AI infrastructure is the next generational asset class — but the industry is building it with the wrong kind of money. Over the past decade, hyperscalers and emerging GPU-cloud providers have financed 30-year assets using 3-year capital. The result is predictable: a cycle of scarcity, volatile pricing, and under-investment in the physical systems that determine cost and reliability.
Radiant changes this equation.
By pairing infrastructure-grade capital (≈ 5 % cost) with a vertically integrated AI infrastructure platform, we convert compute from a speculative commodity into a stable, utility-like asset.
It has to be this way. Expensive capital delivers expensive intelligence - and with it scarcity. Inexpensive capital deliver inexpensive intelligence - and with it abundance.
The 20 Percent Problem
Every dollar of compute cost ultimately traces back to the cost of capital. The “neo-cloud” market has been built on short-term, high-yield funding structures designed for software startups, not power plants.
The Cost-of-Capital Cascade
At a 20 % hurdle rate, providers must:
- Favor short-duration, high-margin customers over stable offtakes.
- Avoid capital-intensive generation or transmission projects.
- Lease “powered shells” instead of owning real assets.
This model couldn’t fund abundance if it wanted to, but that is what the market is demanding.
The 5 Percent Solution
Radiant finances AI infrastructure the way power utilities, ports, and fiber networks are financed — with long-term, low-cost, patient capital.
Through Brookfield, we access a global balance sheet that prices risk like infrastructure, not venture.
- Hurdle rate: ~5 % vs. 20 % industry average.
- Tenor: 20–30 years.
- Security: hard assets — land, power, and compute — not equity velocity.
This 15-point spread is not incremental margin; it is a new physics of the business.
It realigns incentives from speculation to construction.
Exhibit 2 | Financial Leverage of Capital Efficiency
At 5 %, we can own what others must rent.
What the Spread Unlocks
Behind-the-Meter Power
Energy now represents 30–40 % of AI operating cost.
Most GPU clouds depend on grid power subject to volatile pricing and multi-year interconnection queues. With 5 % capital, Radiant builds and owns generation - solar, wind, small modular nuclear - directly adjacent to compute.
Owning electrons eliminates both volatility and congestion.
Each AI Factory becomes its own self-sufficient micro-utility.
Powered-Land Bank
Land and permitting, not GPUs, are the gating variables for AI infrastructure.
At a 5 % carry, Radiant can augment Brookfield’s massive existing bank of powered land to acquire and hold pre-entitled, power-adjacent land globally — adding to an already rich inventory of compute-ready locations.
Timeline Compression
Time is the highest-order derivative of capital efficiency. We transform a construction problem into a logistics problem.
The Financial Market for Compute
Stable, long-term production costs have a derivative benefit - they make compute a hedgeable commodity.
Radiant can issue 3-, 5-, and 10-year compute offtake contracts priced in $/TFLOP-hour — effectively compute futures.
Hyperscalers can lock in multi-year cost certainty; sovereigns can reserve capacity for national AI programs.
Competitors cannot offer this because their debt matures faster than their GPUs depreciate.
We are building the financial layer of the AI economy—a real market for compute, backed by infrastructure economics.
Sovereign Implications
Compute sovereignty depends on ownership of the supply chain: land, power, and capital.
With 5 % capital, nations can deploy AI infrastructure the way they built power grids — as public utilities, not vendor dependencies.
This changes AI from a strategic vulnerability into a domestic capability. Every sovereign should be racing to this outcome.
Conclusion
The last decade’s cloud was defined by speed and scarcity. The next decade will be defined by duration and abundance.
By aligning financial time horizons with technological lifecycles, Radiant establishes the first platform where intelligence can compound like infrastructure, not depreciate like hardware.
The arithmetic is simple:
This is the math of Radiant:
We are not another cloud; we are a utility for intelligence production.
