Power to People: AI Hasn’t Picked Energy’s Winner

In 2024, a mid-sized grid operator in the American Southwest received an interconnection request from a data center. Single project. Load equivalent to a mid-sized city. Timeline: aggressive.

Little did the grid operator know this would become the new normal. Microsoft, Google, and Amazon are redrawing where transmission infrastructure gets built, which generation assets get financed, and how fast the grid has to move. Demand is now driving supply at a scale that rewrites the physics of grid planning.

This is what it looks like when AI-driven demand begins to concentrate and shape physical infrastructure.

We have seen this movie before, just on a different network. Web 2 economy seemed to promise decentralization: open architecture, distributed participation, value for everyone. What it delivered was a small number of chokepoints. Google controlled search discovery. Apple and Google together controlled app distribution. Meta controlled social reach. Amazon controlled e-commerce surfaces. Each sat between suppliers and their end users, extracting value proportional to that position because they owned the surface through which those assets reached the market. Digital technologies, then, AI now, were not designed to be born with that outcome. The technology accelerated it, rewarding whoever already controlled the distribution layer.

The energy sector is now approaching the same inflection point. The chokepoints are different: interconnection queues instead of app stores, dispatch platforms instead of ad networks, carbon registries instead of search indices, but the structural logic is identical. It can follow that same path toward concentration, or it can evolve differently.

We spend a lot of time at the intersection of economic structure and technological change in our podcast. The central question we keep returning to is not what AI can do in energy, but as AI reshapes the energy system, where does bargaining power go, and who captures the value?

Right now, that answer is not yet determined. The power structure of an AI-driven energy system is still being formed. That creates a window of opportunity, but one that will not remain open for long. This article proposes a way to read that window: by identifying the conditions under which value concentrates, and the conditions under which it remains more distributed.

The Gravity of Concentration

Before mapping where small players can win, it is worth being direct about what the default path looks like.

Technology does not democratize markets naturally. It optimizes them. The optimization rewards whoever already has the best inputs: the most data, the deepest relationships, the strongest balance sheet. In the internet (Web 2) economy, that meant digital technology accelerated the dominance of whoever controlled distribution. The companies that owned search discovery, app stores, social graphs, and cloud infrastructure captured value that vastly exceeded what the underlying asset creators, i.e., the developers, the publishers, and the sellers, ever earned. The assets were distributed, but the control points were not.

Energy has structurally equivalent control points. Instead of app stores and ad networks, the chokepoints are grid interconnection queues, industrial system interfaces, regulatory approval cycles, and long-term offtake agreements. These are the surfaces through which energy assets reach markets. And like the internet’s distribution layer, they tend to consolidate over time, rewarding whoever sits between the asset and the end user.

The default trajectory, absent deliberate countervailing conditions, is concentration at those chokepoints. The rest of this piece is about what makes a different outcome possible.

Six Conditions. Three That Are Decisive.

We have identified six conditions that, when present, create genuine space for new entrants to build durable positions and capture value within the energy system. These conditions are not a simple checklist where any single factor is sufficient. They reinforce each other. The more conditions are present simultaneously, the harder it is to displace the position.

1. Niche Demand Incumbents Structurally Cannot Serve

Large energy companies have minimum market size thresholds. Below a certain scale, a market is not worth entering because its cost structures make it irrational. AI has made fragmented, sub-scale demand economically aggregable for the first time. Voltus packages thousands of small commercial and industrial electricity users, each too small for a utility to manage individually, into a grid resource that commands real market prices. Software-enabled coordination has made this kind of fragmented aggregation commercially viable in ways that manual customer-by-customer management could not easily support.

The same logic applies at the geographic extreme: Husk Powerserves rural communities in India and sub-Saharan Africa that often sit below the investment threshold of conventional grid-extension models, backed by development finance filling a gap the market will not touch on its own.

2. Speed and Depth Advantages in Specific Problem Types

Large organizations are not slow because of incompetence. They are slow because their processes were built for a risk environment where the cost of a wrong decision outweighed the cost of a delayed one.

AI has inverted that calculus in specific problem types. GridCARE‘s work with Portland General Electric identified more than 80 MW of incremental capacity for data center interconnections in 2026, showing how AI-enabled flexibility analysis can shorten the path from grid constraint to usable capacity.

The depth dimension is equally important: Envision Digital builds wind optimization models calibrated to specific turbine models, specific wind regimes, and specific microclimates that many broad-based equipment or platform providers may struggle to customize at asset-level granularity. Even small generation or availability gains on a large wind asset can translate into meaningful financial value. For incumbents, that precision is uneconomical. For a focused entrant, it is the entire business model.

3. Bypassing Physical Asset Requirements (Decisive)

The traditional entry barrier in energy is physical: you need assets to have a position in the value chain. If you did not own infrastructure, you were a vendor, not a player. AI is weakening that equation in three distinct ways, each progressively more structural.

The first is simulation replacing physical validation. ThinkLabs uses digital twins and physics-informed AI to run grid planning analyses that previously required months of engineering work and proprietary infrastructure. It owns no transmission lines and is nonetheless inside the grid planning value chain, through work with Southern California Edison, a position that did not exist for asset-light players a decade ago. The intelligence layer has separated from the asset layer.

The second is data replacing physical surveys. Traditional wind resource assessment often requires long on-site measurement campaigns, frequently measured in months and sometimes 12–24 months for meteorological data collection. AI combining satellite data, atmospheric modeling, and terrain analysis can compress early-stage screening and pre-feasibility assessment before full physical validation. For a small developer without a local team in a new market, this is the removal of an entry condition. The gate was physical. AI took it off its hinges.

The third, most radical form is virtual assets replacing physical ones. A demand response aggregator using AI to coordinate distributed batteries and flexible industrial loads can deliver some peak-capacity, flexibility, and ancillary services that would otherwise require conventional generation or grid upgrades. The entry barrier to providing dispatchable capacity has moved from ten figures of capital to a software platform. This is the bypass logic at its most structurally significant: the product and the asset are becoming more separable than traditional energy market structures allowed.

The key caveat is that this bypass is still partial. Simulations ultimately lead to physical build-out, and virtual capacity relies on assets owned by others. However, partial does not mean insignificant. What has changed is when and where value can be captured. The critical question for investors is whether companies can establish a defensible position in the intelligence or coordination layer before the economics consolidate back around asset ownership. Early examples such as ThinkLabs and GridCARE indicate that this is not only possible, but already underway.

4. Distribution and Interface Control (Decisive)

The internet (Web 2)’s concentration happened at the interface layer: whoever controlled the surface through which users and suppliers met controlled the market. Drawing the parallel, energy has an equivalent layer, less visible but equally consequential: grid interconnection queues, industrial energy management systems, demand response dispatch platforms, carbon registry interfaces. These are the points through which products reach end users and assets reach markets. And like the internet’s distribution layer, they are structurally prone to concentration.

Tesla’s virtual power plant makes the dynamic explicit. The asset layer is genuinely distributed, with distributed home batteries and solar systems, individually owned. But Tesla’s Autobidder platform decides when each asset charges, discharges, and enters the market. While the assets could be democratic, the remote control is not. Here, AI created a more efficient version of centralization, one layer up from where it previously sat. This is the pattern that concentrated the internet’s value at the interface layer. It is repeating in energy, in real time.

The industrial version is less visible but more entrenched. Many commercial and industrial assets are increasingly mediated by platforms such as Schneider’s EcoStruxure, which connects data from buildings, data centers, infrastructure, and industrial systems. For an AI entrant with a genuinely superior optimization model, the barrier lies in whether the incumbent platform opens its interface. If it stays closed, the capability has no path to market. This is why distribution control is decisive: a company can satisfy every other condition and still find itself structurally locked out. The due diligence question, therefore, shifts from model performance alone to interface power: who sits between this product and its end user, and what incentive do they have to let it through?

5. Capability Formation Outside Incumbents

In fields new enough that no incumbent holds a knowledge monopoly, the capability formation process is distributed by default, and that creates a durable structural opening. In early carbon removal, the institutional knowledge base was thin, fragmented, and still being built. What emerged instead were open communities like Airminers and OpenAir Collective, where lawyers published standardized contracts, engineers shared methodology frameworks publicly, and early market validation happened through collective iteration rather than proprietary R&D. This dynamic is particularly effective in fields such as carbon removal and other early-stage climate infrastructure markets, where knowledge is still being developed openly and collaboratively. In these environments, early entrants benefit not just from participation but from learning faster than others. As they do, they begin to build proprietary insights that, in more mature industries, would typically reside with incumbents. Over time, this accumulated knowledge becomes a competitive advantage, gradually shifting the field from open collaboration to more closed, differentiated competition. That transition marks the point at which the initial window of opportunity begins to close.

6. Regulatory Design (Decisive)

Regulation is the condition that most investors underweigh, and most operators overweigh. The reality is more precise: regulation is not a monolithic force. Its impact on market structure is entirely a function of its design, and it can both pull new entrants in or push them out. Three distinct dynamics are worth separating.

The first is a regulation that pulls in new entrants by creating markets no one could previously enter. When the EU’s Corporate Sustainability Reporting Directive mandated detailed carbon and ESG disclosure, most large companies could not fulfill the requirement internally, and the data infrastructure simply did not exist. This accelerated demand for carbon and ESG data-management software, including AI-enabled tools that help companies collect, structure, and report sustainability data. CSRD initially opened a competitive window because the requirement was specific enough to create demand, technically demanding enough to expose data gaps, and broad enough to support a software market. Subsequent EU simplification efforts show that reporting rules can also narrow, delay, or reshape the market they create.

The second dynamic is regulation that can still concentrate markets even when it is designed to accelerate deployment. The US Inflation Reduction Act expanded clean-energy incentives and introduced transferability, which allows many project owners to monetize credits even if they do not have sufficient tax liability themselves. However, one has to admit that it does not eliminate concentration risk. Smaller developers can still face fixed legal, diligence, financing, interconnection, and transaction costs that scale poorly.

Hence, when evaluating any new energy regulation, investors would need to probe into who can move fastest to turn compliance into commercial advantages. The better test is whether the cost of accessing that subsidy is fixed or scalable. Fixed costs favor large players. Costs that scale with company size, or can be distributed through platforms and standardized processes, keep markets more open.

The third dynamic is regulation as a precondition for market existence, where, without it, institutional-scale markets struggle to form reliably. Before the EU Carbon Removal Certification Framework, carbon removal had functional technology, willing buyers, and operational pilots. What it lacked was a government-backed certification framework that could make quality, additionality, storage durability, and verification legible to institutional capital. For investors in nascent energy sub-sectors, the CRCF logic matters: the absence of a regulatory framework is not always a sign that a market is free. Sometimes it means the market cannot form at all until someone writes the rules.

Where the Conditions Stack Today

Applied as a screen, the six conditions cluster favorably in the following areas right now.

Carbon and voluntary markets, where regulatory frameworks are still forming, data is not yet locked, and no incumbent has consolidated the analytical layer;

Rural and off-grid distributed energy, where the demand is structural, the incumbents are absent by design, and the bypass logic on physical assets applies most cleanly; and

The aggregation layer of virtual power plants, and more specifically, the optimization and dispatch intelligence sitting above distributed assets that no single player has yet locked up.

Grid-edge intelligence, including behind-the-meter optimization of buildings and industrial loads, where data remains fragmented and control points are still emerging;

Climate risk and physical risk analytics, where standards are still evolving, datasets are incomplete, and early movers can still build differentiated data advantages.

These are not permanent windows and may close quickly. The conditions that make them open, regulatory flux, unconsolidated data, and absent incumbents, are precisely the conditions that attract consolidation once the opportunity becomes legible.

AI has not yet determined a system-level winner in energy. But the timeline for that statement to remain true is measured in years.

Why This Is Not Just an Investment Thesis

There is a case for democratization in energy that goes beyond portfolio construction. Investors should hold it alongside the financial case.

Going back to the parallel, new entrants built the internet’s most valuable applications on top of open access. Yet the companies that controlled those access points captured a disproportionate share of the total value created, often more than the builders themselves. Openness at the application layer did not prevent concentration at the distribution layer. And because the internet operated without physical constraints, it never had to reckon with what happens when a concentrated system meets a real-world shock.

Energy has no such exemption. When the internet’s chokepoints failed or extracted too much, developers found workarounds, built new platforms, or moved to different distribution channels. The underlying infrastructure kept running. When energy chokepoints fail: a dispatch platform goes down, a single interconnection queue backs up, a dominant aggregator misprices flexibility, the lights go out. Distributed systems fail in smaller, contained ways. In a physical grid, that containment is the difference between a recoverable loss and a stranded asset, between a regional disruption and a systemic one. Concentration in the internet was a value distribution problem. Concentration in energy is an infrastructure risk.

The financial case for an open system goes further than resilience. A closed system optimizes the existing pie. An open system grows it. Energy is a larger and more complex system and the bread and butter of society and the economy. The total value available in an open, AI-coordinated energy transition dwarfs what a concentrated system would permit to exist. Investors positioned early at the coordination layer of an open system are not making a charitable bet on democratization. They are positioning for a larger return pool than concentration would ever produce.

There is also a durability argument that belongs in every investment memo on energy transition assets. Transitions that concentrate gains generate political resistance. Transitions that distribute gains build constituencies. If the communities hosting wind farms, the industrial operators providing demand flexibility, and the small developers bringing new capacity online are participants rather than bystanders, the transition is harder to reverse. For long-duration infrastructure investments, where regulatory risk compounds over decades, political durability is not an ethical preference. It is a material investment consideration.

AI has not yet determined a system-level winner in energy. The chokepoints are forming, the coordination layer is being built, and the window is open but measured in years, not decades. The investors who understand that an open system produces more to capture are the ones most likely to be on the right side of where the value goes.

Cover photo by Yang Dawei on Unsplash