Methodology

How we measure, and what the data can't tell us

This page will grow as the dataset does. It's a working document: we'd rather show the limits than overstate the findings.

Source

Our data comes from TED (Tenders Electronic Daily), the EU's official journal for public procurement, run by the Publications Office of the European Union. We use contract award notices published from January 2024 onwards, retrieved through the TED API. Every chart cites TED as its source.

What TED covers, and what it misses

• TED is only part of the picture. It covers tenders above the EU thresholds. Mercell estimates it holds around 20% of all European tender data. National portals, purchases below the thresholds and call-offs under existing framework agreements are largely missing.
• Data quality is uneven. Some awards have no value, some values are outliers or framework ceilings rather than actual spend, and some notices don't name the winner.
• Cloud is usually bought through resellers. The company that wins the contract is often a reseller or integrator, not the cloud provider whose platform is used. In the UK, Tussell could identify only AWS under its own name.

Why we start with AI

For AI contracts, the service and often the supplier are named in the title and description. For cloud, they often aren't. So we start with AI and only add cloud where the data lets us identify the actual provider.

Before publishing any findings, we're running a data test on a random sample of award notices to measure how often the winner, the value and the underlying provider can be identified. The results will be published here.

Classification

Notices are selected by CPV code and keywords in the main EU languages, then classified with a large language model (Claude, by Anthropic) by category, vendors mentioned and whether the underlying provider can be identified. We keep the prompt, model and version for every classification, and check a manual sample against the model's labels. Accuracy will be reported here.

Colour code

In every chart, blue means US hyperscalers and AI labs, orange means European providers, and grey means integrators and resellers.

What's next

The EU's Public Procurement Data Space (PPDS) links TED with national portals and could fill many of the gaps above. We'll add it when it's usable.