aFRR, FCR, and the alphabet soup of ancillary markets

A DC energy manager in Germany who asked me last year about ancillary revenue needed three pages of notes before the conversation could get useful. That's not because the markets are especially complicated. It's because the acronyms got specific before the markets got named.

The soup, fastest to slowest.

FCR — Frequency Containment Reserve. The first response to any frequency deviation. Fully automatic, responds within 30 seconds, holds the grid steady while slower reserves mobilise. Minimum bid size in Germany is 1 MW. FCR is probably the easiest product for a site with a battery to bid into because the prequalification is lightest and the activation signal is simple.

aFRR — automatic Frequency Restoration Reserve. Kicks in within 5 to 15 minutes of a frequency deviation, via an automatic signal from the TSO. This is the workhorse of secondary response. German minimum bid size is 5 MW, though smaller providers can participate if they submit only one bid per product per LFC area. aFRR is where most of the ancillary revenue volume lives in Europe.

mFRR — manual Frequency Restoration Reserve. Activated manually by the TSO within 15 minutes. Also 5 MW minimum bid in Germany. Less frequent activations than aFRR, but each activation tends to be longer and is priced pay-as-cleared on the MARI platform, which standardised at a 15-minute resolution when it went live in 2022.

RR — Replacement Reserve. Tertiary response, activated after 15 minutes if imbalance persists. Handled by the TERRE platform in the markets that use it. Not all European markets do.

Above those four products sit the cross-border platforms that now coordinate them at an EU level.

PICASSO handles aFRR exchange across participating TSOs. Operational since June 2022. Uses a 4-second market time unit and clears more than 7.8 million markets per month. Currently includes Germany, Austria, Netherlands, Belgium, Denmark, Czechia, Spain, Portugal, and a growing list as further TSOs accede through 2026.

MARI handles mFRR exchange. Same operational window, different product. 13 operational TSOs as of May 2025 with 16 more in various states of integration. Belgium's Elia joined in May 2025. TenneT NL connected in December. MARI prices repeatedly hit ±€10,000/MWh in the Baltics during late 2025, which gives you a sense of how much price signal the markets can carry when they go short.

IGCC is the imbalance netting layer that sits on top of PICASSO. It uses the same IT infrastructure and works by cancelling out simultaneous up and down activations between TSOs before either side actually fires. Quiet but important.

Four products, four platforms. Not conceptually complicated. Annoying to look up.

What the soup means for a data centre is that behind-the-meter flexibility, when it exists, has multiple places to earn money. A 10 MW battery isn't just a peak-shaving asset. It's a qualifying asset for FCR, aFRR, mFRR, and imbalance settlement, in most Western European markets simultaneously. Optimisation is the game, and the optimisation is non-trivial because the products interact. Reserve the battery for FCR and you can't use the same capacity for aFRR in the same hour. State-of-charge decisions bind across products. The revenue stack is a constrained optimisation problem, not an accounting sum.

This is the part of the story that's shifting fastest. Through 2024 and into 2025, European ancillary service revenue for a 2-hour BESS touched €1M per MW per year in the best markets, with peak projections to €3.6M in particularly volatile conditions. Those numbers are symptoms of early-stage scarcity, not a sustainable run rate. New storage is being built aggressively. Margins will normalise. Somewhere between today and 2028 is the window where participation earns above-market returns for operators who can run the optimisation. After that the returns normalise toward the cost of capital like every other optimised commodity market.

For a DC operator specifically, the frictions worth understanding are local ones.

Prequalification is per-asset, per-TSO. A battery qualified for aFRR in Germany doesn't automatically qualify in the Netherlands. This is a local engineering and paperwork exercise, handled via the regelleistung.net platform in Germany or its national equivalent elsewhere. Aggregators can do the qualification on behalf of smaller sites. Larger sites sometimes choose direct participation.

The products also aren't fungible with DC load in the way people sometimes assume. You can't bid your production tenants' servers into aFRR. You can bid behind-the-meter storage, non-critical compute clusters, or thermal storage with enough inertia to shift cooling draw by several minutes. The addressable asset class is narrower than "the site's total MW", usually by a wide margin.

And the product definitions are changing faster than most operators have internalised. ACER is actively harmonising prequalification requirements across PICASSO and MARI participants. The French balancing mechanism extends mandatory participation to generation assets above 10 MW from 2026. ALPACA brings joint aFRR procurement between Austria, Czechia, and Germany from September 2025. Anyone modelling ancillary revenue on 2023 product definitions is modelling a market that doesn't exist anymore.

So what does participation look like for a DC with a 10 MW behind-the-meter battery in Frankfurt, realistically, today? Qualify on FCR as the entry product. Add aFRR in year two once the dispatch logic and the prequalification are proven. Run passive imbalance alongside — this doesn't need prequalification, just fast price access and a willingness to react. Total revenue contribution to the site's energy P&L somewhere between 5% and 15% depending on volatility, optimiser quality, and how much capacity is reserved for peak shaving.

That range is wide because the market is still finding its shape. Which is the honest answer to the underlying question: is it worth it?