What Radiant sells — and what it buys
AI factories under long-term contract — for sovereign governments, telecom providers and select global enterprises, spanning fully managed, turnkey AI Factory operations down to high-performance bare-metal delivery, built to the NVIDIA DSX reference design with liquid-cooled 200kW-class racks. Above the factories sits the Ori Global AI Cloud for on-demand capacity: pre-configured bare metal, GPU instances, and as-a-service inference, serverless Kubernetes, fine-tuning, model registries and storage — so Radiant spans from Rung 1 to the tokens rung on one balance sheet.
The two scarcest inputs, one layer below everyone else on this rung: permit-stabilized land banks and behind-the-meter power generation — a mix of hydro, wind, geothermal, biomass and dispatchable gas — plus NVIDIA systems via the DSX design and Ori's inherited NVIDIA Cloud Partner status. It buys no grid power at retail and no cloud capacity at all: Radiant is a power generator, not a reseller, offering fixed-rate PPAs against its own electrons. The capital comes from a direct pipeline to Brookfield's BAIIF program.
Scale proof
The bet
That AI compute becomes a utility — and that in utilities, the durable moats are cost of capital and cost of electrons, not GPU allocation. Everyone on the ladder above pays for power and money at market rates; Radiant's wager is that Brookfield-scale capital plus owned generation structurally undercuts them, the way regulated utilities undercut merchant generators. If token prices keep compressing, margin migrates to whoever holds the lowest-cost land, power and balance sheet — and the Ori software layer exists to convert that infrastructure advantage into product, from bare metal up to inference. The risk is symmetrical: the company entered full operations in February 2026, so the thesis is capitalized but not yet proven in racked, revenue-generating GPUs.
Natural counterparty
Sovereign governments first — national-scale compute programs that want in-country AI factories with island-mode power resilience under planning horizons measured in decades — then telecom providers and select global enterprises on long-term contracts. On-demand builders enter through the Ori Global AI Cloud. On the buy side: landowners and permitting authorities, power developers, NVIDIA (via DSX and NCP status), and — unusually for this ladder — Brookfield's LPs, since the fund structure makes institutional capital itself the capacity-funding instrument.
Buyer fit
You need utility-scale, in-country AI factory capacity under a long-duration contract, with power economics locked by fixed-rate PPAs against on-site generation — or on-demand NVIDIA capacity via Ori's cloud with a path to dedicated infrastructure behind it.
You need large deployed capacity with a public operating track record today — Radiant entered full operations in February 2026, and the headline numbers are pipeline and access, not racked GPUs — or you want per-second serverless or a pure pay-per-token API as your primary unit.
Five fields we track but don’t publish
Everything above is public — figures are Radiant-published unless otherwise sourced, and this card is a draft pending direct provider verification. The fields below change weekly and are verified directly with each provider — they live in a private supply-demand ledger, not on this page.