Whoever Powers the Chips Wins the Race

The hardest megawatt to deliver is never the one inside the building. It's the one at the fence.

I've spent more than fifteen years building and energising power and data-centre infrastructure across the Gulf, North America and Europe - over 300 megawatts delivered, including a 240 MW project in Abu Dhabi. In all that time, the part that broke schedules was never the switchgear, the chillers, or the racks. It was the stretch of work nobody puts on the cover of the brochure: getting enough power to the boundary of the site, and the paperwork that allows you to switch it on. That was always the job. What's changed is that the AI race has taken this old, unglamorous constraint and made it the whole game.

The constraint the market keeps underestimating

Everyone is counting GPUs. Analysts track chip allocations the way commodity traders track oil barrels -  who has the Blackwells, who's in line for the next Vera Rubin shipment, whose cluster will train the next frontier model. It's a reasonable thing to track. It's also no longer the constraint that decides who wins.

Contracted demand for AI compute is running well ahead of the grid's ability to power it. In the United States alone, generation and storage projects seeking grid interconnection have piled up into a backlog north of 2,000 gigawatts - more than the country's entire installed capacity, sitting in queues rather than producing power. The median wait from application to commercial operation is approaching five years; for large loads like data centres, some utilities are quoting seven. In Texas, data centres now make up the large majority of a queue that dwarfs the grid they're waiting to join. PJM's capacity auction failed to clear its own reliability target for the first time in its history. None of that is a chip story. It's a wires-and-permits story, and it's the story that will actually determine where AI infrastructure gets built over the next decade.

The bottleneck has moved from silicon to substations from chip fabs to planning offices. Nobody wants to write that headline, because it's less exciting than a keynote about the next model. But it's the one that matters to anyone actually trying to break ground.

The powered-site ladder

Over the years I've come to think about this as a ladder with five rungs. You need the right land. You need control of it. You need a firm grid connection. You need planning and permits. And you need energisation power actually flowing, tested, and live.

Here's the thing almost nobody outside the industry appreciates: the risk and the time are not evenly spread across those five rungs. Finding land is comparatively easy. Energising a site, once everything above it is in place, is largely mechanical. It's the middle two rungs, grid and permits where almost everything that can go wrong, does. Projects don't die dramatically on a construction site with cranes and headlines. They die quietly, in an interconnection queue, or in a planning office, months or years after the announcement that made the news.

If you want to know whether a data-centre project is real, don't ask who the anchor tenant is. Ask where it sits on this ladder.

Where the value actually gets created

Land is available to anyone with capital. GPUs, eventually, are available to anyone with a purchase order and patience. Neither of those is scarce in the way that matters — they're expensive, but they're not hard in the sense of requiring judgment, relationships, and years of accumulated scar tissue to get right.

The scarce thing, the thing that's actually hard to buy is the work of converting an unpowered parcel of land into a grid-connected, permitted, energised site. That work retires a specific kind of risk: the risk that a project stalls indefinitely in a queue, or gets rejected at planning, or wins its permit but discovers the substation upgrade costs more than the project itself. Once that risk is gone, you're not looking at the same asset anymore. A permitted, powered site is fundamentally different from — and far more valuable than — the field it started as, because someone has already absorbed the years of uncertainty that would otherwise sit on the buyer's balance sheet.

Why this is a craft, not a capital game

Here's what capital alone can't do: it can't jump an interconnection queue. It can't shorten a utility's study timeline. It can't make a planning authority move faster because the applicant has a larger balance sheet. Interconnection processes, utility relationships, grid studies, planning regimes, energisation sequencing these are won market by market, by people who have sat across the table from the same utilities and the same planning officers before, who know which studies to commission early, which objections to pre-empt, and which relationships actually shorten a timeline versus which ones just feel like they should.

I've learned this the direct way on live projects, under real deadlines, in jurisdictions with very different rules for how power and permits actually get granted. There's no shortcut for that kind of pattern recognition. It's earned one project at a time.

Where Storm fits

This is the work Storm does. We operate on the hardest rungs of that ladder: sourcing strategically located land, securing grid capacity, and taking projects through planning and permitting to deliver energised sites that are genuinely ready to build on. The people who run compute, and the people who finance what runs on top of it, shouldn't have to carry land risk, grid risk, and permitting risk themselves. Our contribution is taking that risk off the table before anyone else has to touch it.

We're building this Europe-first, with a small number of adjacent markets under active evaluation. That's a deliberate choice, not a default. Europe combines real, growing demand for AI infrastructure with power and permitting realities that are genuinely local market by market, utility by utility, planning authority by planning authority. That's exactly the kind of terrain where disciplined, patient site development turns scarcity into durable value, rather than just a longer queue.

Why now, and why here

AI demand is global. Power and land are not, they're stubbornly, unavoidably local. That mismatch is precisely where the opportunity sits. My own track record was built in the Gulf and beyond; Storm's home market, deliberately, is Europe. The lesson travels even when the geography doesn't: the fundamentals of getting a site powered and permitted are the same whether you're in Frankfurt or Aragón. What changes is the specific utility, the specific regulator, the specific queue. Getting those specifics right, market by market, is the entire job.

The work we've chosen

The AI economy will be built by whoever can actually deliver power and permitted sites at scale, not by whoever announces the biggest chip order. That's unglamorous, difficult, multi-year work. It doesn't fit neatly into a keynote slide. But it's the work that decides, in the end, whose projects get built and whose stay stuck in a queue. It's the work Storm has chosen to do, and it's the reason I've joined to lead it.

The AI race won't be won by whoever buys the most chips. It will be won by whoever can power them.

This report may contain forward-looking statements, projections, estimates, market data, and information obtained from third-party sources. Such statements are inherently subject to risks, uncertainties, assumptions, and changing market conditions, and actual outcomes may differ materially. Third-party views, forecasts, and data referenced herein do not necessarily reflect the views of Storm Group.

While Storm Group believes the information contained herein to be reliable at the time of publication, no representation or warranty, express or implied, is made as to its accuracy, completeness, or continued relevance. Readers should conduct their own independent assessment and consult appropriate professional advisors before making any commercial, financial, or investment decisions.

Copyright 2026 Storm Group AG. All rights reserved.*