THE STRATEGY
If you own the compute you own the outcome
Software competes on price against bigger incumbents. Infrastructure does not. Raion raises institutional level capital and deploys it into physical AI infrastructure: compute that our partners directly own with us.
Why This Position
Raion partners directly with the owners of energy — not brokers, aggregators, or intermediaries standing between capital and the power source. That relationship is the scarce resource.
THE THESIS
The capital allocator for sovereign compute.
The largest asset managers built their position by becoming the default route into an asset class, the entity capital moves through, not around. Raion is built on the same logic, applied to physical AI infrastructure: the counterparty leading institutions and governments call us when they want direct ownership of compute, structured and custodied to institutional standard, not indirect exposure to whoever happens to operate it.
“This isn’t a thesis about renting cycles from hyperscalers. It’s a thesis about who owns the machines the next decade of AI runs on, and how that ownership gets financed, titled, and governed.”
Laurence Cohen
THE POWER CONSTRAINT
GENERATION 1 MODULES
0.5MW
NEXT-GEN TARGET SPEC
~2MW
The hyperscale build-out has become a grid, water, and community challenge. Distributed, transportable modules carry a fraction of that footprint: each draws little enough power to connect without straining a single point of the grid, and deployments spread across many sites instead of concentrating demand in one place.
THE OWNERSHIP LAYER
You own the server rack. Not the frontier labs.
Raion finances physical modules, sited and powered, then structures the capital behind them so the holder has a direct, traceable claim on the hardware itself. This is the difference between renting the future and owning the machine that builds it.
Cloud compute is an allocation:
GrantedMeteredRevocable
Direct Title, Not Synthetic Exposure
Each module sits inside a segregated structure with the capital behind it. Investors and sovereign partners hold a claim that traces to physical GPUs on a secured site, while the buyers of that compute get dedicated hardware with direct access rather than a virtualised allocation. A better product for the client, and a stronger asset behind the claim.
Government and Commercial, Same Standard
Deployment supports national compute mandates including data residency, export-control-clean supply, and sovereign capability, alongside private commercial buyers who need dedicated infrastructure they don’t have to share, negotiate for, or wait on.
Deployed Where the Economics Are, Not Where the Lease Already Is
Modules are standardized and portable: sited wherever power, land, and sovereign mandate make sense, not constrained to the footprint a single hyperscaler already committed to.
HARDWARE INFRASTRUCTURE
Direct Ownership of Hardware
Segregated Custody
Independent Audit
Institutional Reference Minimum
Module-Backed Mandate
Modular Powered Shells
Tier-1 Hardware Spec
No Co-Mingled Assets
Direct Title to Hardware
Special Purpose Vehicle landscape
Launching our own fund doesn’t make sense yet
Ownership takes different legal forms depending on who's investing and how. The platform is a small family of vehicles, introduced in sequence — each with a recommended default and the condition that would change it.
HOW WE OPERATE
We Invest in AI.
Our Operation Is AI.
AI runs the back and middle office under human-in-the-loop control whilst humans make every decision that matters.
THE INSTITUTIONAL STANDARD
Insights & Resources
Research notes, structural memos, and diligence frameworks on the financial architecture of AI infrastructure.
REQUEST MARKET REPORT
Ask for the Phase One study before capital moves.
Structure & Economics. Operating model note. AI-native operations note. Leadership & team note. Available on request.
