Main finding
Building the same capability independently means funding a multi-year frontier programme whose outcome is uncertain. Licensing buys time and removes most of the technical risk: months instead of years, roughly a tenth of the capital, and a provider who carries the cost of failed experiments. The build option is not lost, and exit terms are what keep it alive.
What supports it
- The cost comparison is arithmetic, not a forecast: $100M to $300M of frontier development against a $20M three-year licence, both as planning estimates.
- Time-to-capability is the decisive variable: a frontier programme of 18 to 30 months begins delivering value long after a licensed integration, which starts earning in the first year.
- Frontier capability has concentrated in a small number of well-resourced teams, which is why the realistic alternatives are licence, narrow applied build, or doing neither (Epoch AI; Stanford AI Index).
- A licence is a usage right for a defined term, not an asset: it does not transfer ownership, and it ends unless renewed on agreed terms.
What this does not claim
- The $20M figure is an internal planning assumption. No supplier has quoted it, and negotiation may move it materially.
- Licensing creates provider dependence. If the platform is withdrawn or degrades, capability leaves with it.
- A licence can lag behind general models that keep improving on commercial terms, so the advantage has to be re-tested at renewal.
Planning assumptions used
- The platform is operational today and remains available for the full three-year term.
- Integration effort is costed inside the licence, not treated as free.
- A narrow internal build remains possible for one or two capabilities if they prove decisive.
Open before signature
- What is the earliest date at which a licensed integration produces a paying service?
- Which capability must we be able to rebuild if the provider disappears?
- How do exit, portability and price-at-renewal terms protect the decision?
Sources
Figures marked illustrative are planning assumptions. This note separates published evidence from project assumptions and does not present a supplier quotation, a valuation or a forecast of returns.