Flexible compute for Iberian power producers

Turn renewable power into AI compute revenue.

TCP develops modular AI compute at renewable-power sites in Portugal and Spain. Start with contracted batch inference, fine-tuning or other workloads whose power envelope is agreed with the customer. Scale after validating demand, connectivity, cooling and supply continuity. Interruptible ASIC mining remains a separate option for surplus-only sites.

Iberia

One site. Two markets.

Sell electricity locally when the price is attractive. When verified prices and site conditions make export weak, flexible compute can create a second route to market. A zero or negative day-ahead price is a market outcome; curtailment is a separate physical or operational event. Renewable sourcing matters mainly as a decarbonization and social-license benefit; the commercial case still depends on actual pricing, contracts and operations.

Global compute demandAI/HPC contracts and crypto networks can value digital output beyond the local electricity market. Revenue still depends on real contracts, utilization and market conditions.Contract and market dependent

TCP product scope

Three standard product scopes

AI compute leads all three scopes. Facility references use a provisional 20% nameplate ratio and PUE 1.20; verified power availability and interconnection headroom determine final IT capacity. Example budgets below include GPU ownership and are not offers.

01

Pilot

0.5–2 MW
Target IT load
0.5–2 MW
Facility reference
≈3–12 MW

An AI pilot for batch inference and fine-tuning. Validate customer demand, data security, fibre, cooling and an agreed interruption schedule before adding capacity.

Load an AI compute example
1 MW · AI/HPC
Electrical input at PUE 1.20
1.2 MW
GPU equipment + cluster IT
€26,000,000
Site infrastructure
€10,000,000
Illustrative CAPEX incl. contingency
€41,400,000
Use AI example · 1 MW
02

Modular

>2–10 MW
Target IT load
>2–10 MW
Facility reference
≈12–60 MW

Repeatable AI clusters for contracted inference and training, with dedicated operations, cluster networking and validated power continuity.

Load an AI compute example
5 MW · AI/HPC
Electrical input at PUE 1.20
6 MW
GPU equipment + cluster IT
€130,000,000
Site infrastructure
€45,000,000
Illustrative CAPEX incl. contingency
€201,250,000
Use AI example · 5 MW
03

Campus

>10–50 MW
Target IT load
>10–50 MW
Facility reference
≈60–300 MW

A staged AI/HPC campus with anchor customers, firm power arrangements, redundant connectivity and engineered cooling. Energize each phase against contracted demand.

Load an AI compute example
20 MW · AI/HPC
Electrical input at PUE 1.20
24 MW
GPU equipment + cluster IT
€520,000,000
Site infrastructure
€160,000,000
Illustrative CAPEX incl. contingency
€782,000,000
Use AI example · 20 MW

Above 50 MW: bespoke campus engineering, grid study and contracting—not a fourth standard offer.

TCP site planner

Transparent site and economics screen

Edit every assumption. The result is a feasibility screen, not a quote, forecast, investment recommendation or promise of grid-service eligibility.

Illustrative H100-class budgets · reviewed 21 September 2026. Loading an example replaces the inputs and clears readiness evidence. These are planning cases, not verified sites or supplier quotes.

Benchmark references and budget basis

80 eight-GPU nodes per IT MW allows 12.5 kW per node including shared IT overhead. €300,000 per node plus €2m/MW for network and storage; site infrastructure is €10m / €9m / €8m per MW for Pilot / Modular / Campus. Add 15% contingency. These allowances require an equipment list, local design and quotes; they do not price B200/B300 systems.

7,000 powered hours, 70% billable utilization and €2.50/GPU-hour are separate commercial assumptions. Electricity is charged for every powered hour at full design load, including unbilled time. The €60/MWh blended energy input must cover the actual power arrangement; low-price surplus alone does not establish 7,000 hours.

1 · Site and load
2 · Economics
3 · Readiness evidence

Boundary conditions

What the screen will not assume

  1. Nameplate capacity is not available compute capacity.
  2. AI/HPC value requires contracted demand, firmness, fiber and cooling.
  3. Compute can curtail load quickly, but ancillary-service participation requires specific qualification and contracts.
  4. BESS stores energy; compute consumes it. The comparison must be use-case and site specific.

Commercial path

From facility data to signed project

1

Share 12–24 months of interval generation, export, price and constraint data plus the single-line diagram.

2

Validate the electrical boundary, supported MW, dispatch envelope, telecom, cooling, noise and permits.

3

Run a metered pilot and confirm compute settlement, stop/start behavior and any BESS role.

4

Negotiate definitive hosting, equipment, revenue-share and operating agreements before scale.

tcp@bittrees.io

Start with a copyable email draft

Create a project enquiry below. Nothing is transmitted by this page: copy the message or open a draft to tcp@bittrees.io in your own email application, then review it before sending.

Browser-only template. TCP receives information only if you explicitly send the email.

Sources and assumption dates

Market facts use official 2025 REN and Red Eléctrica system summaries plus OMIE Portugal and Spain day-ahead files dated 5 July 2026. Site-envelope estimates use published 0.5–1 MW prefabricated systems and a roughly 1 MW liquid-cooled field module as planning references. ASIC scenario inputs (€48/€77/€121 per site MWh, 2% fee, €40k/MW-year O&M and €1.8m/MW base CAPEX) are editable research snapshots dated 19 June 2026—not live quotes.