THE SHORT ANSWER
● Government — £100 million cash, plus £65 million for pathogen countermeasures
● OpenAI and Anthropic — Up to £200 million in compute credits, model access and staff
● Spendable anywhere — Only the £100 million
● Headline figure — £300 million, mixing two kinds of commitment
Britain's new AI biosecurity lab is being reported at roughly £300 million. The composition matters more than the total.
| Government £100M | Lab commitments, up to £200M |
|---|
| Form | Cash | Compute credits, model access, staff |
| Spendable on | Anything | One vendor's infrastructure |
| Survives a vendor change | Yes | No |
| Appears in a budget as | Expenditure | In-kind |
| Withdrawable | No, once committed | Depends on terms, unpublished |
| Buys frontier model access | Not at this price | Yes |
The case for credits
A biosecurity lab needs frontier model access, and frontier model access at research volume is not something £100 million buys outright at list rates. Credits deliver the thing the lab actually needs, at a price no procurement process could negotiate. OpenAI's contribution runs through its Rosalind Biodefense programme, which means dedicated capability rather than generic API quota.
There is also precedent that this works. CERT Polska used Daybreak model access from OpenAI to find six MikroTik RouterOS vulnerabilities, some chainable to unauthenticated device takeover. That was credits, not cash, and it produced patched infrastructure.
The case for reading it carefully
Credits create a dependency that cash does not. A lab built on one vendor's models develops processes, tooling and institutional knowledge specific to those models. If the terms change, the alternative is not "spend the money elsewhere" - the money was never there.
Nor are the terms public. Duration, renewal, what happens at £200 million, and whether the commitment survives a change in either lab's commercial priorities are all unstated. For a facility meant to operate across decades of biological risk, that is the detail to watch rather than the headline.
The context it landed in
Two days earlier, the Council for Science and Technology wrote to AI minister Kanishka Narayan warning that Britain risks becoming a "high-value AI customer" rather than a creator, and recommended investment in domestic compute and a publicly governed British model. A lab funded substantially in credits from two American companies is either a pragmatic answer to that warning or a demonstration of it, depending on how much weight you put on who owns the infrastructure.
How to read the next one
Three questions settle most of these announcements. How much is cash? What can the non-cash portion be spent on? And who decides whether it continues? Apply them and a £300 million headline resolves into £165 million of public money and a commercial relationship.