A battery in a shipping container looks much the same whether it is sitting next to a wind farm, bolted onto a suburban substation, or behind a hyperscaler data centre. The economics behind each one are not the same. Three distinct BESS business models now compete for capital in 2026 - standalone, co-located with renewables, and data-centre - and each has a different revenue stack, a different grid interface and a different design envelope. The winner in a given project is almost never about the hardware. It is about which of the three is best-suited to the site.
Three BESS configurations, one underlying technology
2025 - the year European BESS contracts broke through
The number that stunned the European energy market in 2025 was not how many GWh of batteries were built; it was how many were contracted. According to Pexapark’s 2026 Renewables Market Outlook, almost 12 GW / 24 GWh of battery storage capacity was signed under Flexibility Purchase Agreements (FPAs) and optimisation contracts - roughly three times 2024’s volume. Long-term renewable PPA volumes fell to 13.1 GW in the same period; for the first time, BESS offtake came within shouting distance of PPA offtake. Capital did not leave clean energy in 2025; it pivoted from generation to flexibility.
Inside the BESS contract bucket, fixed-revenue Flexibility Purchase Agreements (FPAs) - tolls, floors, revenue swaps - did the heavy lifting. Pexapark recorded 6.5 GW of FPA volume in 2025 versus 3.3 GW in 2024 (+97%) and 38 disclosed FPA deals versus 12 the year before. Great Britain still dominated with 4.5 GW (~75% of European FPA capacity), but the activity broadened: Germany, Italy, the Netherlands, and even first deals in Bulgaria and Poland. Three large utilities (EDF, Statkraft, SSE) account for 77% of the GB market; the EU is more competitive but still concentrated, with the top three at 66%.
The shift from speculative merchant BESS pipelines to bankable contracted volumes is the defining commercial story of 2025. The next three sections walk through each of the three configurations (standalone, co-located, data-centre) inside this new contractual landscape - because the same hardware now competes for capital under three very different revenue logics.
Standalone BESS - the default grid-connected asset
A standalone BESS is a grid-connected battery sitting on a dedicated substation with no generator attached. It buys energy from the grid when prices are low and sells back when prices are high, bidding into day-ahead, intraday and ancillary-services markets exactly like any other dispatchable asset. For most of the 2021–2024 boom, this was the only model with a serious European pipeline - the UK built its entire 12.9 GWh operational fleet out of standalone projects; Germany crossed 2 GW of utility-scale standalone capacity in mid-2025, heading for 3 GW by year-end.
The economics are straightforward in principle and brutal in practice. A standalone project stacks revenue from three sources: wholesale arbitrage (the day-ahead spread), ancillary services (FCR, aFRR, dynamic containment), and capacity or tolling contracts. When spreads are wide, like 2022–23 when UK T-1 capacity prices cleared above £60/kW/yr, standalone projects print cash. When spreads compress - German arbitrage fell from average €120/MWh in 2022 to €80–90/MWh in 2025 - unlevered IRRs on standalone fall from double digits to 5–7%, which is why the unsubsidised standalone market has been described as “unbankable in 2025” by more than one analyst.
Standalone is still the structurally simplest BESS. One interconnection agreement. One grid code. One revenue model. One point of failure. It remains the default where grid connections are abundant, where wholesale spreads are wide, or where a capacity market provides a floor - the UK, Ireland, the Nordics, and increasingly Italy.
What changed in 2025 is the shape of the spread the standalone BESS is bidding into. Pexapark’s 2-hour top-bottom (TB2) tracking shows daily arbitrage potential rose materially in every major market: Germany +6% (already-elevated 2024 base), Italy +3%, Great Britain +12%, and Spain +38% - the largest single-market jump in Europe. Spain’s 2024 negative-price hours (247) more than doubled in 2025, widening the bottom of the spread without compressing the top. Germany’s 2-hour spread now averages €232/MW/day, up 25% from 2023’s €186.
Spain’s widening spread is the defining 2025 standalone signal. It is also the reason why, even before the FEDER auctions clear, operators rather than developers are now the constraint - a cell does not capture a €200/MW/day spread by itself, it captures it via the dispatch policy and the forecast quality of whoever runs it.
Co-located BESS - solar plus storage, or wind plus storage, on shared steel
Co-location puts the battery on the same site as a generator, usually solar PV. The two assets share the grid connection, the transformer, the switchgear, the access road and often the SCADA. Depending on the grid code, they may be AC-coupled (two separate inverters, both feeding the same grid connection) or DC-coupled (battery and PV share the same inverter with power combined on the DC bus).
| Aspect | AC-coupled | DC-coupled |
|---|---|---|
| Inverter | Two separate inverters; PV and BESS each have their own. Pool on the LV / MV AC bus. | One bidirectional inverter; PV connects to its DC bus through a DC-DC converter. |
| Round-trip eff. | ~85–88% AC; PV power inverts twice when stored. | ~88–92% AC; saves one DC↔AC step on charge. |
| Capex | Higher hardware count; simpler controls. | Single inverter, fewer transformers; ~5–10% lower capex on small/medium projects. |
| Retrofit | Easy — add BESS to an existing PV plant without touching the inverter fleet. | Hard — needs new shared inverter rated for both. |
| Clipping recovery | Cannot capture pre-inverter clipping. | Captures DC-side clipping (5–15% of solar yield in oversized arrays). |
| Where it wins | Greenfield + retrofit, modular sizing, pure ITC stacking. | High-DC ratio PV plants; new builds with shared MV BoP; data-centre microgrids. |
The economic logic is powerful. Germany’s cable-pooling rules allow the combined project to share a single grid connection point, cutting infrastructure cost by 20–40% versus building the two assets separately. In Spain, where roughly one hour in eight in 2025 saw zero or negative prices, co-location turns a solar asset from a price-taker into a time-shifted merchant: generate at midday, store the surplus, sell into the evening ramp at 4–6x the price. Co-located BESS deal volumes in Europe grew 676% year-on-year into the first three quarters of 2025, roughly twice the growth rate of standalone.
- Grid connection cost - one substation, not two. 20–40% capex saving where cable pooling is allowed.
- Capacity factor of the connection - a solar site typically uses only 20–25% of its MVA rating annually. Sharing it with a battery lifts utilisation to 60–70% with zero extra grid cost.
- Cannibalisation hedge - the solar gets paid for stored energy at evening prices, not zero. The battery gets charge energy at cost of generation, not market.
Co-location has design costs too. The combined project has to share the grid connection capacity, which means the solar occasionally curtails to let the battery export. The BESS inverters typically need to be grid-forming so the combined asset still behaves well at the interconnection point. And the two pieces of hardware have different depreciation lives - batteries 15–20 years, PV panels 25–30 - which complicates refinancing and augmentation.
Data-centre BESS - the category that barely existed in 2022
The largest single change in the BESS market in the last 18 months has been the emergence of the hyperscaler as a buyer. The Volta Foundation’s 2025 Annual Battery Report documents that global BESS deployments crossed 100 GW for the first time in 2025, with 104 GW and 257 GWh added in that single year - and that data-centre operators have shifted from treating batteries as a passive uninterruptible-power accessory to treating them as mission-critical infrastructure integrated with both onsite generation and the grid.
The numbers from the hyperscalers themselves give the shape. Microsoft has announced 21 BESS-adjacent data-centre projects; Google 19; Amazon 18; Meta 12. US data-centre power demand alone is on track from 62 GW in 2025 to 76 GW in 2026, and a quadrupling by 2030. Volta Foundation’s report puts the total hyperscaler BESS opportunity at roughly 20 GW through 2035, with about 9 GW of it by 2030.
What is actually different about a data-centre BESS
A data-centre battery is not a UPS. A traditional UPS provides seconds-to-minutes of backup while a diesel generator starts. A data-centre BESS is engineered around three overlapping jobs, and the Volta Foundation’s 2025 analysis is the first public report to document this shift at scale.
This is what the industry means by the shift from UPS to BESS inside the data centre: the same chemistry as a utility-scale project, the same grid-forming PCS, the same SCADA - but dimensioned for hours rather than seconds, and financially justified by grid-service revenue as well as reliability. In practice, 2025 data-centre BESS installations range from 30–50 MW / 1–2 hours to flagship designs in the 150–500 MW range.
From 415 V AC to 800 V DC - the architecture shift
The Volta Foundation 2025 Annual Battery Report frames the deeper shift: batteries are now embedded across every layer of the modern data-centre power architecture, from facility-level BESS connected to the grid down to rack-level battery backup units (Rack BBU) sitting inches from AI servers. As power densities rise and uptime tightens, hyperscalers are migrating from centralised 415 V AC backup to distributed 800 V DC, with NVIDIA leading the ecosystem push for AI factories.
The architecture shift matters because batteries become not one box but two layers, with new performance requirements at each. Facility-level BESS still connect to the grid and earn from grid services; the new layer - rack-mounted battery backup units - sits inches from the GPUs, absorbs millisecond load spikes that even the fastest grid-tied BESS cannot smooth, and is increasingly specified by hyperscalers in the same procurement cycle as their compute. NVIDIA’s 800 VDC ecosystem push, Microsoft’s 800 V Open Compute reference design and Google’s ±400 V architecture are converging on the same broad direction: more cells, distributed across the campus, and tied to a single high-voltage DC bus that minimises conversion losses on the way from the substation to the silicon.
There are two structural consequences. First, hyperscalers are becoming some of the largest single offtakers of LFP cells in the world - placing multi-GWh supply contracts directly with manufacturers, a buying pattern that was almost entirely utility-led three years ago. Second, hyperscalers are driving novel commercial structures: virtual power plants aggregating multiple data-centre BESS, tolling deals with grid-service providers, and co-location inside data-centre campuses of both BESS and onsite solar or gas generation.
Revenue stack · how the same hardware earns differently
The cells, BMS, PCS and thermal system are the same in all three configurations. Where the money comes from is not. The chart below stacks indicative revenue contributions by source for an unsubsidised European 100 MW / 2 h project in 2025–2026 - directional, drawn from McKinsey, Wood Mackenzie and Volta Foundation.
Architectural differences at a glance
| Feature | Standalone | Co-located | Data-centre |
|---|---|---|---|
| Primary revenue | Wholesale + ancillary + capacity | Arbitrage on host generator + grid services | Reliability + grid services + deferral |
| Grid connection | Dedicated substation | Shared with solar/wind | Behind data-centre meter |
| Typical size range | 20–400 MW / 1–4 h | 20–300 MW / 2–4 h | 30–500 MW / 1–4 h |
| Typical PCS | Grid-forming increasingly required | Grid-forming; AC- or DC-coupled | Grid-forming + islanding capability |
| Counterparty | TSO, market, utility offtaker | PPA offtaker + TSO + trading desk | Hyperscaler + TSO |
| Key risk | Revenue stack compression | Shared-infrastructure availability | Load profile volatility, tech obsolescence |
| Where it wins | Mature markets, deep spreads, capacity auction | Solar-heavy systems with cannibalisation | Hyperscaler markets, constrained grids |
The economics in three numbers
For a 100 MW / 2-hour project, the Volta Foundation 2025 report puts turnkey BESS cost at roughly $117/kWh, down 31% year-on-year from 2024 and 70% since 2022. Applied to each configuration, that implies similar hardware bills of order €25–30 million for 200 MWh of capacity, but dramatically different revenues:
- Standalone (Germany 2025): arbitrage-led revenues of €80–90/MWh spread, unlevered IRR 5–7% at today’s spreads; resilient 9–11% in markets with a capacity floor.
- Co-located (Iberian): shared-infrastructure capex savings of 20–40%, plus a solar PPA hedge that lifts unlevered IRR by 1–2 percentage points. 2025 deal volume grew 6.8x year-on-year.
- Data-centre: commercial structure dominated by bilateral tolling; revenue not primarily set by wholesale spreads but by the value of reliability and deferred grid investment to the hyperscaler; double-digit unlevered IRRs on the best sites.
Which model wins where
Standalone wins where the grid is deep, the market is liquid, and the auction calendar is predictable - which is why 2025 capacity awards concentrated in the UK (18 GWh), Poland (20 GWh), Italy (10 GWh), Spain (9.4 GWh) and Bulgaria (13.7 GWh). Co-location wins where the grid is congested, the solar or wind generator would otherwise be cannibalised by its own abundance, and where cable pooling or hybrid permitting is allowed - Germany, Spain, Italy, the Netherlands. Data-centre BESS wins where the host load is massive, volatile, and unable to wait for a firm grid connection - which increasingly describes the hyperscaler pipeline in Ireland, the Nordics, Iberia and northern Germany.
The three models are not substitutes. They are three different applications of the same technology, chasing three different market signals. The European market in 2026 is the first year where all three are scaling at the same time, and the winners in each bucket will not be the cheapest hardware vendors - they will be the developers who matched the configuration to the grid signal.
The bottom line
A BESS is a chemistry. The business is a configuration choice laid on top. Standalone optimises for market participation and depth; co-location optimises for grid-connection scarcity and cannibalisation hedging; data-centre optimises for reliability-plus-flexibility under hyperscaler load. The cells inside all three are almost the same. The capital structure, the offtaker, the interconnection and the revenue are not.
Over the next three years, the fastest-growing of the three is likely to be data-centre: Volta Foundation’s 2025 numbers put the hyperscaler opportunity at 20 GW and the growth rate above anything else in the BESS market today. The most capital-efficient of the three will be co-location: the 676% annual deal-volume growth in 2025 Europe is almost certainly structural, not a one-off. Standalone is the steady core - but in 2026 it is no longer the whole market.