes tomorrow.
On a transaction, I would test that bridge in four places: revenue, the contracts supporting it, the economics underneath it, and what the business owns, controls and depends upon.
- Revenue
What is signed? What is live? What is still being implemented? What is usage-based, and what is genuinely committed? What do retention and expansion look like, and how concentrated is the customer base?
Rather than ask how much revenue exists we need to know how of it is recurring, contracted, live and likely to remain. Two businesses can report the same headline revenue and have very different revenue quality. And in my experience the difference shows up in the contracts, the customer concentration and the economics.
- The contracts supporting it
Despite being a lawyer, I do not need every clause. I want the provisions that tell me how durable the revenue really is: termination and renewal, minimum commitments, pricing and price adjustment, exclusivity, change of control, customer concentration, and whether any pilots or implementation arrangements are being treated as recurring revenue.
A surprising amount of value sits in provisions that never appear in the fundraising deck. On a transaction, a termination right, a pricing mechanism or a contractual dependency can tell you far more about the durability of revenue than the headline ARR number.
- The economics underneath the growth
What does another £1 of revenue actually cost to generate? It doesn’t come out of nowhere. We need to ask what the model, inference and infrastructure costs are and what happens to them as usage increases? How much sits with third parties, and does the business produce genuine operating leverage as it scales?
Growth creates value when the economics improve with it.
- What does the business actually own and control?
I always have many questions here too.. Who owns the intellectual property? What rights exist over the data, and what customer consents are required? Which technology licences, suppliers, integrations or partners are critical? What regulatory permissions are necessary, and which parts of the operating model depend on third parties?
This matters particularly in regulated businesses, where revenue may depend not only on the product but on licences, regulatory permissions, banking or payment partners, critical suppliers and outsourced infrastructure. Those are not peripheral compliance details. They may be part of the commercial architecture that allows the business to generate revenue at all. Their durability, transferability and vulnerability to regulatory change can bear directly on value.
A dependency can be doing a great deal of the heavy lifting without appearing anywhere in the headline metric.
This is also where the moat question becomes interesting.
By moat, I mean the part of the business that remains difficult for competitors to replicate even as the underlying technology becomes cheaper and more widely available.
If access to the same or better models became more widely available tomorrow, what would the business still own, control or do better than its competitors?
In many AI businesses, I suspect the durable advantage will not be in the model itself. It may lie in how deeply the product is embedded in customer workflows, distribution, trusted brands and customer relationships, proprietary data and rights, regulatory positioning and switching costs created over time.
In other words, what is genuinely difficult to copy?
Much of that would have been recognisable to a corporate lawyer twenty years ago:
Contracts. Rights. Relationships. Control.
What has to become true?
At valuations of this order, investors are not simply pricing the business as it exists today. They are underwriting assumptions about future revenue, margins, market position and continuing defensibility.
The question is not simply whether the valuation is right:
What has to become true over the next three to five years for today’s valuation to work?
The revenue must prove durable. The contracts must support it. The economics must improve with scale. Critical dependencies must remain available. And the business must retain something valuable as the underlying technology becomes better, cheaper and more widely accessible.
The valuation gets your attention. The data room tells you what you’re actually buying.
