Norges Bank launches Europe's first Enterprise AI tender. We haven't seen anything like it.

By Even, CTO and co-founder of Cobrief

When you work with public procurement every day, you get a fairly unique window into how the public sector actually operates, and which tools it buys. As a company that itself leans heavily on AI, internally, in product development and in the product itself, I have long kept a close eye on how the public sector adopts artificial intelligence.

Honestly? I have seen very little. What does exist is mostly Microsoft agreements where Copilot comes along for the ride, almost as a default in larger organizations.

That is why a notice that recently appeared from Norges Bank made me sit up: Enterprise AI Platform.

What Norges Bank is actually asking for

In the notice, Norges Bank writes that the purpose is to expand its AI portfolio beyond what exists in the Microsoft 365 ecosystem, in order to reduce dependence on a single vendor and secure access to leading agentic AI capabilities built on open standards.

That is a strikingly forward-leaning order from a government institution. They are not asking for "a chatbot" or more Copilot licenses. They explicitly ask for agentic capabilities, open standards and reduced vendor lock-in. The requirements specification shows they have done thorough groundwork, and it could easily serve as a template for other organizations, public or private, weighing a similar purchase.

So I checked: has anyone done this before?

I set Cobrief's AI agent loose to find out whether similar tenders exist, not just in Norway but across Europe. The agent went through the European TED, Finland's Hilma, Denmark's udbud.dk and Norway's Doffin.

The answer: We have not found any comparable competition in TED, Doffin, Hilma or Udbud.

The closest we get is Sykehusinnkjøp HF, which in January 2024 announced a qualification competition using competitive dialogue for a framework agreement on artificial intelligence platforms. That one, however, was about putting AI applications to work in clinical routines, a completely different use profile from the general, agentic workspace Norges Bank is after. And an important caveat: the absence of hits in our searches is not the same as proven absence.

What we do find is a clear pattern. Every public purchase of agentic AI tools we can trace has gone around ordinary competition, via voluntary ex-ante transparency notices and direct awards:

• Norsk Helsenett recently bought Claude Enterprise via direct award, after documenting that the product was not available through existing framework agreements or other suppliers in the market.
• Sikt has published two voluntary ex-ante transparency notices: one for the AI tool Claude and one for the coding tool Claude Code.
• The same pattern shows up in purchases of other AI tools across several public bodies.

The common denominator is that the buyers have concluded that the tools at the front of the pack are simply not available through the usual channels, and used that as objective grounds for buying directly.

Why this is such an important fork in the road

Norges Bank has chosen a different path: a full, competitive process announced on Doffin and TED. That is demanding. Among other things, there is a mandatory requirement of documented experience from at least two deliveries of an Enterprise AI platform of comparable complexity, something few suppliers in Europe can point to today.

But that is exactly why this tender matters. It does two things:

1. It establishes a public, reusable requirements specification for what an organization with high demands on security and complexity expects from an Enterprise AI platform. Other buyers no longer have to start from zero.
2. It sends a signal to the market that there is real, paying demand for agentic platforms outside the established license agreements, and that vendors need to build a sales and bid organization capable of handling public procurement processes in Europe.

The public sector has the most to gain

We already see that several of the suppliers competing in tender procedures are early adopters of the latest in AI. My impression is that the public sector itself still lags behind, and that is a shame, because that is precisely where the productivity gains from agentic tools could be greatest.

If more public organizations follow Norges Bank's example, with thorough, open requirements specifications instead of waiting for the tools to turn up in existing agreements, I believe we could see a real lift in public-sector efficiency. The procurement team at Norges Bank deserves credit here. More should take inspiration.

Read the full notice here: Enterprise AI Platform – Norges Bank

This analysis was done with Cobrief's AI agent, which searches across Doffin, TED and the Nordic tender databases. Try it yourself.