China sank its data centers into the Pacific to solve the water problem. Microsoft solved the same problem without leaving the parking lot. Here's which one is actually coming to a data center near you.
Back in a prior life, I was in charge of my employer's IT department, among other things. We switched to Voice-Over-Internet-Protocol (VOIP) telephone systems, which in short, meant lots of extra network switches getting loaded into our broom-closet server room. The heat was tropical (and also unexpected in our planning - well, we did expect a heat rise, just not the furnace that resulted!). Ironically, I was also in charge of the office environment on the other side of the wall - where almost all of the clerks spent most of the year (winter, definitely), in layers of sweaters and personal desk area heaters.
I remember saying to the building's engineering staff, "gee, it's too bad we cant somehow put a water-based server rack cooling system in place here, and pump the heat into the other department. Win-win!" (though not sure it would have been so during the summer months, where some of the space heaters were replaced by cooling fans).
Unfortunately, that idea was easier said than done -- the one-time cost to do it would have been excessively high, though I did point out that our "4-year life" for all the network switches and servers would likely fail downward to about 2 years if we don't deal with the heat problem -- and that would be (back then) about a $20,000 cost. Every. Two. Years (or so).
I had shifted to other employment focuses a year or so later, so this became somebody else's problem.
The organization was mid-sized -- about 150 FT people, with about another 100 PT. Big for the 8,000 population of the community, but not exactly ginormous. And the heat in the server room, AFTER we did everything we could think of to cool the space, was at best around 35C.
This was amateur hour compared to what a modern small data centre would experience.
But the take-away for all of these modern centres that are using the closed-loop approach: That's not conceptually a new idea -- buildings have used "radiator heating" for well over a century now (before central air and electrical heating "modernized" buildings). Or dare I say, the cooling system for internal-combustion engines (like cars). The system is great until it breaks -- ask anyone who actually lives/works in such a heated environment. But my main point -- instead of chillers (or alongside the chillers), the main emphasis is how do you cool the water BEFORE pumping it back into the system. Answer: heat sink it to heat buildings, communities, greenhouses, vertical growing environments, etc., etc., At scale, this is back to the same problem I had in my earlier story -- theory is good, but how do you rewire (or rather, "re-pipe") entire cities so that you can heat everywhere with the data centre heat? (and what to do for the summer, when everyone needs air conditioning more than heat?). Not sure if heat-sinking this into the ground is practical -- and somehow drawing on that heat during winter (if the data centre heat isn't sufficient).
Thank you for this. The broom closet story alone is worth the price of admission, and yes, the "well, we did expect a heat rise, just not the furnace" line made me laugh out loud.
You're right that closed-loop cooling isn't conceptually new. Radiators, car engines, same physics, just applied to silicon instead of pistons. And you've landed on exactly the right follow-up question, because "cool the water before it goes back in" and "heat sink it somewhere useful" are two different engineering problems, and the second one is where this actually gets interesting.
Turns out you don't have to re-pipe an entire city. It's happening already, at the neighbourhood or campus scale, wherever the plumbing already exists or gets purpose-built. Two examples close to home: an Equinix data center in Markham, Ontario now heats nearby condos, a university campus, two high schools, and two community pools, through Markham District Energy. In downtown Toronto, Telehouse Canada's data centers feed into Enwave's district system, pairing deep lake-water cooling with the same waste-heat recovery. Quebec's QScale is piping surplus heat from its Lévis campus straight into neighbouring greenhouses, which is basically your idea, just for tomatoes instead of clerks.
Alberta is the cautionary counterpoint. A University of Calgary study flagged the province's limited district heating infrastructure as the main thing standing in the way of doing this at scale there, which is your "re-piping the city" problem showing up exactly where you'd predict it.
One more thing that validates your instinct: data center waste heat is "low-grade," usually around 35 to 40 degrees Celsius, well short of the roughly 65 degrees most district heating systems want. So a heat pump has to boost it, which costs electricity. It's not a free lunch; it's the same tradeoff logic as everything else in this series, just relocated one step downstream.
Your summer question is the one nobody's actually solved yet. A few projects are experimenting with seasonal thermal storage, banking summer heat underground to pull back out in winter, but it's early and unproven at scale. If you want to watch one industry actually wrestle with your exact question in real time, that's the thread to pull.
Thank you for your question. It actually caused me to write a 3rd part to this series: District Heating Turns Data Center Waste Into Warmth. Coming out in a couple of days!
Thank you for your question. It actually caused me to write a 3rd part to this series: District Heating Turns Data Center Waste Into Warmth. Coming out in a couple of days!
I'll look forward to reading that. As far as I can tell the biggest barrier is simply the relatively low value of low-temp heat (below 50C). And the answer to that is complex - after thinking about it a fair bit, my thoughts are a) it needs a network effect to create enough value to be worthwhile (eg a somewhat standardized 'system' / method to have access and get value), and b) in locations where there is no existing infrastructure, "government" will have to be involved to incentivize, build, or mandate that it happen.
Now leave aside for the moment whether this is feasible in a world where builders of data centres are choosing locations based on cost (what I suggest above will cost money), and what the cost point might be to keep them from just going to lowest-common-demoninator: what's actually interesting is what this might look like. A regional/municipal 'heat dump and pump' network (excess heat and withdrawal at some base rates)? A 'field of dreams' for greenhouse operators / breweries / vineyards (what other industries might value?) ? Just collocating with district heating? I don't know. [And within these, what are the characteristics of local industries that are needed - universities / hospitals / might be some examples.]
And are there any essential services that need to be incorporated eg 'heat/cold provider of last resort'? Price setting?
This is kind of thinking out loud more than a plan.
I think I was being particularly dense when I asked that question, thinking it was some Microsoft novelty, now on looking again I realise it's the submerged di-electric liquid approach. Which I'm sure Microsoft has tweaked some but there is info about it out there.
Back in a prior life, I was in charge of my employer's IT department, among other things. We switched to Voice-Over-Internet-Protocol (VOIP) telephone systems, which in short, meant lots of extra network switches getting loaded into our broom-closet server room. The heat was tropical (and also unexpected in our planning - well, we did expect a heat rise, just not the furnace that resulted!). Ironically, I was also in charge of the office environment on the other side of the wall - where almost all of the clerks spent most of the year (winter, definitely), in layers of sweaters and personal desk area heaters.
I remember saying to the building's engineering staff, "gee, it's too bad we cant somehow put a water-based server rack cooling system in place here, and pump the heat into the other department. Win-win!" (though not sure it would have been so during the summer months, where some of the space heaters were replaced by cooling fans).
Unfortunately, that idea was easier said than done -- the one-time cost to do it would have been excessively high, though I did point out that our "4-year life" for all the network switches and servers would likely fail downward to about 2 years if we don't deal with the heat problem -- and that would be (back then) about a $20,000 cost. Every. Two. Years (or so).
I had shifted to other employment focuses a year or so later, so this became somebody else's problem.
The organization was mid-sized -- about 150 FT people, with about another 100 PT. Big for the 8,000 population of the community, but not exactly ginormous. And the heat in the server room, AFTER we did everything we could think of to cool the space, was at best around 35C.
This was amateur hour compared to what a modern small data centre would experience.
But the take-away for all of these modern centres that are using the closed-loop approach: That's not conceptually a new idea -- buildings have used "radiator heating" for well over a century now (before central air and electrical heating "modernized" buildings). Or dare I say, the cooling system for internal-combustion engines (like cars). The system is great until it breaks -- ask anyone who actually lives/works in such a heated environment. But my main point -- instead of chillers (or alongside the chillers), the main emphasis is how do you cool the water BEFORE pumping it back into the system. Answer: heat sink it to heat buildings, communities, greenhouses, vertical growing environments, etc., etc., At scale, this is back to the same problem I had in my earlier story -- theory is good, but how do you rewire (or rather, "re-pipe") entire cities so that you can heat everywhere with the data centre heat? (and what to do for the summer, when everyone needs air conditioning more than heat?). Not sure if heat-sinking this into the ground is practical -- and somehow drawing on that heat during winter (if the data centre heat isn't sufficient).
Lots of engineering questions!
Thank you for this. The broom closet story alone is worth the price of admission, and yes, the "well, we did expect a heat rise, just not the furnace" line made me laugh out loud.
You're right that closed-loop cooling isn't conceptually new. Radiators, car engines, same physics, just applied to silicon instead of pistons. And you've landed on exactly the right follow-up question, because "cool the water before it goes back in" and "heat sink it somewhere useful" are two different engineering problems, and the second one is where this actually gets interesting.
Turns out you don't have to re-pipe an entire city. It's happening already, at the neighbourhood or campus scale, wherever the plumbing already exists or gets purpose-built. Two examples close to home: an Equinix data center in Markham, Ontario now heats nearby condos, a university campus, two high schools, and two community pools, through Markham District Energy. In downtown Toronto, Telehouse Canada's data centers feed into Enwave's district system, pairing deep lake-water cooling with the same waste-heat recovery. Quebec's QScale is piping surplus heat from its Lévis campus straight into neighbouring greenhouses, which is basically your idea, just for tomatoes instead of clerks.
Alberta is the cautionary counterpoint. A University of Calgary study flagged the province's limited district heating infrastructure as the main thing standing in the way of doing this at scale there, which is your "re-piping the city" problem showing up exactly where you'd predict it.
One more thing that validates your instinct: data center waste heat is "low-grade," usually around 35 to 40 degrees Celsius, well short of the roughly 65 degrees most district heating systems want. So a heat pump has to boost it, which costs electricity. It's not a free lunch; it's the same tradeoff logic as everything else in this series, just relocated one step downstream.
Your summer question is the one nobody's actually solved yet. A few projects are experimenting with seasonal thermal storage, banking summer heat underground to pull back out in winter, but it's early and unproven at scale. If you want to watch one industry actually wrestle with your exact question in real time, that's the thread to pull.
Thank you for your question. It actually caused me to write a 3rd part to this series: District Heating Turns Data Center Waste Into Warmth. Coming out in a couple of days!
Is there a good source to read more about the microsoft approach, i.e. basic explanation of the technology?
Thank you for your question. It actually caused me to write a 3rd part to this series: District Heating Turns Data Center Waste Into Warmth. Coming out in a couple of days!
I'll look forward to reading that. As far as I can tell the biggest barrier is simply the relatively low value of low-temp heat (below 50C). And the answer to that is complex - after thinking about it a fair bit, my thoughts are a) it needs a network effect to create enough value to be worthwhile (eg a somewhat standardized 'system' / method to have access and get value), and b) in locations where there is no existing infrastructure, "government" will have to be involved to incentivize, build, or mandate that it happen.
Now leave aside for the moment whether this is feasible in a world where builders of data centres are choosing locations based on cost (what I suggest above will cost money), and what the cost point might be to keep them from just going to lowest-common-demoninator: what's actually interesting is what this might look like. A regional/municipal 'heat dump and pump' network (excess heat and withdrawal at some base rates)? A 'field of dreams' for greenhouse operators / breweries / vineyards (what other industries might value?) ? Just collocating with district heating? I don't know. [And within these, what are the characteristics of local industries that are needed - universities / hospitals / might be some examples.]
And are there any essential services that need to be incorporated eg 'heat/cold provider of last resort'? Price setting?
This is kind of thinking out loud more than a plan.
Thank you again….it gave me some inspiration to dig a little further. https://www.thesanity.org/p/district-heating-turns-data-center?r=8ic6c2
That is an interesting solution, isn't it? Let me take a look.
I think I was being particularly dense when I asked that question, thinking it was some Microsoft novelty, now on looking again I realise it's the submerged di-electric liquid approach. Which I'm sure Microsoft has tweaked some but there is info about it out there.
Thats ok..... Its still an interesting topic. I enjoyed the research and writing about it. Thank you again