AI's Energy Problem Was Here Before the War. It Will Stay After.

The conflict in Iran forced a structural constraint into the open. But the math was already broken.

Mar 25, 2026

The conflict in Iran, now entering its fourth week, has rekindled a debate that was already building quietly for 24 months in technology circles: energy. Not as a footnote to the AI story, but as a structural constraint at its center.

cable network — AI Factories are one part of the story only, Photo by Taylor Vick on Unsplash

The Subsidy That Funds Today’s Prices

There is an important structural fact about current AI pricing that rarely surfaces in business conversations: the prices being paid today do not cover costs.

According to Axios, OpenAI is projected to burn $14 billion in 2026, up from $8-9 billion in 2025. Anthropic’s margins, while improving, remain under pressure from higher-than-expected inference costs. The comparison Axios draws is well-chosen: this is the “millennial lifestyle subsidy” applied to AI, in the same tradition as VC-funded Uber rides or Amazon’s years of zero profit. The gap between what users pay and what it costs to run these models is being funded by investor capital, not sustainable operations.

Both OpenAI and Anthropic are expected to go public. Public market investors will demand margin expansion, not subsidised pricing. The floor is explicitly temporary.

This matters because businesses building on current AI pricing are making a structural assumption that has not yet been tested. A company pricing its product around $3 per million input tokens for a frontier model is working off a cost floor being held down by investor capital and intense competitive pressure. When that floor shifts, the margin calculations shift with it simultaneously across every deployment. The question is not whether repricing happens, but when and at what pace.

The Energy Variable

Beneath the investor subsidy question sits a harder and older problem: electricity.

European wholesale electricity prices have been highly volatile in the past years. They went from roughly €35/MWh in 2020 to above €500/MWh at their peak in March 2022, driven by the Ukraine conflict and the collapse of Russian gas supply. They have since moderated, but the IEA’s latest data shows EU electricity prices for energy-intensive industries in 2025 still running at roughly twice US levels and nearly 50% above China. The Ember European Electricity Review 2026 notes that an increase in gas generation in 2025 pushed the EU’s fossil gas import bill up 16% year-on-year. Now, with the Middle East conflict disrupting LNG supply, Wood Mackenzie estimates that roughly 19% of global LNG exports have been removed from markets weekly, pushing TTF gas prices sharply upward and feeding directly into European electricity prices.

Data centers consume an extraordinary amount of power. By 2026, total global data center electricity consumption is projected to exceed 1,000 TWh per year, equivalent to the annual electricity consumption of Japan. AI inference is what is driving that growth, and the more capable the models become, the more compute each query requires.

The hyperscalers understand the exposure. Microsoft reopened Three Mile Island under a 20-year supply agreement to provide nuclear baseload for its data centers. This carries a historical weight worth noting: Three Mile Island was the site of the worst nuclear accident in US history in 1979, an event that effectively froze new nuclear construction in the United States for four decades. The fact that Microsoft is now reviving it signals something significant about how seriously these companies view the energy constraint. Amazon and Google are making equivalent bets on small modular reactor technology that does not yet exist at commercial scale.

A more anecdotal but telling signal: Google recently posted a Director-level role for Head of EMEA Clean Energy and Power, requiring 15 years of experience navigating European electricity markets, regulatory frameworks, and energy purchasing strategy. The role sits within Google’s Cloud and Technical Infrastructure organization. When a technology company creates a senior executive function dedicated to managing the electricity market, energy has moved from an operational cost to a strategic variable.

LNG Supply is at the heart of both the Iran conflict and energy cost structure. Photo by Chris Pagan on Unsplash

The Predictability Problem

The level of energy costs matters less than their predictability. A business can build a viable model around high, stable costs. What breaks cost models is volatility: it ruins multi-year pricing decisions.

This is a compounding problem for AI business models specifically. Token prices are set by providers who must forecast energy costs years in advance. When those costs are highly volatile, providers face an unpalatable choice: absorb energy cost swings in their own margins, or pass volatility through to customers whose own pricing models cannot absorb it either. Right now, investor capital seem to be absorbing the gap. That arrangement will not survive an IPO.

Under additional pressure from the Middle East conflict, this is not an abstract concern at all. It is a practical obstacle to the ROI calculations that justify AI adoption in the first place. If the cost of running AI at scale is unpredictable beyond an 18-month horizon, it becomes structurally difficult to build the business cases that drive enterprise implementation. The energy problem does not just constrain infrastructure. It constrains adoption itself.

Efficiency as the Structural Response

The discussions currently happening around more efficient models, whether Small Language Models, Edge AI, or World models, are fundamentally about energy economics expressed in technical language. A model running at the edge, on local hardware with a known amortisable cost, has decoupled from the energy market uncertainty of centralised cloud infrastructure. A smaller, specialised model achieving adequate performance at a fraction of the compute cost of a frontier model changes the energy equation for every deployment at scale.

The most efficient model is also the one most insulated from energy cost volatility. The race to the largest model turning into race to inference, the infrastructure constraints become more visible and the business case for efficiency intensifies. It may be the return of energetic frugality, albeit not just for CSR goals.

Three Things Worth Tracking

For those making decisions impacted by this, three questions follow from this analysis.

First, ask what share of current AI costs reflects structural pricing and what share reflects temporary investor subsidy. The Axios data on OpenAI and Anthropic burn rates is a starting point. Stress-testing unit economics against a 2x to 3x increase in compute costs may just be a three-year planning horizon exercise.

Second, treat energy access as a location and architecture decision, not just a procurement question. Operators with access to nuclear or stable renewable baseload at predictable prices are in a structurally different position from those depending on gas-linked electricity in volatile markets. On-premise or edge deployment for appropriate workloads is not only a privacy argument; it is increasingly an economic one.

Third, watch efficiency metrics as closely as capability metrics. The next wave of competitive advantage in AI infrastructure will not go to the largest model. It will go to the model that delivers adequate performance at the lowest and most predictable cost per query. That is what survives a repricing cycle.

Jeremy Beaufils is Executive Director of the Digital Disruption Chair at ESSEC Business School.