The frontier model race is increasingly an infrastructure race. The latest compute agreement places capacity, energy and long-term supply alongside model quality as the decisions that shape who can build at scale.
For teams in Portugal, the practical question is not whether to reproduce hyperscale infrastructure. It is how to build products that can move between providers, measure their real demand and negotiate capacity before it becomes a constraint.
“Compute is no longer a background service. It is becoming part of the product strategy.”
ai.pt briefing
That shift rewards smaller teams with clear workloads. Products that know when they need a large model, when they can use a compact model and when they should avoid a model call entirely can protect both speed and runway.
Portugal has a particular view onto this change. New Atlantic data-centre projects sit beside a growing software ecosystem, creating a chance to connect infrastructure investment with companies that use it productively.