Devilishly Advocative: Microsoft Heats Ocean; Builds Skynet’s Safe Haven

Have you heard that Microsoft is testing underwater data centers? On the surface (well, actually on the ocean floor) it’s not a bad idea. Project Natick seals a node of servers in a steel pipe for an undersea adventure planned for at least 10 years. The primary reason is to utilize cold ocean temperatures to keep the machines cool as they crunch through your incessant Candy Crush Saga sessions.

microsoft-project-natick-squarePassive cooling is wonderful, and really drops the energy footprint of a data center, albeit a very small one which is being tested. Scaled up, I can think of another big impact: property taxes. Does anyone know what the law says about dropping a pod in the ocean? As far as I can tell, laying undersea cabling is expensive, but once installed there are no landlords holding out their hands for a monthly extraction. Rent aside, taking up space with windowless buildings sucking huge amounts of electricity isn’t going to win hearts and minds of the neighborhood. Undersea real estate make sense there too.

But it’s fun to play Devil’s Advocate, and this one immediately raised my eyebrow. I read as much Sci Fi as time allows, and am always interested to see which authors are registering the best technology predictions. This is the second time in short order that I turn to [William Hertling’s] work. Back in November, Google announced a project to add predictive responses to Gmail. This parallels the premise of [Hertling’s] Singularity Series which begins with Avogadro Corp. Another major point in that novel is the use of offshore data centers.

In the story those data centers are basically unmanned floating oil rigs. The installations become targets of theft as pirates board and cut their way into the shipping containers to get at the valuable user data stored within. Avogadro’s response is to outfit the rigs with weaponized robots, thereby giving the Artificial Intelligence a way of defending itself.

I see a big parallel if Project Natick takes off. Presumably there will be thousands of the data center pods. They don’t need to be weaponized, recovering them from the floor of the ocean would be more risk than reward for a would-be thief. Tapping undersea communications has long been a tool of spycraft so data security is something to consider. And if total deployment numbers were to reach millions we may have an ocean warming problem to face. But if we ever face a threat from an all-controlling AI these are the perfect safe-haven for it. In the Microsoft promo video below, Research Engineer [Jeff Kramer] even says:  “You can’t do anything about it if it screws up”.

Well, for now these pods will have hardline connections for both electricity and data, so you can always pull the plug. Unless the AI is smart enough to order wireless and solar (perhaps geothermal) redundancies before our small-minded species can realize the threat.

79 thoughts on “Devilishly Advocative: Microsoft Heats Ocean; Builds Skynet’s Safe Haven

  1. The great advantage to convective cooling as you describe it is that you no longer are losing energy with the inefficiencies of the vapor-compression refrigeration systems that most centers use – in theory, heat pipes would be sufficient to get heat to the outer shell. Generate 100W of waste heat, lose 100W of waste heat, not 130+ while paying for the difference with consumed electricity.

    1. and the only thing i can think about in the last 5 minutes is:

      You could have this SO much easier if you just put the Datacenter beside a River.
      If there happens to be a hydro power station nearby, allthebetter (–> cheap electricity).
      No need to drop a few tons of valuable raw materials in the sea and leave them to rot (do you really THINK theyre gonna pick them up later?).
      Much easier maintainance. no need to lay expencive underground wiring.
      No need to build a waterproof enclosure for your servers…

      1. I think the big advantage that they are looking at with this is flexibility, sure a datacenter on a river has the ability to be more cost effective, but it is also significantly larger in size, wich is sometimes unnessicarry, similarly many areas where this would be most useful do not have rivers, but are by the sea.

      2. Big data centres may raise a river’s the water temperature enough to kill its inhabitants. Raising the temperature of say the Pacific ocean will take a bit more effort (though it can be done by crowd sourcing the oxidation of carbohydrates).

  2. I hope they know how to deal with the microscopic marine juveniles of barnacles, anemones etc who love to cling where their filterable food comes past them with no work at all. Over time they will grow to fill the pipe and slow down heat transfer.

        1. There is a galvanic anti-fouling technology.

          Basically you run a small current (but high voltage) through the outside shell. I know its effective against barnacles, I would imagine its effective against a wide range of benthic life…

    1. pretty sure most of those sea life still needs lots of oxygen and therefor can only live near shoreline or at the surface of the water if the pods are sunk deep enough they should not matter…

  3. One such pod on the ocean floor is innocuous enough, and a pretty good idea. But if it’s successful and they deploy hundreds or even thousands of them, what would the thermal impact be to the ecosystem, ocean currents, etc?

      1. Reminds me of this dead serious discussion between two environmentalists about the impact on the environment from ozone generated by the electric motors in electric cars.

        1. Funny. Shaun has a point though… A single test block won’t be a huge problem, but an Apple, Microsoft or Google sized DC will raise the local water temperature by several degrees. This will screw up the local ecosystem significantly. Would also modify local currents and lets face it, how many of our douchebag corporation are going to pay to actually decommission these things? They’ll be left to rot. So much for environmentally friendly.

          1. This happens with nuclear power plants and desalination plants. It’s fine you just mess up a body of water that isn’t important or useful. Not everything is sacred.

          2. “The will be left to rot”
            It is not uncommon to intentionally sink ships to help rebuild reefs. It provides surface area that much see life lives on and provides shelter for fish.

    1. Insolation is on average about 1000W per square meter (during daytime, so lets include night and make it 500). The oceans cover about 3/4 of earth, or some 3.6×10^14 square meters. That’s 1.8×10^17 Watt, or 180 PW. US datacenters are expected to be using about 16 GW by 2020 (“140 billion kWh/yr”,, about half of which would be for just operating the cooling. So let’s say globally data centers will be dumping an extra 25 GW of heat into the ocean by 2020, if all those centers are moved to the ocean. I’d say it’s still insignificant, especially if you account for reduced radiative forcing due to the fossil fuels that now don’t have to be burned to generate the power for cooling. That ignores local effects of course, but still.

      1. The problem with TLAs and TLIs is that they mean something different to everyone.

        UAV := Unmanned Aerial Vehicle

        or is it

        UAV := Underwater Autonomous Vehicle

        or maybe it’s

        UAV := Urban Assault Vehicle


        UAV := University of Antelope Valley

        or even

        UAV := Unstoppable Allied Vanguards

  4. The fundamental flaw in the premise of the post is the implicit assumption that data centers are not already heating the oceans. Of course they are: the oceans absorb all of the excess heat from the atmosphere. As already was pointed out, with passive cooling you lose the inefficiencies of active cooling currently used in above ground data centers, so you are actually releasing less heat to the environment (and that is the only number that matters in the long run).

    Furthermore, many, if not most thermal generation stations use a natural water source as a coolant, so the power consumed by data centers already is contributing to the heating of water bodies at the point of generation.

    The only real concern is about local heating. Does creating a local hot-spot cause a problem? Probably, but since the vast majority of the ocean floor is more like a desert than it is like a lush garden, I would suggest that there is a large set of candidate sites where a local hot spot would not matter. Not matter to the ocean, that is. Create a local hot spot and I can guarantee one thing will come running (swimming/crawling?): sea life! Forget about bugs – they’re gonna have barnacles to worry about…

        1. No need to read it since it was already reported on slashdot, reddit, eweek, techdirt and host of other generic tech sites several days before it appeared here. And, yes, I took time to comment because I care about this site and I hate the drift towards becoming just another generic tech blog.

          Not everyone here agrees with us not-a-hack commenters. That’s cool. Feel free to scroll past my comments in the future.

      1. You don’t think putting a datacenter on the ocean floor is a hack? I do. Just because they fundamentally have far more resources to pull it off doesn’t mean it’s not a hack.

        1. This isn’t a hack. A hack is reusing something in a way it wasn’t intended or designed to be used. This is just a tech story. Also, this is a very costly idea. The amount of energy saved in cooling below water and cost of cooling above water are far exceeded by the cost of building and setting these things up underwater in a pressure tight vessel. It’s more costly than space travel. For data. Get real.

          1. So, by your definition of a hack, servers are intended or designed to be used on the bottom of bodies of water.

            Who care if it doesn’t fit an arbitrarily tight definition? Those that cry ‘not a hack’ in the comments pollute this site more way more than the accused articles Would you rather read an article about a generic mineral oil PC?

          2. Did you even read my comment? Also, the fact that “not a hack” appears so frequently is because HaD has changed gears on what pulls in ad revinue. Finally, mineral oil PC’s are not a hack. PC’s and server racks need cooling as part of their operation. Cooling a computer with oil or a server rack under the ocean isn’t a hack, it’s a solution to a necessary evil. Hacking is the process by which you change something in a way to benefit you, even though the object was never intended to be used that way. Hacking is when you unlock the FLIR E4 to use the full sensor instead of the intended low resolution. Hacking is when you tap into a car on the road and control it remotely by finding an exploit that gives you escalated privileges. Cooling a server with water isn’t interesting or new. Cooling a computer with mineral oil isn’t interesting or new. These aren’t hacked events. These are intended behaviors with an alternate solution to a problem that exists by design.

    1. I wonder if it would be worth it to apply this same idea to traditional steam-turbine power plants; use the heat from a data center to pre-heat the boiler water, gaining some of that heat back as work. You’d also reduce transmission losses by building your data center next to your power plant.

      1. I like it. Pre-warm the water before boiling. As Microsoft said, powerplants are already near water (as are much of the population of the world). The biggest saving isn’t cooling, but rather real estate. With cooling, they already noted that Google has a datacentre which uses external water as a heat sink. Don’t put the datacentre in the water, pump the water around the datacentre – much easier to maintain. I don’t see underwater datacentres working any time soon – there is plenty of land still available and can be bought as capital which will likely appreciate unlike investment in underwater vessels which will depreciate.

    2. Not a flaw. If, as in Microsoft’s home state, the power is from hydro or wind, there is no net difference in heat. The work done by the water is the same if it flows to the ocean or runs turbines on part of the trip. And if it is from hydrocarbon, it depends on the time scale. If it is fossil fuel, it is stored solar energy. If the hydrocarbon is from comets or other cosmic sources, then it is stored nuclear energy. Just change your integration interval and you will get zero eventually.

    3. If you co-locate the datacenter with a big OTEC (Ocean Thermal Energy Conversion) power plant, the datacenter will even *cool* the local surface water, and maybe even mitigate global heating. Also, the AI won’t have to worry about dependence on onshore power sources to power its defensive systems…

    1. The way these big data systems work is through redundancy and virtualization, each storage and computing unit is like the cells in your body, if one dies it does not matter so long as there are enough still operating for the overall system to function well enough. These machines have a life that is shorter than the mean time to failure anyway, they become technologically redundant before 50% of them have failed. So what do they do at it’s end of life? Sell their remaining functionality off as a second or third tier product, or if the link is needed for a newer unit they just unplug the old unit and leave it down there until there is enough in scrap value on the ocean floor to fund a mass recovery and recycle program.

  5. Am I the only one that sees the issues with servicing something like this? In Avagadro Corp the floating barges make sense, all the benefits of passive cooling, while still maintaining the ability to service or upgrade the systems down the road. Would it be feasible to pull these up from the ocean floor for maintenance, or would it be cheaper to just build another one and leave the old stuff to Davey jones? Seriously i’m asking, I have no idea what the lifespan of a commercial server is, or the cost for that matter.

    1. Maybe build base to connect the data centre to the cables and when you need to upgrade the data centre automaticly disconnect the data centre, inflate a balloon on top and it floats to the surface, then just refurb or replace as needed and use a UAV to reinstall.

    2. I don’t believe regular maintenance is part of their plans here, instead their probably banking on good parts and a certain degree of redundancy to keep one of those pods operating for some years or maybe even a decade, then replace the pod and have a team go over the old pod to see how it fared in order to determine what can be improved with the next model.

      However considering how fast technology is advancing I have to wonder if they’re really going to go with that pod design or if it could be only part of a larger offshore data center design in the works. Paul brought up OTEC and now I’m wondering if microsoft and other companies might not be heading towards that route with the OTEC system bringing up cooler water which is first used to generate power (which can be used to power the servers) before being passed on to the server pods to carry away their waste heat. Doing it that way would allow the company to more easily service and more importantly replace old parts with newer technology (not to mention making it easier to dispose/recycle of those pods without costly searches and salvaging).

    3. If you eliminate spinning hard disks (use solid state disks) and presumably there are no fans in this operation (forced water cooling), you essentially have no moving parts to fail. In that scenario, you should have very few failures within a 4-8 year life span of the hardware being deployed. Not zero, but very low.

      My educated guess is that they build a certain amount of redundancy into the capsules, drop them into the water, and will never go back to retrieve them, ever.

      1. You do not need to eliminate hard disks, just use good bearings and keep them spinning :-). Solid State disks have their failure mode too. But think about electrolytic caps. I have already had too many failing PSUs.

  6. One beef. “cold ocean temperatures” should be “low ocean temperatures” or “cold ocean water”. There is not such thing as cold temperature, no matter how often the TV weather announcer says it. What is the4 air temperature today? It is cold degrees. Like saying today we will have shorter lengths.

      1. To me, “Hot water heater” can be correct. The water that most people consider ‘cold’ coming into the water heater is quite warm, compared to absolute zero…. and it takes this ‘hot’ water and heats it up ‘hotter’. I guess ‘hot’ is a relative value, not absolute.

  7. Strikes me as overkill and doesn’t look like it produces enough advantages to offset the problems created, thus I suspect that cooling is only the public explanation for why this is being considered.

  8. You can probably have these self powered by utilising a combination wave/tidal power, convection and thermo-acoustic generators and heat-pumps. The waste heat drives the pulse engine that then vents water above the unit thereby drawing in the colder water in from around and below it. The wave power is from a, still very deep buoy, that rises and falls due to the pressure changes caused by waves above. It would have a huge amount of torque so a gearbox would be needed unless some sort of massive piezoelectric device was possible, in fact that would be preferable as it reduces the number of moving parts. The pressure can be managed by filling the unit with cooling fluid so that the skin has internal and external cooling cycles but no penetrations, even the power and data can be transmitted through titanium walls via magnetics (transformer effect). This arrangement makes the unit self powered, disaster resistant and less vulnerable to bit-flipping errors caused by cosmic rays etc. Remember it could be 5 km down and still 1 km off the ocean floor giving it a minimum of 1 km of water based radiation shielding. Only the data link need travel down to the anchor point and then across the bed of the ocean.

    As for the paranoid crap, I don’t care how smart it is if I can cut off all it’s actuators.

    1. Yes, they ‘can’ make it self-powered, but if wave/tidal power, convection, thermo-acoustic generators and heat-pumps were a cheaper way to generate electricity we would have fewer coal power plants.

      1. We don’t have any coal powered plants, at the bottom of the ocean, and the power cables to run these things start costing more than they do if they are in the deep trenches where they are safest and where the thermal differential is greatest. You do know what a thermal differential is, a gradient? It is the phenomena that makes all forms of thermal power station work, even coal ones.

        So thanks for your expert advice, LOL.

        1. so like, you’re in agreement with him, but you need to throw in an insult as well. Awesome. I love to insult people who know less than me.

          Your expert correction is even more laughable, LOL. It betrays your primitive understanding of thermodynamics, a total ignorance of oceanography, and a willingness to make unsubstantiated assumptions (dangerous and stupid by itself). I’ll go out on a limb and suggest you don’t know much about power plants or their economics either.

  9. I guess this makes sense if you’re in california and can’t find a cool place in the summer. Check some google datacentres in the EU and you’ll find watercooling but the water is cooled by evaporation after running it over some fancy fountains outside. The cost in water itself is probably pretty hefty but at the same time you keep the maintainability as it’s above ground and make a saving in energy by not having to run compressors and whatever else for a normal air conditioning system.

  10. “I can think of another big impact: property taxes.” Yeah, you just wait. As soon as our Socialist Big Government learns about your new undersea data center, they’ll figure out a way to tax your ass off. Then the EPA or some such will regulate the hell out of you too.

  11. Heating the oceans directly is not a good idea. There was a great 2-part show on NHK world that looked at the current state of the art knowledge about climate science as a global multi-physics system. The biggest worry was ultimately that the oceans, specifically the deep ocean, was rapidly heating. There were predictive models, conceptual experiments, and large scale models that all showed that hot oceans were major factors in the formation of super-hurricanes.

    Distributed computing units that provide useful heat energy is the way to go, “co-generation”. And here’s where the synergy can get really crazy: using big data, running on the huge compute clouds, we should be able to identify certain segments of roads and highways that suffer the most of freeze & ice effects, and install a datacenter there, and use the waste heat to just keep the road safe.

    1. OK, do the maths, 360 million square kilometres of ocean on earth, so lets just look at the equatorial 1/3 of that and we have about 100 million square km of ocean under sunlight for 10 hours a day (or more) at 1 KW per square meter, so how much energy per day is that ocean absorbing from sunlight? Now what percent of that is 1000 times the current eorld computing energy requirements (energy pushed through the actual chips, not cooling costs as we get that for free.)

      See what I mean, over heating the oceans is the least important issue! At worst you may get some localised convection driving a column of water to become more oxygen and nutrient rich thereby producing more phytoplankton which absorb CO2. i.e. Potentially a good thing.

      1. Right, an estimated 1 kilowatt/m^2 of solar energy penetrates the atmosphere, and since (according to google) the surface area of the earth is 5.101*10^12 m^2 (about 5 trillion square meters) the earth constantly receives about 250 petawatts of power. from the sun. (it only receives power on half the surface area at a time) . now its difficult to estimate how much of that is reflected back into space instead of warming the earth, but some sources say it is as high as 75%, which means that of the 250 petawatts, only about 62.5 petawatts actually go towards warming the earth. In addition, the earth also generates its own heat from internal sources (totaling about 47 terawatts according to wikipedia which is only). Now assuming the average server uses 500 watts and loses all of that energy as heat (which is an overestimation on both counts) and if we put one billion of them in the ocean, the total heat power that we would add would only be .5 terawatts, only 1 /125,000 of what is absorbed from the sun.

  12. How about the “save harbour” data laws (or the soon to be “Privacy Shield”) ? If the pod is dropped outside the 12 mile zone and in international water, to what country’s rule should the data protection adhere ? What if the pod has multiple interconnects to several exchanges and someone starts to host a torrent site there ?
    A big save would be the realestate, then power cost but by far the biggest would be to lay off some lawyers….

  13. Comparing these to space craft is ridiculous. Damn thing quits working you send the crane crane that placed back out to retrieve it. The heat the created by servers is going to added to the atmosphere. As ell as water co ducts heat I would expect localized heating that would extend very far from the equipment. one way or another. Refrigerant cooling would more heat.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.