Power, Water, and Permission
Listen to multiple data-center presentations in a row, and the marketing falls away to reveal a candid picture. The slides smile on growth – rack densities increasing to figures that would have sounded absurd five years ago, AI workloads stretching every forecast upward. But beneath the optimism, the same three fears continue to emerge, no matter who is on the podium: can you get the power, can you manage the heat, and will the town let you build at all?
The eleven presentations from this year’s summit by operators, developers, engineers, equipment makers, and the consultants who orbit them scarcely agree on anything. However, they align when it comes to the shape of the problem. And the most revealing moments stand to be the ones where they silently contradict.
The wall is the wire
The two national organizations that began with a broad perspective converge on the same uncomfortable fact: Spain is not limited by supply. The nation obtains much of its power from renewable sources. The challenge is transporting it - grids, substations, connection points, the unglamorous middle of the electrical system. The capacity on the highest voltage levels is already full; the cheaper connections come with waits of several years.
So the map is redrawing itself. The saturated triple-hub of Madrid, Zaragoza, and Barcelona is being replaced with smaller regional nodes of between five and twenty megawatts. It is chasing places where a slot for a connection is still available, and the land is cheaper. So, you want a place cold enough to cool for free. A cable manufacturer, of all the presenters, made the point most literally: its job is to bring that connection into the building. It used up its spot to remind the room that the wire itself is now a strategic option – copper or aluminium, embodied carbon, a twenty-five-year life that, properly installed, stretches beyond forty. Copper, by its numbers, is a fraction of aluminium’s carbon emissions but several times its price; treat the figures as the vendor’s, not gospel.
When the people paid to be optimistic can’t agree within a factor of two, the honest answer is: nobody knows, & it depends on the hyperscalers.
Here is the first crack worth noticing. When asked where Spain ends up by 2030, the two macro speakers are wildly at odds – one draws a path of just a few thousand megawatts, the other considers around one gigawatt conservative and two ambitious. The 2025 anchor points are nearly the same for both; their end points differ by more than twofold. Both are cheerleading figures from organizations that have a stake in the line going up. The useful takeaway is the range, and the reason for it: the wild card in this is how excited the hyperscalers get.
Then comes the heat
If power is the exterior wall of the building, heat is the interior wall. The cooling-equipment manufacturer presented the curve that is now being referenced by everyone: a conventional rack consumes 5-15 kW, an AI-training rack consumes 50-100 kW, and the highest density machine-learning racks exceed 200 kW, with a headline slide stating 600 kW/rack coming as soon as this year. At these densities, air cooling just goes blind. The room is turned to liquid, in stages: rear-door exchangers, then cold plates pressed against the chips, then full immersion in dielectric fluid.
A refrigerant chemist articulated the compromise at play in the chemistry: As chips approach a thousand watts each, the industry is caught between coolants that are friendly to the climate and coolants that are flammable – you get one or the other, essentially. And a helpful reframing emerged multiple times: the traditional efficiency ratio, which measures a facility’s total power draw against its IT load, is biased in favor of air-cooled designs since the server fans consume a fifth of that IT load. Measure by the compute that actually performs the work, and the numbers get even uglier, and the argument for liquid just keeps getting stronger.
The new battleground is water
The newest concept at the summit had nothing to do with electricity. As loops go liquid, the water inside them is also becoming safety-critical – and nobody is looking at it. One monitoring company made its whole pitch around that blind spot: coolant goes bad. Biofilm forms, metal rusts, and the very bacteria that building-water regulations focus on are now, according to vendors’ interpretations, making their way into data centers, too. Its differentiator is a patented method to score that risk on real-time; whether the claim holds up is a question for the technical track, not the marketing slide.
Water also turned into a currency that people compete for. One manufacturer’s entire product range is essentially one trade-off curve: dry coolers consume water and energy; open towers consume energy and water; and the rest of the options fall somewhere in the middle. You can be great at saving water, or you can be great at saving energy. Excelling at both, in the same climate, is the thing no one has figured out.
Build it like a product, not a cathedral
The forward-leaning developers share a world view that would make an old-school project manager want to run for the hills: the custom, hand-built data center is an end product. A private-equity-backed developer described what it terms “black-box construction” – open systems, integrated delivery, and discipline borrowed directly from the factory floor: error-proofing, marginal gains, modular pods that repeat. The logic is unsentimental. Capital and AI timelines don’t wait for an artisanal build.
The same instinct appeared in cable pitching, of all places – harmonise across sites, enforce transnational equivalence, demand the data ahead of time, and compare like with like. The days of the one-off are ending because no one has the schedule for it.
Who’s really in the room
Remove the logos and three characters emerge: the operator, with load and uptime ownership; the builder, with schedule and cost ownership; and the advisor – the engineering trade body and the real-estate consultancy – with diligence, permits, and politics ownership. Now a fourth voice is pushing in: the hardware vendor that doesn’t sell a box anymore but a partnership, making modelling ability and global scale the real product.
Beneath all of them sits the money, and it is consolidating. There is one operator who sold its estate; one developer is private-equity-backed; one equipment manufacturer is a branch of a sprawling industrial group. The industry is being financialised, and the new gospel of repeatability and selecting partners comes straight from there: when an investor owns the balance sheet, discipline stops being a virtue and becomes a requirement.
From completion to contribution
The common theme emerged, appropriately enough, in the soft-spoken consultancy slot: the notion of a “social license to build.” A permit and a connection to the grid are required, but they aren’t enough. The community has to want the thing or at least tolerate it. The reframe is from completion to contribution: stop delivering a sealed black box and start delivering a neighbour. The most tangible manifestation is heat reuse – using a district system to carry the waste heat, rather than sending it up in smoke. It’s a rare lever that’s both a carbon win and an acceptance win, and that right there is why a handful of decks reached for it.
The real subject
Read left to right, and the top line is actually about neither cooling nor cabling. It is about a sector reaching the boundaries of the material world more quickly than it anticipated, and cobbling together a new rulebook as it goes: get the power, tame the heat and the water, build like a factory, find the money, and make peace with the neighbours. The businesses that will still count in five years are the ones already solving these five as one problem.