I've been writing about data centers for a long time. For all those long years, I've been hearing about how great liquid cooling is, and how everyone will soon be using it - to enable high power densities.
At the same time, for all those long years, I've seen a powerful lot of white space resolutely cooled by air, with no intention of ever being cooled by anything else. According to an Uptime Institute survey (see more on Monday), 85 percent of enterprise IT operators don't use direct liquid cooling, and the number who do use it (16 percent) is outnumbered by the number (22 percent) who do not plan to use it ever.
The sector is conservative, and no matter how great something is, data center operators won't adopt it till they have to. If air cooling works, they stick with it. If it ain't broke, why fix it?
So I am not about to give you one of those prediction pieces that say "this year will be the year of liquid cooling" But there are a couple of things to say.
Go with the flow?
Firstly, it's always worth mentioning high performance computing (HPC). Liquid cooling is normal there, and all the top supercomputers use it. That's one place where power densities are high enough to make liquid cooling the only sensible option. Similarly, hyperscalers are also prone to using it, when the situation demands
Next, this week, I saw two comparatively small announcements which signal increased interest.
One is an investment in a liquid cooling company. GRC (formerly Green Revolution Cooling) got a $28 million round of funding led by a whopping $25 million from SK Lubricants of Korea. Some people clearly see a serious future for the concept. There's a little more to this: SK Lubricants is a petrochemical company keen to gather some green credibility, and it sees GRC as a way to develop a new business line that has environmental credentials.
Secondly, this week. Inspur opened up a liquid cooled rack it makes for Chinese cloud giant JD to more customers. It's built in a factory, with a capacity of 100,000 liquid-cooled servers per year, and it will be sold pre-bundled into racks. So Inspur, too, thinks liquid cooling is heading for a bigger market.
Why might that be? Let's look beyond the details to a bigger picture.
In order to stave off global warming the world has to stop burning fossil fuels. We have to "decarbonize."
The difficult sectors to decarbonize are transport, which uses oil, and heating buildings, often done with natural gas. This has come about for sensible reasons. In the UK, for instance, gas is about six times cheaper than electricity for heating. If you have a carbon-based grid, it's expensive to burn gas, make electricity, and then use that electricity to make a much smaller amount of heat.
But we do need to decarbonize our home heating. That means we need to shift from burning gas to using electricity from renewable sources. And we need to find a way to do that economically.
That changes the value of things, says Max Schulze of the Sustainable Digital Infrastructure Association: . “Now, heat is a very valuable commodity, especially when made from electricity.”
Data centers use electricity. They use lots of it, and in some places, like Ireland for instance, that amount is causing serious concern. Because data centers are rich consumers and can afford to switch to renewable tariffs, and that renewable electricity is needed to decarbonize the grid. In Denmark, data center development went more slowly when it was realized that the rapid growth of data center projects jeopardized the country's renewable targets.
This is leading to a backlash against data center building.
But whenever electricity is used, heat is ultimately generated. So data centers could be creating an answer to the problem. Temperate nations need heat for homes and offices, and that heat should be made from renewable electricity. Data centers, by buying renewable power and using it, are doing just that. They are making heat from renewable energy. That heat needs to be shared - or sold - to the people that need it.
The trouble is, data centers make heat in the wrong places, and they make the wrong kind of heat
Data centers can offer to share heat, by pumping their hot air into district heating systems. But even in northern European countries, there may be no district heating system near enough to take that heat, because even in insulated pipes, hot air can't travel very far before it becomes useless warm or cold air.
In Finland, Microsoft agreed to site its new region next to a Fortum energy plant, in order to make sure its hot air could actually be used. In Sweden the Stockholm Data Parks invites data centers into a site already connected to district heating. In practice, that's they only way air-cooled data centers can avoid wasting heat - unless you take the more extreme approach of Qarnot and similar companies, and break your data center into pieces which can be installed in homes as "digital boilers"
In Norway, data centers are required to "consider" reclaiming their waste heat, but this will often just mean getting a report done to prove that it isn't economical to do so. A lot of data centers are reportedly being built with an external vent for their heat, so they can be called "heat reuse ready", making it someone else's problem. Across the whole of Europe, the Climate Neutral Data Center Pact is driving data centers to become carbon neutral. But its words on heat reuse are simply a promise to the "look for" ways to reuse heat - in most cases, that search will confirm that it's not economical.
At some point, these regulations and promises may have to start delivering - and at that point, the only way to share data center heat on a massive scale will be to improve the quality of that heat.
Hot water is much more transportable, and usable by other customers such as district heating systems.
It will take a lot more joined-up thinking than I see at the moment but, at some point there will be a market for data center heat, and at that point it will pay data centers to adopt liquid cooling to produce the best heat. They may even be required to do so.
Data centers don't always follow logic, but there is a reasonable argument that data centers will eventually have to go with the flow.