08
^{FLEX MARKETS}
Unlocking energy market flexibility with AI

^{MARKET FLEXIBILITY}

Unlocking energy market flexibility with AI

Power markets globally are facing the same challenge: governments and energy companies are adding renewables faster than grids can absorb them.

The resulting stress is visible in prices, wasted energy and intervention costs in many energy markets. On May 1, 2026, German day-ahead prices hit the exchange floor of -€499.99/MWh. In 2025, Spain, the Netherlands and Sweden’s SE2 zone recorded high levels of negative pricing, while billions of watts of wind and solar were wasted in Australia. In the same year, Germany spent €3.07 billion on managing grid congestion, and Spain spent between €3.3 and €3.8 billion.

For companies operating in these markets, this is a margin story more than an environmental one. Volatility on this scale destroys value for participants who can’t respond and creates value for those who can. The same hour that wipes out revenue for a generator selling at market rates pays a battery operator to absorb energy. The gap between those two outcomes is widening every year, and it’s decided by operational capability, not asset ownership.

Germany’s power market is a case in point. It holds one of the largest flexibility fleets in the world, with roughly 16 GW of batteries, two million electric vehicles and two million heat pumps. Yet it still can’t exploit this potential. The barrier isn’t the hardware, but everything between it and the market: metering, market access and the ability to act on a price signal within fixed time slots.

Flexibility readiness depends on six dimensions

To better understand what’s holding participants back, we assessed the flexibility readiness of 11 energy markets across six dimensions:

D1:

How much stress the grid is under.

D2:

Whether the market creates meaningful price signals.

D3:

How extensively smart meters are used.

D4:

How much flexible
capacity exists.

D5:

Whether assets are approved to take part in energy markets.

D6:

Whether models exist for monetizing that participation.

We then divided these into three clusters:

1

Market design readiness:

grid and system stress (D1) and price signal formation (D2).

2

Technology readiness:

metering and data infrastructure (D3) and installed flexible capacity (D4).

3

Market participant readiness:

market access and prequalification (D5) and commercial models (D6).

The result: participants influence or control five of those dimensions

It’s common for market participants to assume that readiness is a regulatory problem that only governments can fix. But our analysis finds that only one of the six dimensions is genuinely outside of their influence.

Figure 1: Where energy market participants can act on their own initiative

Degree of control of market participants	Dimensions	What this means in practice
Direct control	D4 Flexible capacity D5 Market access and prequalification D6 Commercial models	Investment, process and business-model decisions. No regulator has to move first.
Strong influence	D2 Price signals D3 Metering and data	Retailers decide whether price signals reach customers. Any asset owner can install its own sub-metering ahead of the national rollout.
Response only	D1 Grid and system stress	Set by network operators and regulation. Participants shape their exposure by choosing where they site their assets and how they bid when congestion strikes.

This finding matters because the three dimensions participants control directly are those where value can be created. In Germany, only 2.55 GW of a 16 GW battery fleet is prequalified for trade. And across Europe, under 20% of an estimated 100 GW of demand-side flexibility reaches the market.

Neither of these gaps require a regulator to close them. They require someone to do prequalification work, build the trading capability, and stand up the commercial model.

Where AI creates value

The main case for AI here is feasibility, not efficiency. A team of traders can dispatch a conventional portfolio of 10 thermal units. They can’t dispatch 200,000 heat pumps, home batteries and vehicle chargers. Each asset has its own availability, state of charge, customer preferences and network constraints. And the best course of action changes every 15 minutes, across several markets at once.

Only AI can coordinate assets on that scale. That’s why algorithms dispatch every large distributed portfolio in operation today – from Kraken's 2 GW residential fleet to Tesla's Autobidder across 33 sites.

But even for participants working with smaller portfolios, AI can create value across all five dimensions where they have influence or control. This chart maps practical use cases against the effort needed to implement them and the business value they create.

Figure 2: Where AI creates value for energy market participants – effort against benefit

Four ways to unlock that value

1. Go after the quick wins. Prequalification and compliance automation is the most underrated item on the chart. Automating this document- and process-heavy bottleneck would unlock otherwise inaccessible revenue, yet none of the markets we examined has done it.

Price and imbalance forecasting is the highest-value low-effort case. That’s because better forecasts reduce the reserve a portfolio must hold and improve every subsequent bidding decision. Google DeepMind raised the value of a 700 MW wind fleet by around 20% by improving its day-ahead bids. And the UK’s National Energy System Operator (NESO) reports a 33% more accurate solar forecast from satellite nowcasting. Both build on data companies already hold.

2. Commit to orchestration – it requires more effort but brings the biggest gains. Orchestrating a distributed fleet is one of the hardest activities in our assessment, requiring device connectivity, real-time data pipelines and a change in how the trading floor works. But it also separates the aggregators gaining market share from those losing it.

The same applies to automated bidding across multiple markets. Peer-reviewed work on intraday battery trading finds that optimizing continuously earns 58% more than optimizing hourly. Modo Energy has recorded individual batteries earning up to 70% more than the fleet average in a single month, in the same market. The difference is better optimizing software, not better hardware.

3. For network operators, the prize is avoiding capital expenditure. The UK’s National Grid uses dynamic line rating across 275km of transmission lines; Terna in Italy uses it across roughly 20 connections. Both TSOs claim it allows existing lines to carry 40% more current and, in one case, avoided a €24 million line replacement.

Meanwhile, in France, RTE’s “Learning to Run a Power Network” challenge showed that optimizing grid topology relieves congestion by reconfiguring the network, rather than paying generators to redispatch. With markets including Germany and Spain spending upwards of €3 billion on congestion management, even small percentage gains make a material difference.

4. Don’t let poor metering stand in the way. AI acts on data, and data means meters – presenting a challenge for markets where coverage is low. Load disaggregation (estimating what individual devices consume based on readings from a single meter) can partially fill the data gap. That’s why it appears on the chart, despite modest returns. But it’s a workaround, not a substitute. In a market without widespread metering, the only way to build a successful business case for AI is to install sub-meters ahead of a national rollout.

Where you start depends on what you own

For utilities, retailers and aggregators: the first move is forecasting, because it improves every downstream decision and needs no new assets. The second is deciding whether to build a data & AI orchestration platform in-house or partner with a specialist.

For generators and storage owners: the fastest return is to automate prequalification and bidding across every product the asset is technically capable of serving.

For network operators: dynamic line rating and congestion forecasting release capacity from existing infrastructure years before new lines can be built. And both are defensible to a regulator on avoided-cost grounds.

Final word

Flexibility is the binding constraint on the energy transition, and market design is the global bottleneck. But waiting for market design to improve isn’t a strategy. Participants can influence or control five of the six dimensions that determine flexibility readiness – and AI can help turn that readiness into commercial advantage.

The uncomfortable truth, though, is that it can’t create readiness where it doesn’t already exist. Applied to a portfolio that’s metered, market-accessible and price-exposed, AI compounds returns. Applied to one that isn’t, it produces impressive models and no margin. Companies that close the readiness gap first will find that the same volatility punishing their competitors becomes their most reliable source of value.

^Sources:

^{Capgemini World Energy Markets Observatory flexibility readiness analysis across 11 markets, drawing on Bundesnetzagentur, EPEX SPOT, REE, RTE, TenneT, Nord Pool, NESO, AEMO, CAISO, FERC, regelleistung.net, IEA, Modo Energy and BCG Platinion.}
^{Company-published figures from Octopus Energy, Tesla, National Grid, Terna and Google DeepMind. Company capacities and optimization uplift claims are self-reported.}

Authors

Torben Schuster

Vice president, Head of Energy Transition and Utilities, Capgemini

Christian Bussar

Manager, Energy Transition and Utilities, Capgemini