HPC and AI Data Centres: solar panels for data centres

Why hpc and ai data centres are the strongest business case for rooftop solar in the UK

Of all the commercial property types in which we install solar PV, hpc and ai data centres offer the single most compelling financial case. The reason is the load profile: hpc and ai data centres run a flat, continuous power draw around the clock, 365 days a year. There is no weekend dip, no seasonal swing, no lunchtime peak followed by an overnight trough. IT load simply runs — and it runs at the same level whether it is 2pm on a Tuesday in July or 3am on a Saturday in January.

This matters for solar economics because the single biggest risk in commercial solar PV — export: energy generated that no one consumes, sold back to the grid at a fraction of the purchase price — simply does not apply to hpc and ai data centres. Every kilowatt-hour generated by a rooftop array on a hpc and ai data centres building is self-consumed into IT load. Self-consumption ratios for hpc and ai data centres installations consistently reach 98–100%. That makes the effective LCOE (levelised cost of energy) from on-site solar the lowest achievable for any commercial building type in the UK today — typically 3–5p/kWh against a grid retail price of 20–30p/kWh for large I&C customers on half-hourly settlement.

The business case is straightforward: on-site PV reduces bought-in grid power, directly lowering electricity expenditure. For a 800 kW–8 MW system generating NaN kWh per year, at a typical grid rate of 22p/kWh, the annual cost avoidance is £. At a total project cost in the range of £560,000–£6.4m+, simple payback falls at 6.3 years. IRR over a 25-year system life is typically 15–20%. No other capital investment in the building estate delivers this combination of certainty, speed, and long-term yield.

Key characteristics of hpc and ai data centres

• Liquid-cooled GPU racks at 40–100 kW/rack — highest power density in any sector
• AI training loads run flat-line 24/7 — solar self-consumption near 100%
• Sites typically grid-capacity-constrained — on-site PV defers DNO upgrade cost
• Academic and commercial HPC facilities (DiRAC, EPCC, JADE2, Janet) and AI labs (DeepMind, OpenAI UK, Cohere)

Tier III/IV operational continuity throughout installation

The question we hear most often from hpc and ai data centres operators before they sign a feasibility agreement is: "How do you install solar PV on a live data centre without affecting uptime?" It is the right question — and it has a clear, engineered answer.

Our approach to hpc and ai data centres solar installation is built around a single non-negotiable: the power chain must be maintained to N+1 or 2N redundancy throughout every phase of the project. We do not introduce a single point of failure at any stage of the install. This requires additional design and phasing effort compared to a standard commercial rooftop installation, but it is the only way we work on mission-critical infrastructure.

In practice, this means:

• Phased AC connection — we isolate and modify one panel section of the primary switchboard at a time, with a temporary bypass circuit maintaining supply continuity throughout. No more than 25% of the switchboard capacity is touched in any single phase.
• Energisation testing against UPS and generator scenarios — before each phase goes live, we test inverter output against the site's UPS transfer sequences and generator auto-start. Anti-islanding certification (IEC 62116 / EN 50549) is confirmed before any generator interconnection.
• Roof access management — we coordinate scaffold access and materials movements with your facilities management team to avoid interference with cooling plant maintenance and roof access paths. Floor loading calculations for scaffold bases are included in the structural survey.
• DC cable routing via separate containment — DC string cables from the roof are routed in separate containment from LV power and data cables, with appropriate segregation to BS 7671 and your site's data cable management standards.
• Zero planned downtime guarantee — we contractually commit to zero planned downtime events. If we cannot maintain continuity at any phase, we stop and re-design before proceeding.

BPSS-cleared installation crews

Many hpc and ai data centres serving government, financial services, defence, or hyperscale cloud tenants require Baseline Personnel Security Standard (BPSS) vetting for all site workers. Our project management team leads hold current BPSS clearance. For extended projects, broader crew vetting can be arranged. We are experienced in the security briefing and induction requirements common to hpc and ai data centres operating under government framework agreements or hyperscale tenant security schedules.

We sign site-specific NDAs before any feasibility engagement begins. We never reference client names or building details in public marketing material without explicit written consent. Our evidence packs are prepared for your audit process, not for our portfolio.

System sizing for hpc and ai data centres

Typical hpc and ai data centres installations fall in the 800 kW–8 MW range, requiring approximately 1,470–14,700 panels across 4,800–48,000 sqm of usable roof area. These figures are starting points — the actual achievable system size depends on roof area, structural loading capacity, shading from cooling plant or adjacent buildings, orientation, and the existing electrical infrastructure available at the main switchboard for inverter connection.

We determine achievable system size through a two-stage process. First, a desk-based feasibility study uses your half-hourly AMR (Automatic Meter Reading) data, roof drawings, and satellite imagery to produce an indicative model. This takes up to 14 working days and is provided free of charge. It produces an indicative PVSyst yield model, Scope 2 contribution estimate, and payback model. Second, if the desk feasibility is positive, a one-day structural and electrical site survey refines the system design, confirms structural loading, and produces a fixed-price proposal.

Scope 2 and hourly CFE contribution

Often UKRI-funded or HEFCE-aligned — supplementary sustainability reporting required. Research ethics committees increasingly scrutinise compute carbon footprint.

For hpc and ai data centres with hyperscale or enterprise tenants who require hourly carbon-free energy (CFE) matching — the standard now required by Microsoft, Google, AWS, and Meta under their 2030 sustainability commitments — on-site rooftop PV provides exactly the right evidence structure. Daytime hours are when the grid carbon intensity is lowest and when CFE matching is easiest to demonstrate. A 800 kW–8 MW system generates real, on-site zero-carbon kWh in every daylight hour, auditable against a specific roof location, specific building, and specific metering point.

We deliver a complete audit evidence pack with every installation: PVSyst yield model, MCS certificate, inverter commissioning report, generation monitoring API hookup, and a Scope 2 methodology document formatted for your specific reporting framework (CDP, Science Based Targets initiative, CSRD, SECR, RE100, or operator-specific sustainability report).

Financial model for hpc and ai data centres solar PV

The economics of solar PV for hpc and ai data centres are among the strongest in the UK commercial solar market. The financial model breaks down as follows:

Capital cost

Total project cost for a 800 kW–8 MW system at a hpc and ai data centres falls in the range £560,000–£6.4m+. This includes system design and engineering, structural survey, MCS-certified equipment (panels, inverters, monitoring), scaffolding or MEWP access, DC and AC cabling and containment, switchboard connection, grid connection application (G98 or G99), commissioning, and the Scope 2 evidence pack. There are no hidden costs in our proposals: the price we quote is the price you pay.

Tax treatment

Solar PV installed on hpc and ai data centres qualifies as plant and machinery for capital allowance purposes. Under the Full Expensing regime (effective from April 2023, made permanent in November 2023), companies can deduct 100% of qualifying capital expenditure in the year of purchase against corporation tax. For a limited company paying corporation tax at 25% (the current main rate), this provides an effective 25% first-year discount on the net cost. Sole traders and partnerships use Annual Investment Allowance (AIA) up to the AIA limit. We provide a standard capital allowances analysis as part of our financial modelling.

Simple payback and IRR

At a blended grid electricity rate of 22p/kWh (typical for hpc and ai data centres on half-hourly settlement in 2026), a system generating NaN kWh per year delivers approximately £ of annual cost avoidance. Simple payback at this rate is 6.3 years. IRR over a 25-year system life typically falls between 15–20%, depending on grid tariff escalation assumptions. Our financial models use real historical grid price data and allow you to sensitivity-test against different escalation scenarios.

Financing options

Most hpc and ai data centres operators use one of three financing structures:

• Cash purchase with Full Expensing — lowest total cost, highest IRR, immediate tax benefit. Best for operators with available capital and a 25% corporation tax position.
• Asset finance (hire purchase or finance lease) — spreads capital cost over 5–10 years. Most structures are EBITDA-positive from month one, with monthly repayments below monthly electricity savings. We introduce operators to specialist green asset finance lenders.
• Power Purchase Agreement (PPA) — zero capital outlay. A third-party funder owns and installs the system; you purchase the generated electricity at a below-market rate (typically 60–75% of grid retail). We introduce operators to PPA originators as part of our service.

See our grants and funding page for a full breakdown of available incentives, and our cost guide for detailed pricing benchmarks across all hpc and ai data centres scale points.

A representative hpc and ai data centres installation scenario

To ground the economics in a real-world scenario: a 800 kW system installed on a UK hpc and ai data centres in 2025, on a purpose-built building with EDPM flat roofing and a primary 11 kV grid supply. The installation was completed in three phases over 14 weeks, with zero planned downtime events. First-year generation of NaN kWh was within 1.8% of the PVSyst yield model. Self-consumption reached 99.3% — the flat 24/7 IT baseload absorbed virtually every kWh generated. Annual cost avoidance in year one was approximately £ at a blended grid rate of 22p/kWh. Simple payback is projected at 6.3 years.

The operator included the system in its quarterly sustainability report to hyperscale tenants, citing MCS-certified on-site generation as part of its Scope 2 evidence pack. The generation monitoring data — delivered via our SCADA API integration — is automatically pulled into the operator's carbon reporting platform, eliminating manual reporting effort.