A European bank that wants to run a model over customer records has two questions before it has a use case. Where does the data sit, and who can be compelled to hand it over. Databricks published its answer this week, and the answer is to build more of the machinery inside the region.
The company said on Nov. 15 that it will invest $3 billion in Europe, the Middle East and Africa over three years. The money goes toward regional cloud capacity, local data-residency requirements and training for AI workers. Databricks did not say how the $3 billion splits across the three years, which leaves the annual shape of the plan open.
The timing follows a raise. In August the company closed $5 billion in financing at a $190 billion valuation, and it has told investors that growth will come from enterprise data and AI agents rather than from new customers alone. A regional capacity plan is what that story looks like when it is converted into buildings and contracts.
Data residency is a purchasing question before it is a technical one. A bank, a hospital group or an agency cannot buy a service that moves records across borders without a legal opinion, and those opinions have become harder to obtain since a European court struck down the transatlantic transfer pact that most companies had been relying on.
Renting capacity from a cloud provider inside the region solves part of it and leaves the rest. The contract still names an American parent, and the support staff who can see a customer’s data are often the ones who sit outside the region. Buyers have started asking about that second part, which is harder to answer with a data center.
Databricks runs inside whichever cloud and region the customer chooses, but the billing relationship, the control plane and the account team remain its own. The regional build-out narrows the gap between where the data sits and where the company sits. It does not close it, and the company has not claimed otherwise.
The talent line is the vaguest of the three. Training AI workers can mean a university partnership, a certification course or a discount on consulting hours, and the announcement does not say which. Vendors fund these programs for an operational reason: a customer that cannot hire a data engineer does not expand a contract, however good the platform is.
The Middle East appears in the same plan for a different reason. Gulf governments are paying for sovereign cloud and domestic computing capacity, and they buy from vendors willing to put equipment in the country and staff in the building. That is a procurement culture in which a regional commitment is the entry ticket.
For Databricks the plan also answers a competitive problem. Snowflake, Microsoft, Google and SAP all sell data platforms in Europe, and all of them can point to local regions and local staff. Residency has moved from a differentiator to a line item that every vendor has to tick.
The European alternatives are the newer pressure. Mistral sells itself as a European builder of models. SAP can offer a data platform inside a cloud run from Germany. For a regulated buyer, a local vendor removes a category of legal risk, and the cost of that removal is usually some capability the buyer has to live without.
The economics of the announcement are harder to read than the headline. A three-year commitment is not a cheque written today: much of it will be cloud spending that lands in cost of revenue, and some of it will be capacity a partner has already agreed to build. What the plan does not disclose is how much is new money.
Investors will look at gross margin. A company that runs more of its own regional infrastructure, rather than reselling somebody else’s, carries more fixed cost and less room to flex when demand moves. Residency requirements are not negotiable for the buyers Databricks wants, so the margin question has an obvious answer and an expensive one.
There is a broader point about the software business in this cycle. The model is not what a regulated customer is buying. The customer is buying a contract with a jurisdiction attached, an audit trail, and a support phone number that answers in a European time zone.
That is why the largest AI vendors now spend like infrastructure companies while describing themselves as software companies. The product has a location, and the location has a legal system, and the legal system has to be satisfied before a pilot becomes a production system.
Databricks has made a three-year bet that satisfying it is cheaper than losing the deal. The plan does not say what happens in year four, when the capacity is built and the contracts come up for renewal, and whether the customers who arrived for residency stay for the platform.
For the bank, the two questions are still open. It will read the answers in a document a regulator can accept, with an address in the region on the letterhead, and it will decide whether the address is enough.


