That Runaway Datacenter Power Grab Is The Best News For Net Zero This Century

Opinion Datacenter power is a shocking business. The latest report from the International Energy Agency makes some hair-raising predictions, such as Irish datacenter electricity usage making up a third of that country's total juice budget by 2026.

Globally, datacenter infrastructure is expected to more than double over the same period, going from two to four percent of all electrical demand and adding the equivalent of a brand new medium-sized European nation. 

All this is down to the fashion for intensive data-crunching through cyber currencies, AI adoption, and plain old corporate data hogging. These may seem poor reasons to lumber the environment when we're supposed to be lightening the load, but it also highlights one of long-term facts of life for the big IT shed builders. You can't just plug these things in.

It isn't easy building a datacenter, even if you ignore the bits that chew on the data. Viable installations need big parcels of land that must be near to high voltage supplies and network connectivity. They must also be well away from risk factors like other industrial installations that could go messily wrong, environmental problems like flooding or fire, safely away from residential areas, and so on. 

Of all these, the high voltage power interconnects are often the most troublesome. If you're in the market for tens of megawatts, you really need to connect to your supplier at the network's core transmission line voltage, in the hundreds of kilovolts. For redundancy, you need two such connections. The commercial and practical implications of this can lead to many years' lead time. That's if the capacity's going to be there at all in the time frame. 

The nub of the matter is that datacenters aren't like any other electrical load. They're industrial systems that don't fit alongside other industries, they want to be close to financial, business and domestic consumers, just not physically, and their future demands for power absolutely must be sustainable and carbon neutral, but they demand extreme reliability. 

These aren't problems, these are opportunities. Compared to other infrastructures, such as transport or power, information is not beholden to legacy. We can't rip out all the roads and railways and rebuild them to match the needs of the 21st century: we're having enough trouble with replacing internal combustion with electric motors. We can't put Amazon warehouses or FedEx hubs on remote islands. Aviation and marine transport are going to pin that sort of logistics to hydrocarbons for the foreseeable. 

The logistics of datacenters, however, rely on photons. These are conveniently massless and thus extraordinarily cheap to transport very long distances. Go out under a clear dark night sky and look at the Andromeda galaxy. That's 24 million trillion kilometers away, yet its photons are delivered to your eyeball without emitting so much as a gnat's belch of carbon dioxide along the way. Admittedly, latency's a bit of a problem - you wouldn't want your gaming server out there. 

You can put a datacenter anywhere you like: as long as it can be hooked up to some glowing glass strands, it can do its job. As for the need for lots of land away from potential harm, there's a near infinite supply of the stuff that's no good for agriculture or gracious living, and most of it is therefore very cheap. 

Which leaves power. Everyone in the datacenter and hyperscalar business has an environmentally friendly story to tell, some sensible if limited, and some farcical. What none of them are is proportional to the threat of climate change, a threat which you may have noted is universal and terrifying. It's not something that companies can fix through competition.

• Europe's first exascale system will be slotted into modular containerized datacenter
• Microsoft hires energy mavericks in quest for nuclear-powered datacenters
• Mystery German chip fab sips on Gradiant's ultrapure water
• NVMe consortium polishes its specs to support computational storage

If only there was some lesson from history to guide their thinking.

That lesson exists, and is 70 years old almost to the day. The existential threat to humanity was nuclear armageddon, and the strategy to avoid this was mutually assured destruction. That relied on having retaliatory strike platforms that could survive a first attack, and submarines looked very promising. But submarines are vulnerable if they need to surface to replenish the oxygen diesel engines need - so navies seized upon the suggestion to make them nuclear powered. That way, they could run silently, and run deep – for as long as there are tins of beans to feed the crew. The US Navy decided it needed a small, powerful nuclear generator, and by 1954 it had one. USS Nautilus, the world's first fission-powered submarine, launched on 21st January that year. That was just a decade after the Manhattan Project produced its proof of concept: you can do it if you try. 

Seventy years on, this technology is being looked at as a civil power source, but is struggling to free itself from the stagnating legacy of earlier, far more massive nuclear programmes. There is cautious optimism and cautious funding. Not good enough. So imagine if the world's datacenter industry got together – yes, a fantasy, but these are needful items - and issued a specification for a standard, modular, small nuclear power plant. 

The key word, as always in IT, is standard. That will define the market and provide a stable financial model against which to raise funding. It has to be created in conjunction with regulators around the world, but none of the concepts are new and those conversations are already well established by those who've been trying to sell the concept for decades. It's the kind of low-risk, low-cost, environmentally positive move that governments can get behind without massive commitment of public funds, and made part of a (whisper it) industrial strategy that includes easing fiber provisioning and simple licensing. We've seen the UK change the rules to encourage spaceports, for heaven's sake, so why not for something that really matters? 

There are many more conversations to be had about sustainable datacenter growth, such as discouraging toxic loads and encouraging efficiency beyond just maximizing the bottom line.

But the kudos and practical impact of jump-starting cheap, flexible nuclear power that can be deployed far beyond the datacenter, providing the localized baseline support for renewables that the sector badly needs, would be a shining halo for an industry that badly needs one. ®