Why AI Raised More in One Year Than a Decade
Capital formation at this scale usually follows a proven business model. This time it preceded one, and much of it went to companies with no products, customers or stated plan to earn revenue.
The volume of capital committed to artificial intelligence companies over a short period has few precedents. What makes it more remarkable is that much of it went to companies without products, customers or, in some cases, a stated plan to generate revenue.
A call option on every application the model might enable
The logic rests on a specific belief: that capability research has option value. A team that can train a competitive frontier model has, in this view, a call option on every application that model might enable. Investors are willing to fund the option even without knowing what will be built.
- Talent is the scarce input, and talent can be bought with capital.
- The window before incumbents consolidate is believed to be closing.
- A single successful application can justify a portfolio’s entire cost.
- Limited partners have pressured funds to have AI exposure.
Compute commitments and unusual liquidation terms
Not all of it is traditional equity. A significant share takes the form of compute commitments, structured vehicles with unusual liquidation preferences, and corporate strategic investments that come with commercial strings. These structures complicate the headline numbers, because a company that has raised at a high valuation on preferential terms may have far less real equity value than the headline implies.
Most recipients of every previous wave ceased to exist
Every funding wave of this shape has been followed by a period in which most recipients cease to exist, and the returns concentrate in a small number of winners. Whether that happens here depends on whether the applications arrive before the capital runs out — which is the same question every previous boom has faced.
Research teams, then product companies, then infrastructure
Capital has arrived in waves with different characteristics. The first wave funded research teams and small experiments. The second funded product companies built on other people’s models. The third is funding infrastructure, and it is larger and more concentrated than the first two combined.
The infrastructure wave is different in kind because it produces physical assets with long lives and predictable depreciation. A data centre is not a bet on a business model; it is a bet on demand persisting for a decade.
Returns require the revenue of the software industry
For these investments to return their cost of capital, the applications must eventually generate revenue at a scale comparable to the existing software industry. That is not an impossible bar and it is a high one, and it requires the current growth rate to persist through several phases of the technology.
The honest summary is that the spending is rational given the beliefs of the participants, and the beliefs are not yet verifiable. Both halves of that sentence matter.
Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.