Investment thesis
From Megawatts to Intelligence
AI infrastructure is moving from a scarcity market, where megawatts and GPUs dominate valuation, toward an operating market, where value depends on contracted demand, utilisation, margin, sovereignty and evidence.
Outpace is being built to sponsor and operate selected projects through that transition.
- 01
AI demand is real, but demand growth does not guarantee attractive returns for every data-centre or compute project.
- 02
Megawatts and GPU counts are inputs. Returns depend on delivery, utilisation, customer quality, financing cost, hardware cycles and operating discipline.
- 03
The value chain is shifting from power and capacity toward inference, orchestration, sovereignty, provenance and enterprise assurance.
- 04
The commercial model must connect customer demand to contract, contract to bankability, bankability to project capital, capital to build, build to utilisation and utilisation to cash flow.
What is changing
Demand is real. Returns are not automatic.
Demand for AI and digital infrastructure is growing, and supply is constrained by grid access, capital intensity and execution capacity. Neither fact makes any single project a good investment.
Data-centre electricity use in 2024 About 415 TWh
Global data-centre electricity consumption in 2024, around 1.5% of global electricity.
Source: International Energy Agency, Energy and AI, 2025
IEA base case for 2030 About 945 TWh
The IEA base case for data-centre electricity consumption by 2030, more than double the 2024 level.
Source: International Energy Agency, Energy and AI, 2025
Planned projects at risk of delay About 20%
The IEA's estimate of planned data-centre projects that could face delays, with grid constraints a leading cause.
Source: International Energy Agency, Energy and AI, 2025
| Market signal | What it supports | What it does not prove |
|---|---|---|
| Rising AI and digital demand | A multi-year infrastructure build cycle | That every announced project will reach utilisation |
| Power and grid constraints | Value in executable power pathways | That power access alone creates durable pricing power |
| Large capital requirements | A need for institutional equity and debt | That capital will accept uncontracted development risk |
| Inference growth | Demand for continuous, distributed compute | That a specific inference platform has product-market fit |
| Low installed rack density | Demand for purpose-built AI-grade facilities | That retrofits cannot compete on price for some workloads |
The valuation trap
Megawatts are an incomplete measure.
Enterprise value per megawatt is useful for comparing physical development, but it conceals the variables that determine the cash yield of AI infrastructure.
A megawatt can be secured but not energised, energised but not fitted out, fitted out but underutilised, or fully utilised by a weak counterparty on uneconomic terms.
| Metric | Question answered | Key limitation |
|---|---|---|
| MW secured | Is a power pathway available? | May be conditional, delayed or uneconomic |
| MW energised | Can the site operate? | Does not show compute fit-out or customer demand |
| GPU-hours available | How much compute can be offered? | Hardware may age before utilisation matures |
| Utilised compute | Is the asset being used? | Utilisation alone does not show margin or credit quality |
| Gross profit per MW | What operating value is created? | Needs consistent cost allocation and hardware depreciation |
| Risk-adjusted cash yield | What does invested capital earn? | Depends on financing, contracts and execution |
The four-layer system
Power to intelligence, layer by layer.
Four layers convert megawatts into intelligence. Each plays a different economic role, and value accrues to whoever controls the scarce element in that layer.
- 01
Power
Available electrical capacity, reliability, connection timing and underlying energy economics.
- Core question
- Can sufficient reliable power be secured at the required time and cost?
- What it includes
- Grid access, generation, power purchase agreements, connection schedule, firmness, resilience, cost per megawatt-hour and energy provenance.
- Outpace's role
- Originate access, underwrite the economics and structure power partners.
- 02
Data Centre
The physical facility that converts power into resilient, cooled and connected technical capacity.
- Core question
- Can power be converted into AI-ready physical capacity on time and on budget?
- What it includes
- Land, planning, substations, power train, cooling, racks, fibre, physical security, commissioning and facilities operations.
- Outpace's role
- Sponsor ProjectCos, secure capital, shape delivery and retain selected project equity.
- 03
Compute
The technology stack that converts facility capacity into usable AI processing.
- Core question
- How efficiently is data-centre capacity converted into utilised compute?
- What it includes
- GPUs and accelerators, CPUs, storage, memory, networking, clusters, scheduling, virtualisation, model serving and fleet management.
- Outpace's role
- Structure procurement, capacity and operating relationships, with refresh and obsolescence discipline.
- 04
Intelligence
Commercially valuable AI workloads, services, decisions and outcomes that customers will pay for.
- Core question
- Does the compute produce useful, contracted and profitable intelligence?
- What it includes
- Inference, training, enterprise workloads, sovereign capacity, dedicated environments, customer outcomes and recurring contracts.
- Outpace's role
- Secure demand, commercialise capacity, improve utilisation and retain recurring operating economics.
Economic result
- Customer demand
- Utilisation
- Gross profit
- Investor return
The economic result sits below the chain. It is an outcome of the four layers, not a fifth layer.
| Layer | Economic role | Where value may accrue |
|---|---|---|
| Power | Makes the facility physically possible | Executable grid rights, low-cost power, connection timing and planning certainty |
| Data Centre | Converts power into resilient technical capacity | Delivery capability, cooling, reliability and customer fit |
| Compute | Converts facility capacity into usable AI processing | Hardware access, fleet management, utilisation and refresh discipline |
| Intelligence | Turns compute into contracted capacity and useful AI output | Contract structure, pricing, workload value, retention, assurance and margin |
Scarcity migration
The constraint is moving up the stack.
Capital tends to arrive after the highest-return bottleneck has moved. Outpace treats power as a gate, not as a source of return on its own, and focuses on the constraints that come next: bankable demand and trust.
| Constraint | Period | Who captured value | Outpace view today |
|---|---|---|---|
| Accelerator access | 2023 to 2024 | Chip vendors and early GPU cloud operators | Easing as supply broadens |
| Power and grid | 2024 onward | Owners of executable grid rights and powered land | Binding. A gate, not an advantage on its own |
| Bankable demand | Emerging | Sponsors who can contract credible offtake | Scarce. Much announced capacity is uncontracted |
| Trust: sovereignty and assurance | Emerging in regulated sectors | Operators who can evidence jurisdiction, control and provenance | A premium that must still be tested with buyers |
This table is Outpace's view of the market, not an independent finding.
Intelligence Yield
Connect physical capacity to the value it produces.
Intelligence Yield is Outpace's working economic framework. It is not an audited industry standard, and Outpace does not present it as one.
It forces each project to connect physical capacity with contracted customer value, and with the capital employed to produce it. In plain terms: risk-adjusted gross profit from contracted AI workloads, relative to total capital employed.
Not all tokens have equal value. Capacity produced for a regulated bank under sovereign controls is not economically equivalent to commodity batch inference.
- Physical efficiency
- Useful compute delivered per energised megawatt
- Utilisation
- Billable workload hours against available hours
- Commercial yield
- Revenue and gross profit per energised megawatt
- Capital yield
- Gross profit per dollar of total capital employed
- Assurance premium
- Incremental margin from sovereignty, privacy, provenance or dedicated capacity
- Risk adjustment
- Customer credit, contract term, concentration, refresh and financing risk
Assured capacity
Premium attributes need defined evidence.
Energy and sovereignty attributes can become part of the commercial product, but the wording must be precise. Grid mix, renewable certificates, annual matching, hourly matching and on-site generation are not equivalent.
| Product attribute | Evidence required |
|---|---|
| Low-carbon capacity | A defined carbon-intensity method, boundary and reporting period |
| Renewable-matched capacity | The certificate or contract basis and the matching interval |
| On-site renewable capacity | Metered generation and a consumption boundary |
| Sovereign capacity | Execution location, administrative control, personnel and data-handling evidence |
| Assured capacity | Independent controls, telemetry and evidence of operating conditions |
What would prove this wrong
A thesis should state its own failure conditions.
Outpace's view rests on conditions that can be tested. If they do not hold, the thesis does not hold.
| What must be true | What would invalidate it | Implication |
|---|---|---|
| Relationships convert into enforceable rights | Access does not become contractual project rights | Access does not translate into enterprise value |
| Enterprise demand converts into instruments | No credible buyer commits to a letter of intent or term sheet | Projects remain speculative and difficult to finance |
| Differentiated capacity earns a premium or a lower cost of capital | Buyers pay commodity prices for sovereign or assured capacity | The operating layer does not justify its cost |
| Project economics survive realistic debt service | Projects cannot service debt without relying on optimistic assumptions | Operating economics do not support the capital intensity |
| Financing cost stays below project yield | Debt cost rises faster than contracted pricing | Growth destroys equity value |
| Hardware refresh is funded within the model | Refresh timing or cost exceeds what the project can carry | Residual value and cash yield decline |
The value lies in the conversion.
A generic project with uncontracted demand produces modest and fragile returns. Outpace's role is to convert access into contracted, bankable projects. The case for Outpace strengthens only as that conversion is evidenced.
Build the next infrastructure project with Outpace.
Outpace works with project owners, infrastructure partners, capital partners and enterprise customers.