Nscale recently announced it will deliver over 66,000 of NVIDIA’s next-generation Rubin GPUs to Microsoft's Sines data center campus in Portugal, starting in late 2027.
This massive deal, estimated to be worth between $5.5 and $8.1 billion in hardware alone, is a direct response to the immense pressure Microsoft faces. The company has been transparent about its AI infrastructure being capacity-constrained amid soaring demand. This pre-order for 2027 isn't just a purchase; it's a strategic move to lock down the essential computing power needed for its Azure AI services well into the future, ensuring it can compete in the fierce AI arms race.
Furthermore, this Portugal deal doesn't stand in isolation. It's the latest in a series of similar agreements that reveal a clear global strategy. In the preceding months, Microsoft and Nscale announced major Rubin GPU deployments in Norway and West Virginia. This pattern shows the two companies are standardizing on NVIDIA’s NVL72 rack-scale system for their global AI factories. It’s a coordinated, multi-country effort to build a consistent and powerful AI infrastructure network.
The foundation for this specific project was laid long before the announcement. It traces back to crucial groundwork that made such a large-scale deployment feasible. First, Microsoft signaled its long-term vision in November 2025 with a $10 billion investment commitment to the Sines AI hub. Second, the partners established an operational rhythm by deploying previous-generation NVIDIA GPUs at the site. Third, the data center operator, Start Campus, has been securing the necessary power and connectivity infrastructure for years, preparing for a 'giga-scale' project like this.
In essence, the causal chain is clear. The project began with securing fundamental site infrastructure, followed by establishing a working partnership with initial deployments. This culminated in a massive, forward-looking investment in next-generation technology, driven by the urgent need for more AI capacity. This deal is a milestone that de-risks both NVIDIA’s Rubin product pipeline and Microsoft's European AI expansion.
- GPU (Graphics Processing Unit): A specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device. They are essential for the heavy parallel processing required by AI models.
- Capex (Capital Expenditure): Funds used by a company to acquire, upgrade, and maintain physical assets such as property, buildings, an industrial plant, technology, or equipment.
- NVL72 Rack: A server rack designed by NVIDIA that integrates 72 Rubin GPUs and other components into a single, powerful system optimized for large-scale AI training and inference.