Data centers are transitioning from AC to DC
spectrum.ieee.org259 points by jnord 19 hours ago
259 points by jnord 19 hours ago
I set up my own home network with a Vertiv Liebert Li-ion UPS a few years ago and was thinking about how inefficient the whole process is regarding power. The current goes from AC to DC back to AC back to DC. Straight from the UPS as DC would work much better, and as I was teaching myself more about networking equipment, I was surprised to learn that most of it isn't DC input by default (i.e., each piece of equipment tends to come with built-in AC-DC conversion).
Then I started routing ethernet with PoE throughout my house and observed that other than a few large appliances, the majority of powered devices in a typical home in 2026 could be supplied via PoE DC current as well! Lighting, laptops, small/medium televisions. The current PoE spec allows up to 100 W, which covers like 80% of the powered devices in most homes. I think it would make more sense to have fewer AC outlets around the modern house and many more terminals for PoE instead (maybe with a more robust connector than RJ45). I wonder what sort of energy efficiency improvements this would yield. No more power bricks all over the place either.
"... throughout my house and observed that other than a few large appliances, the majority of powered devices in a typical home in 2026 could be supplied via PoE DC current as well!"
We installed 120 LED ceiling lights in our home circa 2020, all of which were run with high voltage (romex) and accompanied by 120 little transformer boxes that mount inside the ceiling next to them.
Later ...
We installed outdoor lighting with low voltage, outdoor rated wiring and powered by a 12V transformer[1] and I felt the same way you did: why did we use a mile of romex and install all of those little mini transformers when we could have powered the same lights with 12V and low voltage wire ?
I then learned that the energy draw of running the low-volt transformer all the time - especially one large enough to supply an entire house of lighting - would more than cancel out energy savings from powering lower voltage fixtures.
You don't have this problem with outdoor lighting because the entire transformer is on a switch leg and is off most of the time.
So ... I like the idea of removing a lot of unnecessary high voltage wire but it's not as simple as "just put all of your lights behind a transformer".
[1] https://residential.vistapro.com/lex-cms/product/262396-es-s...
> I set up my own home network with a Vertiv Liebert Li-ion UPS a few years ago and was thinking about how inefficient the whole process is regarding power. The current goes from AC to DC back to AC back to DC.
With double-conversion, generally yes.
I recently ran across the (patented?) concept of a delta conversion/transformer UPS that seems to eliminate/reduce the inefficiencies:
* https://dc.mynetworkinsights.com/what-are-the-different-type...
* a bit technical: https://www.youtube.com/watch?v=nn_ydJemqCk
* Figures 6 to 8 [pdf]: https://www.totalpowersolutions.ie/wp-content/uploads/WP1-Di...
The double-conversion only occurs when there's a 'hiccup' from utility power, otherwise if power is clean the double-conversion is not done at all so the inefficiencies don't kick in.
I think we're slowly, slowly coming around to the idea of domestic DC distribution. The vast majority of consumer electronics would be perfectly happy to consume 12v. It's cheaper, safer, more efficient. Less design work and certification on inbuilt AC adapters.
I think it's highly unlikely we'll see mass scale retrofits, but if enough momentum builds up, I can see it as a great bonus feature for new builds.
I got lucky with my house and every room has a dedicated phone line meeting at a distribution panel (a couple of 2x4s with screw terminals) built in the 50s. I'm in the process of converting it to light duty DC power. The wiring is only good for an amp or two, but at 48v that's still significant power transmission.
800V to each rackmount unit, with hot plugging of rack units? That's scary. The usual setup at this voltage is that you throw a hulking big switch to cut the power, and that mechanically unlocks the cabinet. But that's not what these people have in mind. They want hot-plugging of individual rackmount units.
GE has a paper about the power conversion design, but it doesn't mention the unit to rack electrical and mechanical interface. Liteon is working on that, but the animation is rather vague.[2] They hint at hot plugging but hand-wave how the disconnects work. Delta offers a few more hints.[3] There's a complex hot-plugging control unit to avoid inrush currents on plug-in and arcing on disconnect. This requires active management of the switching silicon carbide MOSFETs.
There ought to be a mechanical disconnect behind this, so that when someone pulls out a rackmount unit, a shutter drops behind it to protect people from 800V. All these papers are kind of hand-wavey about how the electrical safety works.
Plus, all this is liquid-cooled, and that has to hot-plug, too.
[1] https://library.grid.gevernova.com/white-papers-case-studies...
[2] https://www.youtube.com/watch?v=CQOreYMhe-M&
[3] https://filecenter.deltaww.com/Products/download/2510/202510...
The power connectors will be on the far side of the rack from the service side so shouldn’t be a problem for humans touching the third rail so to speak.
With that sort of voltage you should be able to use a capacitive or inductive sensor to activate a relay.
It is a pretty clever design
> When it is detected that the PDB starts to detach from the interface, the hot-swap controller quickly turns off the MOSFET to block the discharge path from Cin to the system. After the main power path is completely disconnected, the interface is physically detached, and no current flows at this time
> For insertion, long pins (typically for ground and control signals) make contact first to establish a stable reference and enable pre-insertion checks, while short pins (for power or sensitive signals) connect later once conditions are safe; during removal, the sequence is reversed, with short pins disconnecting first to minimize interference.
Yes, I read that. There had better be a backup system. MOSFETs tend to fail ON, and there's a megawatt going into each rack.
Somehow this seems the wrong approach to AI.
Fail spectacularly
Data center workers are gonna need those big yoink sticks and those thick arc-fault bibs that furnace operators wear.
McMaster-Carr has the personal protective equipment required for 800V: Flash protection face shields [1] and Class 0 electrical protection gloves.[2]
It's not that bad. It's just ordinary industrial protective gear.
[1] https://www.mcmaster.com/products/arc-flash-protection-face-...
[2] https://www.mcmaster.com/products/electrical-protection-glov...
An EV fast charger can do 1000V, so with a bit of logic that sounds doable.
EV chargers take a different approach. There is no power on the connector while you're plugging it in. It then locks in place before the contactor closes and power is delivered. Unplugging is the same, power is removed before the plug is unlocked for release.
Or maybe you can require technicians to be in full-protection electrical suits.
As long as you can control for fire, electrical safety seems like a temporary condition as robots and intelligent machines are cheaper and more available long term solution to hot swap blades in datacenter racks.
I think you're being downvoted for speaking of a complex future possibility ("robots and intelligent machines ... solution") as if it was a proven commodity. There will be many twists and turns in the path to the possible reliability, scalability, and cost effectiveness of robots and intelligent machines.
It is a proven commodity already. Just not in our behind-the-curve country.
Look at NTT Data or SoftBank.
Yea I literally said long term but HN voters have such a hate boner for anything adjacent to AI
DC power has been an option for datacenter equipment since I was a young lad racking and stacking hardware. Cisco, Dell, HPE, IBM, and countless others all had DC supply options. Same with PDUs. What’s old is new again.
See e.g. https://www.dell.com/support/kbdoc/en-us/000221234/wiring-in...
48vdc was common in phone exchanges. They filled the basement with lead-acid batteries and to could run without the grid for a couple weeks. In turn the phone was 99.999% reliable for decades.
I'm working on stuff in that market, it's still largely is. DC Power System Design For Telecommunications is still a must read and it doesn't even cover the last 15 years or so of development, notably lithium batteries and high efficiency rectifiers.
I will say that this is a surprisingly deep and complex domain. The amount of flexibility, variety and scalability you see in DC architectures is mind-boogling. They can span from a 3kW system that fits in 2U all the way to multiples of 100kWs that span entire buildings and be powered through any combination of grid, solar and/or gas.
Not to be _that_ guy, but it was technically -48V DC.
Honestly, that was pretty surprising to me when I had to work with some telco equipment a couple of decades ago. To this day, I don't think I've encountered anything else that requires negative voltage relative to ground.
I am STILL designing hardware for -48v telco standard. The first thing we do is convert -48 -> 48v. That's 4 square inches of PCB space we waste.
What do you need +48V for?
We go -48 -> 48 -> 12 -> 3v3,1v8 etc etc. If you went 48 straight to POL voltages then you would have horrific converter performance.
> If you went 48 straight to POL voltages then you would have horrific converter performance.
What's horrific converter performance in numbers?
An isolated flyback (to 12V) should be able to hit >92% and doesn't care if it's fed -48V or +48V or ±24V. TI webench gives me 95% though I'd only believe that if I'd built and measured it. What's the performance of your -48V → +48V?
[with the caveat that these frequently require custom transformers... not an issue with large runs, but finding something that can be done with an existing part for smaller runs is... meh]
-48 to 48 claims something like 97% (load dependent of course). It also needs to arbitrate between two input supplies for glitchless redundancy, plus have PM bus and other spec mandated stuff. There is no technical reason why you cant go -48 -> 12 as you state with good efficiceny, but we cant get hold of a part that ticks all the boxes.
Horrific performance by my definition would be 48v to say 1v. We only realistically use buck topologies for POL supplies. Such a ratio is really bad for current transients, not to mention issues like minimum on times for the controller.
I'm just surprised that either input isolation isn't on your spec, or it still somehow works out better with isolated to +48V than straight to 12V... but I guess if your spec requires other things, it makes sense.
(Thanks for the info!)
Well if it's negative 48V the electricty flows out of your circuit and back to the grid, so you need to make it positive to have the electricity come in.
> I don't think I've encountered anything else that requires negative voltage relative to ground
Automotive collectors can probably still relate to cars from the 1920s-50s having a "positive ground."
With positive ground the traditional more-sacrificial spark plugs lasted longer.
Yes, and that tiny little difference can cost you a lot of expensive gear if you run it off the battery and plug in a serial port or something like that. You'll also learn first hand what arc welding looks like without welding glass.
Some old guitar effects used -9V DC.[1] And the convention with guitar effects power adapter is the barrel is center negative (which is motivated with facilitating easy wiring of the socket's switch to connect to a 9V battery inside).
[1] https://www.analogisnotdead.com/article26/what-is-going-on-w...
Can you explain why it's -48 VDC as opposed to 48 VDC with the + and - inputs mislabeled?
Because the chassis is connected to ground (as in, a literal grounding rod hammered into the soil) and by definition your 0V reference point.
The crucial difference is the direction in which the current is flowing: is it going "in to", or "out of" a hot wire? This becomes rather important when those wires are leaving the building and are buried underground for miles, where they will inevitably develop minor faults.
With +48V corrosion will attack all those individual telephone wires, which will rapidly become a huge maintenance nightmare as you have to chase the precise location of each, dig it up, and patch it.
With -48V corrosion will attack the grounding rod at your exchange. Still not ideal, but monitoring it isn't too bad and replacing a corroded grounding rod isn't that difficult. Telephone wires will still develop minor faults, but it'll just cause some additional load rather than inevitably corroding away.
So, there is a true value for 0?
Does that mean when you have electronics and use multiple dc-dc converters all the inputs and outputs share the same ground, it's not just the values for that pair of wires?
And if I want to use a telephone on an incorrectly wired 48dc circuit, I could switch the positive and negative wires, as long as the circuit in the telephone is isolated and never touches ground?
Thanks. Somehow I got in my head that all circuits were just about the delta from neutral and therefore nothing outside them mattered.
There is a true zero potential. You can detect this because two charged objects with zero delta between them will still repel each other.
I think a circuit should mostly care about the deltas, but when you’re talking about things like phone lines, the earth becomes part of your circuit. You can’t influence its potential (it’s almost exactly neutral because any charge imbalance gets removed by interaction with the interplanetary medium) so everything else is going to end up being determined by what you need for their relative potential to that.
Do you also happen to why this is not more common? Must be useful for more than just telephone wires.
Most large scale systems are AC because transformers are relatively cheap, low maintenance, and efficient. When the system is AC ground makes no difference.
With DC systems you generally think about the issues - which is why modern cars are negative ground. However other than cars most people never encounter power systems of any size - inside a computer the voltages and distances are usually small enough that it doesn't matter what ground is. Not to mention most computers don't even have a chassis ground plane (there are circuit board ground planes but they conceptually different), and with non-conductive (plastic) cases ground doesn't even make sense.
> When the system is AC ground makes no difference.
With AC it's about where the ground is attached along the length of the transformer secondary. In the EU they ground one of the ends of the secondary, in the US we ground the center point.
I don't get to say this very often ... but the US way is objectively safer with no downside: 99% of human shocks are via ground, and it halves the voltage to ground (120V vs 240V). A neutral isn't required if there aren't 120V loads.
I agree that the US voltage is safer (with the tradeoff of lower output powers available at your outlets). However, I suspect this is more than negated by the US plug design, which carries a much larger risk of shocks than almost all EU plug designs (Schuko, British/Type G, etc...)
- uninsulated metal pins make contact with supply while partially exposed - much smaller distance between metal pins and the edge of the plug