24/7 Carbon-Free Energy: the new gold standard for hyperscalers
Beyond annualised 100% renewable, 24/7 CFE demands strict hourly coverage. Mechanics, cost and feasibility explained.
The end of annualised greenwashing
Buying 1 TWh of solar RECs in July to cover 1 TWh of December consumption is carbon time-shifting: the grid mix actually consumed remains fossil. Google and Microsoft buried that practice in 2020 by publishing the 24/7 CFE standard.
The exact metric
CFE Score = Σ (carbon-free energy consumed at hour h) / Σ (total energy consumed at hour h), across 8,760 hours.
A data center can be 100% renewable annualised while sitting at 62% hourly CFE. Targeting 95% hourly CFE radically changes PPA strategy.
The winning mix
- 40% solar (day cover)
- 30% wind (night + winter cover)
- 20% nuclear or hydro baseload
- 10% battery storage (smoothing)
Without nuclear or hydro, hitting 95% CFE costs 2.3× more (PPA multiplication + producer curtailment).
The real cost
| CFE level | Cost vs spot | Feasibility |
|---|---|---|
| 100% annualised | +0–5% | easy |
| 80% hourly | +8–15% | reachable |
| 95% hourly | +25–40% | demanding |
| 100% hourly | +60–110% | hyperscalers only |
The Voltarione engine
Our matching engine computes in real time the achievable CFE score given listed producers + optional EDF/Vattenfall nuclear mix. Output: recommended PPA portfolio optimised for a target CFE threshold (e.g. 92%).
Continue reading
Corporate PPAs for data centers in Europe: the 2026 practical guide (types, pricing, clauses, countries)
PPA structures (on-site, off-site sleeved, virtual), country-by-country price ranges, clauses to negotiate and key markets: the practical 2026 guide to securing power for a European data center.
Nuclear PPAs for data centres: the 2026 European guide (SMR, baseload, pricing)
SMRs, baseload contracts, country-by-country pricing: the complete 2026 guide to nuclear PPAs for AI data centres in Europe. Nuclear vs renewables, the post-ARENH framework (CAPN/VNU), and the Microsoft, Amazon, Rolls-Royce and EDF cases.
How much electricity does an AI data center really consume? A 2026 European benchmark (MW, €/MWh, PPA)
A frontier AI training cluster now draws 100–300 MW continuously, inference already accounts for 80–90% of AI compute, and the all-in energy cost of a 100 MW campus ranges from €38–55M/year in the Nordics to €118–149M/year in Dublin. A data-heavy 2026 European benchmark.