Permitting Friction Drives International Buildouts

AI compute is starting to follow permitting friction as much as customer demand. For Together, that means the fastest way to add capacity is no longer just finding more GPUs, it is finding places where power, land, and approvals can actually clear on schedule. That is why Sweden, Europe wide Blackwell deployments, and the Humain build in Saudi Arabia matter as operating capacity, not just geographic expansion.1 2 3 4 5

• This is part of a broader shift across GPU cloud. Fluidstack moved from short term startup rentals to a 1 gigawatt French sovereign AI cluster with 500,000 chips. Crusoe is co building Stargate in Texas and leasing capacity through Oracle. The biggest winners are increasingly the firms that can secure land, power, and permits at national scale.2 3
• The U.S. bottleneck is real, even if uneven. Local opposition groups now span 40 states, cancellations jumped to 25 projects in 2025 from six in 2024, and policy trackers identified dozens of local moratoriums plus state level pauses and proposals. In practice, developers are spending more time solving zoning, grid, and water issues before a single GPU goes live.5 6 7 8
• For Together specifically, international buildout also improves the product. Its Sweden region cuts latency for Northern and Central Europe and satisfies EU data residency, while the Hypertec and 5C partnership gives it a path to deploy Blackwell capacity in France, the UK, Italy, and Portugal. More regions means faster inference and fewer enterprise compliance blockers.1

The next phase of the neocloud market will be won by companies that look less like simple GPU resellers and more like global infrastructure organizers. Together is moving in that direction. As U.S. approvals get slower and sovereign compute demand rises, more of the decisive capacity buildout is likely to happen in Europe and the Middle East first.1 2 4 8