Nvidia lines up a further $2 billion of European AI deals after $8 billion in a year

Eight billion dollars deployed plus the $2 billion expected makes $10 billion ^[12], and the new tranche is 25% of the trailing year (or rather, of what Nvidia says it deployed, a narrower claim) ^[13]. A quarter added within a week ^[1], at the opening of a three-city tour ^[7], is either an acceleration or a bunching of announcements around the itinerary.

Huang's route for Europe is open models. "Open models are important as they can be customized for languages and for your own sovereignty," he said, according to Bloomberg ^[4]. The argument is about weights, and weights have to be trained. Nvidia is a partner to Mistral AI, which designs open-weight models, and Mistral's latest was trained on 4,000 Grace Blackwell chips ^[5].

The $10 billion can be read as placement of scarce supply, as seeding of the investment Huang's own forecast needs, or, for some part of it, as purchase agreements counted as deals. We think the placement reading fits the evidence best. Nvidia has projected revenue growth of 70% in its next fiscal year, with the possibility of doubling if supply could meet demand ^[10], and a seller that says supply is the limit is not obviously using this money to find buyers. The one non-cash term reported is Nvidia's commitment to help Doctolib train AI models, according to Cryptobriefing ^[17].

The second reading is the serious counter. Huang pointed to AI investment across industries and countries as the driver of the doubling in chip sales he forecast last month ^[9], and a supply-limited seller can still rationally pay to keep that investment coming. The third reading would show up in London and Berlin, where a list made mostly of supply agreements instead of equity would mean the $2 billion measures orders.

We would drop the placement reading if the London and Berlin list is mostly supply agreements, or if the week ends short of the $2 billion Huang described ^[1].