The AI Response to Real-Time Energy Markets

Most discussions around AI focus on making knowledge work more efficient. Far fewer companies are attempting something harder: using AI to operate critical systems where decisions carry direct financial consequences.

Energy is one of those domains.

As power markets become increasingly volatile, industrial companies face a growing mismatch between the speed at which markets move and the speed at which organizations make decisions. Prices fluctuate continuously, portfolios span multiple assets and contracts, and procurement decisions can have material financial implications. Yet much of the industry still relies on workflows designed for a slower era.

To better understand what an AI-native operating model for energy might look like, Sif, Our Head of Growth and Community, sat down with Christian Rosan, Co-Founder and CTO of METEORIC, a company building what it calls a software-defined utility.

METEORIC is building an AI-native energy supplier for industrial customers. Its platform combines forecasting, optimization, procurement, and execution into a single operating system designed to help companies manage energy exposure in real time rather than through periodic, manual decision-making.

Q: From my perspective at the Campus, I see industries hit a wall when their operational speed cannot match market volatility. Where exactly does the financial value leak when a legacy operating model tries to handle a real-time, volatile market?

Value leaks when market speed and organizational speed no longer match. In spot power, prices move in 15-minute intervals, but many industrial companies still manage procurement through spreadsheets, emails, and quarterly advisor calls. By the time data is collected, checked, aligned, and approved, the market has often moved.

The leakage usually happens in three places: companies act too late, they do not have a live view of exposure and consumption shape, and decisions are poorly documented. Energy has defined processes and market-standard data exchanges, but execution is still very manual. METEORIC is built to turn that into a faster, transparent operating system.

Q: Meteoric is taking the leap into actual autonomous execution with ACE. What was the hardest part about moving your architecture from a predictive model to an active execution model? How do you design safety guardrails so a model can take financial actions autonomously without hallucinating risk?

The hardest part is accepting that execution is not a UX feature. It is a liability surface. In energy, a wrong decision can become a seven-figure problem. I would not let an unconstrained model place that bet without the right human in the loop.

For us, autonomy means execution inside clear guardrails: customer mandate, hedge limits, allowed products, volume bands, risk appetite, approval thresholds, and full auditability. ACE can analyze the portfolio, simulate scenarios, recommend actions, prepare approvals, and document the reasoning. But it does not hallucinate trades. It operates inside a governed workflow.

That is the realistic path: automate where reliability is high, keep control where consequences are high.

Q: A common hurdle I've heard from founders at the Campus building for physical industries is data fragmentation: legacy hardware, siloed asset data, and inconsistent meter standards. To train an AI operator that handles forecasting, optimization, and execution simultaneously, how did you solve this cold-start data problem? How do you get a new 50GWh client ingested and reliable on day one?

The cold-start problem in energy is less about exotic meter formats and more about fragmented context. Meter data and market communication often follow established standards. The harder part is that the customer’s energy reality is spread across sites, meters, contracts, invoices, approval rules, open positions, and people’s heads.

On day one, we build a structured portfolio view: sites, consumption, contracts, risk position, procurement rules, and missing data. Just as important, the system marks what is known, what is assumed, and what is uncertain. Once that structure exists, ACE can create value quickly. Every forecast, recommendation, approval, and outcome then becomes part of the customer’s procurement memory.

Q: Technical operators and investors are rightfully skeptical of natural language interfaces wrapped over complex enterprise workflows, as it can feel like a gimmick. Behind the conversational interface of ACE, what is actually happening technically? How does the system translate an ambiguous human prompt from a factory floor manager into a deterministic, optimized trading or procurement strategy?

I agree with the skepticism. A chatbot on top of a broken workflow is not a product. It is a demo.

For us, conversation is only the input layer. When a customer asks whether they should hedge more volume, ACE does not just generate text. It checks the portfolio, open position, forecast volume, market prices, risk limits, contract structure, and approval process.

The output is a decision package: current exposure, market context, options, recommendation, required approvals, and audit trail.

That matters because energy expertise is scarce. Some expert profiles take 1.5 years to hire. If AI lets each operator do the work of three or four people, it changes the operating model.

Q: Legacy utilities profit from market opacity and volatility, creating a structural disincentive for them to help customers optimize. By using AI to drive energy waste to zero, you are flipping that model. How do you structurally align your economics with the customer, and if we look ahead five years, what will look most absurd about the way industrial companies manage their operational stacks today?

The traditional energy model often monetizes complexity. If the customer does not fully understand timing, risk, fees, shape, imbalance, or operational friction, that uncertainty becomes margin for someone else.

We want the opposite model. We win when customers make better decisions with less effort: clearer risk, better timing, lower manual workload, and fewer hidden costs.

AI also changes the unit economics. If software reduces the cost to serve, we do not need opacity to create margin.

In five years, it will look absurd that 50 GWh industrial customers managed major energy exposures with spreadsheets, emails, and occasional advisor calls. Energy needs to become a real-time operating system, not a quarterly purchasing process.

‍

---

‍

The "Community Spotlight" series explores the conversations happening inside of Europe's most concentrated AI ecosystem. Through interviews between Sif Björnsdottir, Head of Growth & Community at the Merantix AI Campus, with founders, researchers, and operators that call the Campus their homebase, the series uncovers the shifts, constraints, and opportunities emerging at the frontier of AI—and what they reveal about where industries are headed next.