Unlocking the Hidden Value of our Power Grid: AI-Based Products to Boost Utilisation

Electrification and data centre build out are being delayed by grid bottlenecks. Grid expansion and grid digitisation are critical to solve this, but can be slow, regulated and expensive. In the meantime, new AI-based products to increase grid utilisation arise as a another, fast way to help grid operators and users unlock value from existing power grids.

We've all heard that power grids are increasingly congested. In many European and North American regions, there is insufficient transfer capacity to transmit all available and requested power from A to B. The biggest pain point is in high & medium voltage grids, leading large power consumers and producers to wait ~7–10 years for new grid connections or upgrades. But congestion is also rising at the low-voltage distribution level, causing load spikes and voltage violations which, especially in poorly digitised grids, risk outages and blackouts.

We are stuck in a traffic jam, in a gridlock. This slows down electrification, delays data centre build-outs, and hampers economic growth and competitiveness.

Less known is that power grids are heavily underutilised. Depending on the asset type and location, utilisation typically ranges from just 30% to 50%. The root cause is that grids are designed for peak demand, which typically only occurs a few hours per year. It's similar to motorways, which have lots of traffic during morning and evening commutes, but are barely used overnight. And utilities and grid operators often lack the real-time visibility and analytical, digital tools to change this: many rely on on-premise, static planning tools rather than real-time, cloud-based software.

The problem is huge. The gridlock and underutilisation result in economic losses in three ways:

1.     Long connection queues slow down clean electrification and economic growth. In the US, ~1,480 GW of renewable energy projects are in the queue today, while demand for new data centre capacity is expected to grow with 65-90 GW by 2030, or ~$800 bn of investments.  

2.     Congestion costs consumers billions. Grid operators need to manage the dispatch of power over time and space so it doesn’t overload the grid. This often results in higher generation costs and/or having to pay consumers not to use power – costs passed onto customers. In the US, these congestion costs are estimated at $20 bn per year, and in the UK were about ~$1.3 bn in 2024 – set to rise to $4 bn in 2030.

3.     Capital investments for new grid infrastructure translate into unnecessary high grid costs. Annual grid capex are already high today and expected to increase from ~$350bn today to $500 - $800bn by 2030. These costs are passed on to customers through grid charges. Through better utilisation, grid charges can be contained as i) part of the capex can be deferred and ii) it can be spread over higher volumes of power.

The good news is that new AI-based software and data products are emerging to increase the utilisation of our existing grid. These can unlock grid capacity faster, flexibly, and at lower cost than traditional grid expansion – and don’t rely on the full digitisation of grid operators. These solutions are no silver bullets, and won't replace the need for long-term infrastructure buildout, but they can ease bottlenecks and lower system cost in the short term. While they can’t fully unlock the value of grid utilisation, they are well-positioned to capture a meaningful share. We see two main product types:

1.     Load flexibility mechanisms – solutions that change the behaviour of grid users, such as flexibility grid contracts, competitive auctions or flexibility marketplaces: “adjusting the traffic”.

2.     Grid enhancement technologies (GETs) – solutions that increase physical capacity or optimises power flows of grid assets: “adjusting the roadmap”.

Alongside these, short-interval (e.g. 15-min), locational power prices are key to achieve system efficiency: the “highway tolls”. I won’t dive into this in detail, as it’s slower-moving and regulatory by nature, but it’s an essential piece of the long-term puzzle.

Neither load flexibility mechanisms nor GETs are new, but adoption has been limited to date. Grid operators are understandably risk-averse towards new technology, with security of supply as their top priority. They have also lacked strong incentives to optimise and invest in utilisation: their revenue is largely capex-based, and the economic gains of higher utilisation often mainly accrue to grid users.

So, what makes us excited about these solutions today? Advancements in AI, increasing power demand and willingness to pay for grid access, and evolving regulatory frameworks are shifting fundamentals. And while the lack of grid digitisation has long been an entry-barrier for start-ups in this space, we increasingly see solutions emerge that do not require full software integration.

Let me share two innovations that operate outside of grid operator control systems, allowing them to move faster and help unlock capacity where traditional grid systems aren’t yet able to deliver at scale.

Case 1: software to help data centres unlock early grid access. There is a lot of available capacity in our power grids which can be unlocked by "non-firm" or "flexibility" grid contracts. These contracts involve consumers or generators agreeing on scheduled curtailments (i.e., percentage reduction of power for a particular period) with the grid operator, leveraging available capacity outside peak hours. As these contracts are 'capacity constrained' they typically offer earlier grid access and/or reduced grid tariffs. The potential is huge: in the US, ‘curtailment enabled headroom’ is estimated to be ~100 GW (12.5% of peak winter load). But unlocking this capacity fast and at scale is complex, and requires advanced analytics of grid data (hourly utilisation, network planning, queues, etc.) combined with a good understanding of grid physics and customer needs. This is where new tech companies can come in as a matchmaker. In the US, start-up GridCARE is helping data centre developers to obtain grid connections faster using GenAI, grid data and flex grid contracts – working for developers, and with grid operators – a smart way to extract value in this space.  Emerald AI and PIQ energy are doing similar things. It’s still early days, but we’re excited to see what traction and impact they’ll achieve.

Case 2: AI-powered sensorless dynamic line rating (DLR). DLR is a grid enhancement technology (GET) that allows grid operators to adjust the capacity of power lines based on weather conditions: on colder/windier days, cables can carry up to ~30% more power. DLR has been around for decades, but adoption is limited. Most existing DLR offerings rely on i) hardware sensors to measure cable temperatures in real time (which limits implementation and prevents day-ahead grid planning) and/or ii) simple weather forecasts (which may lack the precision required by grid operators for reliability). But major steps in AI are creating new digital offerings that solve both implementation and precision challenges. Start-up Gridraven, for example, applies machine learning to hyper-local wind forecasts, actual wind measurements, and landscape information to predict temperatures with unparalleled precision. This requires a level of compute power and intelligence that was not possible until recently.

There is enormous value in our power grids, and new software-first solutions are here to unlock it. Whilst grid operators used to focus on putting cables in the ground, software and data offerings to increase utilisation are gaining traction. Tailwinds of public pressure, high power demand, regulatory shifts, and technological advances are reshaping the landscape. Several companies are emerging to help unlock the gridlock by using existing data in a new way and redefining routes to market. The power grid is not just an infrastructure story anymore – it's a data and innovation story. Those who can help unlock its existing capacity can unleash economic growth, electrification and renewable power, and with that immense value.

Are you a founder building something interesting in this space, or otherwise have an opinion? We'd love to hear from you.