I spent last week at CERAWeek in Houston. Nine sessions over two days. Thousands of the sharpest people in energy, all in one building. And I left more convinced than ever that the industry is looking in the wrong direction.

AI was the word on everyone's lips, which was to be expected. What surprised me was how narrow the conversation still is.

The operational obsession

Almost every AI use case discussed at CERAWeek was operational. Predictive maintenance on offshore platforms, drilling optimisation, field worker support tools. The results are genuinely impressive. Companies are compressing work that used to take months into days, turning 30-day planning cycles into hours. The industry is already seeing real impact and real savings.

But here's what struck me: when the conversation shifted to commercial functions, to pricing, margin management, portfolio risk, the room went quiet. One major oil company's chief strategy officer, asked directly about AI in core business operations, essentially said they're still figuring out whether it's worth it.

That gap between operational AI maturity and commercial AI maturity is enormous. And it tells you something important about where the real unsolved problems are.

Why commercial is harder 

A framework came up several times across sessions that I think explains the lag well. Most of the industry is doing what you might call bolt-on AI: taking existing workflows and making them 30-40% faster. Humans still make the final call. It's low risk and easy to justify.

A much smaller group is asking a fundamentally different question: what does the operating model look like when AI handles the pattern recognition and humans handle judgment and exceptions? That's where pricing and margin management sit. A wrong AI-driven pricing call could be a regulatory exposure or a multimillion-dollar margin hit. The stakes are different, so the industry moves slower.

I understand the caution. But caution is not a strategy. The problems are not going away. They're getting worse.

The demand shock is real

The US side of CERAWeek made the scale of what's coming very concrete. Data centres that were requesting 10-20 MW two years ago are now requesting 500-1,000 MW. Multi-gigawatt campuses are being announced. Load is arriving in 12-18 months, but new generation takes 3-5 years. It's the central operational crisis for US utilities right now.

Capacity prices have increased dramatically. Consumer bills are being affected. And the political pressure is building fast. Energy affordability showed up in election campaign ads in New Jersey. Microsoft proactively committed to paying higher rates for data centre load to avoid the perception that AI is raising household bills. When tech companies start volunteering to pay more, you know the political dynamics have shifted.

Gorilla partner Capco flagged several of these dynamics in their North American energy trends outlook earlier this year, before CERAWeek even happened. They highlighted the shift from static capacity to a flexible grid, the need to segment data centre demand by load shape, and the affordability imperative that comes with all of it. CERAWeek confirmed those predictions are playing out even faster than expected. 

Two problems, one theme

What I found most valuable was seeing how the US and European challenges connect.

In the US, the margin pressure comes from the demand side. Who pays for peak capacity? How do you structure tariffs when load shapes are changing this fast? How do you keep capacity markets functioning when forecasts are moving quarter to quarter?

In Europe, the pressure comes from upstream. No domestic energy base, permanent loss of cheap Russian gas, LNG dependency that ties pricing directly to geopolitical shocks. When commodity prices spike, it hits every retailer's procurement cost immediately.

Different triggers, but the same underlying problem: energy retailers and utilities lack the visibility to manage margin in real time across their full value chain. The data exists but the processing speed doesn't. The organisational structures that would connect pricing, forecasting, hedging, and billing into a single decision framework mostly aren't there yet.

The window is open

One speaker put it in a way that stuck with me: "Close enough is never good enough. It can actually be a disaster." He was talking about the limits of general-purpose AI in specialised domains. I agree. General-purpose tools cannot manage the complexity of energy retail margin. You need domain-specific platforms built on the right data, the right logic, and the right understanding of how these markets actually work.

Every panel I attended landed on the same governance point: AI is a copilot, not a pilot. The human stays accountable. The AI makes the human faster and better informed. That's exactly how we think about it at Gorilla. We don't remove judgment from margin decisions. We make that judgment possible at the speed and scale the market now demands.

Nobody at CERAWeek was talking about the commercial margin problem with any real confidence. Plenty of people acknowledged it exists. Nobody had an answer. That's not a warning sign for us. That's the opening.

The industry is solving the operational AI challenge. The commercial one is next. And that's where we live.