The AI for energy track brought nine jury-selected startups building tools for energy systems optimization, intelligence and automation. The format was unforgiving: three minutes to pitch, followed by two minutes of jury questions, with the host interrupting anyone who ran over. Here is what each company said, in pitching order.
Central EMS
Turning building data into asset value
Sonya Hofer, Co-Founder, pitched software that helps real estate companies grow asset value by turning fragmented building energy data into AI-optimized management.
Everyone knows real estate consumes considerable energy. Less discussed is the cost of inaction: energy performance already affects achievable rent, time on market and ultimately asset value. Yet owners still rely on overpriced, static systems.
Central EMS connects to a building’s entire energy ecosystem and continuously optimizes when energy is consumed, stored or sold. Hofer put the results at 20 to 40% lower energy costs and an EPC improvement of up to two bands. At a 5% cap rate, every euro saved on energy adds twenty to asset value.
The gap nobody else is filling
Existing building energy management systems focus almost exclusively on commercial offices, leaving residential multifamily buildings virtually untouched. Two things explain that: technical complexity, and split incentives that leave stakeholders deadlocked over who pays.
Entering through residential therefore means almost no competition, fast ROI and low customer acquisition cost, with expansion into other asset types following.
Traction and pricing
Central EMS is contracted across 500 assets by year end, having reached 150 in Germany in eight months — which Hofer called a data moat rather than a vanity metric, noting it took competitors eight years. The company had raised 500k and was raising 3.8 million.
Pricing is two-tier: €25 per building for the analytics and monitoring needed for ESG compliance, then optimization tiered by consumption, structured as roughly 10% of savings but charged flat so costs stay predictable.
Sonya Hofer, Co-Founder of Central EMS pitching at Energy Tech Summit 2026
Decentral AI
Private models, chosen automatically
Marouen Helali, CEO, started from an enterprise problem rather than an energy one. Companies want AI, yet they fear regulation, carry heavy governance requirements, and need their models fully private.
The company’s algorithm selects the right model for each task. A maths question routes to a maths model, a chemistry question to a chemistry model — which suits energy particularly well, given how much chemistry, physics and maths the sector involves.
Because specialized models are smaller, Helali claimed the system benchmarks better than large general-purpose alternatives. Those alternatives are also centralized, so using them means sending your data out. Decentral AI instead runs behind the firewall.
Why enterprises won’t do this themselves
The commercial insight is that enterprises will not experiment with different models on their own, so the company packages selection as software that works out of the box.
Helali said a recently closed deal took the company to 1 million in annual recurring revenue. It is cash flow positive, having raised 350k in its first two months and nothing since.
On avoiding token costs
Asked how on-premise deployment escapes token costs, Helali explained that open source models can run on your own hardware. Mainstream models need very expensive GPUs because of their size, which is what concentrates the market among a few providers. Smaller specialized models deliver equivalent results locally — and once the hardware runs, there are no token costs at all.
Dyneo Technologies
The networks already in operation
Max Carrel, Co-Founder and CTO, opened with a photograph from his balcony in Geneva, then pointed past the mountains to thousands of buildings pumping CO2 into the atmosphere to meet heating demand.
Geneva, like many European cities, plans to expand district heating to replace fossil boilers. New construction can deliver low carbon heat. Networks already in operation, however, remain vastly powered by fossil fuel — leaving operators to reduce that dependence while keeping prices competitive.
Prioritizing bottlenecks, not just finding them
Dyneo’s software helps operators reduce heat loss while increasing the renewable share, and streamlines the work of running increasingly complex assets.
The platform consolidates data on how the network operates, then detects and — more importantly — prioritizes performance bottlenecks. Recommendations follow.
Does it act, or only advise?
A juror asked exactly that. Dyneo operates in two modes: recommendations where physical intervention is needed, such as a broken pump, and automatic set point optimization where it can act directly. Commercially it is a subscription with an onboarding cost.
Heimdall Power
Operating a dynamic system on static assumptions
Jørgen Festervoll, CEO, opened with the company’s founding conviction: better grids mean better lives.
The grid is under unprecedented pressure. Demand is exploding from electrification, industry and AI data centers, while thousands of gigawatts of renewables queue to connect to a grid built for a different era.
Wires and poles cannot go into the ground fast enough to build a way out. So the industry has to extract more from the grid it already has — starting by knowing how much capacity is there.
The core problem is that utilities operate a dynamic system using static assumptions. They do not know what a line can carry right now, at this location, under these conditions. Consequently they apply conservative margins, and significant capacity sits untapped.
“The real question isn’t whether capacity exists,” Festervoll said. “It’s whether we can see it, trust it, and unlock it.”
The Apple Watch for the power grid
Sensors mount directly onto high voltage lines using autonomous drones, in under a minute, without outages. Combined with AI software, they typically unlock 20 to 40% additional capacity — equivalent to a new transmission line delivered in hours rather than a decade.
The company has deployed with more than 50 utilities across 21 countries. Even so, 99% of high voltage overhead lines worldwide carry no sensor.
Why hardware still matters
Asked whether software-only competitors threaten the business, Festervoll was direct. Heimdall offers a software-only version, yet congested lines need a sensor to verify the data.
He described the company as hardware-enabled software that is really a data play: more sensors mean more data, better products, better adoption. Software-only dynamic line rating is difficult to defend, in his view, because it could be assembled from a weather service and an AI assistant. A sensor on the line cannot.
Inicio
Where should a project actually go?
Romain Batby, CTO and founder, uses AI to answer one deceptively simple question: where should an energy project be located?
Ideally, projects would be built in the best locations rather than the easiest to permit. Today development is fragmented and uncertain, with developers spending months navigating land constraints, regulations and grid connection risk.
Inicio applies AI to automated site analysis, large-scale development planning, and translating maps into GPS-level insights. The platform then brings every risk — from landowners to city councils — into a single workflow.
Data, and the barrier per country
Inicio works with 20 developers across France and Italy, with a team of 15 combining AI, grid systems and energy expertise.
Asked where the data comes from, Batby acknowledged a real barrier to entry in each new country. The data itself comes from across the web: open sources, easily accessible records, and scarce documents that AI makes it possible to gather and interpret. Revenue combines annual licences with fees when projects are developed.
Romain Batby, CTO and founder of Inicio pitching at Energy Tech Summit 2026
Metroscope
One percent, across every plant in the world
Aurélien Schwartz, CEO and founder, is a former researcher on nuclear plant monitoring, and he addressed the audience as though they were nuclear operators.
What such an operator wants is maximum output with minimum maintenance. Showing a real pressurized water reactor case, he pointed to eight or nine simultaneous problems across the steam cycle causing actual losses.
That waste is the pitch. A plant might lose roughly 10 megawatts out of 1,000, with close to 1% of energy wasted at plant level worldwide.
The technology is physics-inspired AI built on an inference engine. EDF uses it across its nuclear fleet and claims a substantial net gain in fleet production. Schwartz also mentioned supporting Ukrainian operations during wartime to deliver electricity as safely and reliably as possible.
In the previous year alone, he said, the company saved 350,000 tons of CO2 equivalent — by displacing fossil generation through improved nuclear output and by increasing gas plant efficiency.
Spun out of EDF, and hybrid by design
Asked how dependent the company is on EDF, Schwartz explained its origins as a spin-off from EDF R&D. The three co-founders were researchers there, and once the technology proved itself they struck a deal to create a startup, with EDF providing seed investment.
On whether this is a frontier model, his answer was hybrid. The company uses years of second-by-second plant history to calibrate digital twins combining machine learning with physics. Where a physical equation exists, they use it. Where it does not, they apply supervised and unsupervised learning instead.
Splight
Enough electricity to power New York
Cheikh Dramé opened with a comparison. Imagine wasting enough electricity to power New York City for a year. That, he said, is roughly what happened in 2024: around 50 terawatt hours of renewable energy curtailed, equivalent to €1.5 billion in unearned revenue.
The cause is capacity that exists but cannot be used. Because distribution and transmission lines operate at around 55% of capacity, renewables get curtailed.
Doubling usable capacity
Splight’s dynamic congestion manager moves the traditional thermal operating limit upward, doubling and sometimes tripling usable capacity when paired with dynamic line rating.
The benefits distribute across the grid. Generators reduce curtailment. Batteries improve their economics. Large loads interconnect faster. Utilities gain reliability, since the system also acts as an automated control scheme.
Dramé cited over 30 customers worldwide, deployment across around 66 gigawatts of assets, roughly €50 million in additional generator revenue, and about 850 kilotons of CO2 avoided. Founded in 2021, the company has raised 26 million.
Regulation, and a live deployment
Asked how the company handles regulation rather than physics, Dramé described engaging regulators directly. In the northeastern US, performance-based ratemaking is pushing utilities toward non-wires alternatives rather than simply building more transmission. Elsewhere, the company works through generators losing money to curtailment, who then bring the solution to their utilities.
In Chile, the company installs monitoring devices alongside the software. The congestion manager works with fast-responding assets — solar, wind, some combined cycle, battery storage — calculating set points so that when a contingency occurs, generation trips down progressively rather than shutting off entirely.
WindBorne Systems
Most of the atmosphere is unobserved
Karen Ye, Director of Commercial Growth, opened with a storm that swept through Portugal and Spain earlier in the year, leaving a million homes without power and costing lives.
Forecast uncertainty drives disaster preparedness, and that uncertainty comes from the data feeding the models. Remarkably, only around 15% of the atmosphere is adequately observed today.
Balloons that stay up for weeks
WindBorne operates what Ye described as the largest constellation of weather balloons on the planet. Traditional balloons fly for two hours. These average over two weeks, continuously profiling the atmosphere and collecting data over oceans that go unobserved.
The company has launched over 7,000 balloons, with more than 300 aloft at any time from over 15 launch sites. That data feeds in-house AI models generating forecasts for utilities.
Returning to the storm, Ye showed that the company’s models picked up its intensity three days ahead of leading global forecasting models, then asked what a utility would do with that foresight.
The economics of selling data
Ye reported over $15 million in revenue last year, primarily from US government work and energy hedge funds. All-in balloon cost including hardware and labour is under $1,000, and because the company sells data rather than launches, longer flights mean more data. Gross margins run 60 to 70%.
A juror pressed on where the value sits and who pays. Ye distinguished two streams: balloon data sold to governments, and forecasts sold to industries like hedge funds. Importantly, the company does not sell all its data, because that is the moat — and as the constellation scales toward 10,000 balloons, that differentiation grows.
Yuon
Sized around the morning peak
Kajo Krummenacher, Chief Product Officer, closed the track on district heating, a cornerstone of Europe’s energy transition attracting billions in new investment.
Today’s networks operate with very limited intelligence. Operating costs are driven by peak demand, which creates three problems: limited capacity, preventable thermal losses and higher emissions. Utilities investing heavily in infrastructure are therefore leaving value on the table.
A typical network shows a large morning load peak, and that peak is what limits an operator’s use of expensive infrastructure. Yuon’s system learns how each network behaves, forecasts demand, and optimizes flow in real time to lower it.
Crucially, it integrates into existing systems: no hardware replacement, no infrastructure rebuild, software only. Krummenacher put the results at more than 20% lower operating costs, up to 60% lower peak loads and around 30% less CO2.
How invasive is deployment?
It depends on the operator’s existing setup. Where a guidance system and smart meters are already in place, it is software only. The work is getting the integration right, which is why the company is building libraries of compatible systems.
Kajo Krummenacher, Chief Product Officer of Yuon pitching at Energy Tech Summit 2026
Takeaway
Across nine pitches, one insight kept reappearing in different clothing: the capacity is already there, and what is missing is the ability to see it.
Heimdall and Splight both argue that transmission lines carry far more than operators dare use. Metroscope finds a lost percentage inside nuclear plants that are otherwise running fine. Yuon and Dyneo both target heating networks sized around a peak that better forecasting could flatten. WindBorne’s entire proposition is that most of the atmosphere goes unobserved.
AI for energy, on this evidence, is less about building new infrastructure than about measuring what already exists precisely enough to stop wasting it.
Energy Tech Challengers returns at Energy Tech Summit 2027 in Bilbao, April 7–8. Do you want to watch the next generation of energy startups pitch live, or take the stage yourself?
