Whitepaper
Sovereign AI, on infrastructure you control
What sovereignty means at every layer of an AI deployment, and how to get it without waiting years for capacity.
- Hardware
- Single-tenant
- Clusters from Q1 2027
- US, EMEA, APAC
- Interconnect
- 17 facilities, 9 countries
- Powered land
- Up to 3 GW
01What it means
What sovereign AI means
Sovereign AI is AI that a country or an organization controls end to end: where the hardware sits, who can touch it, which law applies and where the data travels.
It is decided layer by layer. These are the five layers of an AI deployment, and how we deliver control at each.
- 01
Power and land
Where does the capacity physically sit?
Powered land and campuses in the markets you choose, up to 3 GW.
- 02
Data centers
Who can enter the hall and touch the rack?
Private cages and halls; our access controls are part of the security review under NDA.
- 03
Compute
Whose hardware, whose kernel?
Single-tenant bare metal by default, with kernel-level control.
- 04
Network
Which routes, across which borders?
Our own network, AS215724, with programmable peering to route traffic on your terms.
- 05
Control
Who operates it, and how?
EdgeView and the API, so your team sees and steers every layer.
02Residency
Residency is not sovereignty
Keeping data in a country is where sovereignty starts, not where it ends.
Residency settles
- Where the data is stored
Sovereignty also settles
- Who can touch the hardware
- Which law applies
- Where the traffic goes
A shared cloud region can keep data in the country and still leave those questions open.
Which law governs your contract and your data is set for each deployment and reviewed with your team under NDA.
03Deployment options
Choose how far isolation goes
From a dedicated cluster in a region to a campus of your own.
| Dedicated cluster | Private cage or suite | Inside telecom sites | National campus | |
|---|---|---|---|---|
| Where the data sits | A region you choose | A facility you choose | In-country operator sites | A site you choose |
| Hardware | Single-tenant | Your racks | Set per site | Your campus |
| Network | AS215724, programmable peering | AS215724, programmable peering | Set per site | AS215724, programmable peering |
| Live in | 10–12 weeks | Set per facility | Set per site | 250 MW within 12 months |
04Where
Capacity in the jurisdictions you need
Our network meets the internet at 17 interconnect facilities in 9 countries; dedicated clusters open in the US, EMEA and APAC from Q1 2027. Need another country? We build to suit.
Interconnect17
8 North America · 6 Europe · 3 Asia-Pacific
In development9
3 North America · 3 Europe · 3 Asia-Pacific
As of October 2026. Illustrative map: points show where our footprint is, not exact sites.
05Controls you can verify
What you can check today
Audits, staff access and key management are covered in the security review, under NDA.
- Single-tenant by defaultYour hardware is yours: no noisy neighbours and no shared kernels.
- Kernel-level controlBare metal with your own kernel, drivers and tuning, nothing in between.
- Security pack under NDAArchitecture, controls and DPA, ready for your review.
- No egress feesMove data in and out without bandwidth penalties.
06Who it’s for
For those who answer for the data
07How we get there
Five steps to sovereign capacity
Requirements
Your data classes, jurisdictions and scale.
Site match
Regions, facilities or a new campus that fit them.
Security review
Architecture, controls and DPA, under NDA.
Deploy
Dedicated clusters first; campuses in phases.
10–12 weeks to a cluster
Operate
Through EdgeView and the API, with financing or a joint venture if you need it.
10–12 weeks to a cluster