My city's power company itself has multiple peaking power generators in shipping containers, that it runs because there's isn't enough supply.
Compared to full-size natural gas powerplants they're extremely awful, but they're a necessity when we otherwise wouldn't have reliable power. If it wasn't for NINBYism, we'd have plenty of hydro and nuclear power to meet all of our needs, but with NINBYism the only option is to build smaller, which is also dirtier and less efficient, but avoids the NIMBY blockades.
Hey now, give the environmentalists some credit too.
Even if the NIMBYs evaporated overnight these projects would still be hobbled by all manner of "you're gonna need an engineered stormwater solution" and "gonna need an impact study" type crap wherein firms and trades that manage to market what they do as being good for the environment (but at what cost?) are empowered sometimes by law, but more often by bureaucratic policy to leech large sums out of money at every step of the process.
Ideally these engines should be deployed near the NIMBY’s houses so that they can appreciate the symphony and aroma they produce. If it’s a problem, offer them the possibility to bid a certain amount (either in terms of current money or future “property value”) to replace them with quieter, more efficient alternatives if the music/fragrance isn’t up to their taste.
Modern aero derivative power turbines are capable of operating at below 15 ppm NOx when properly fitted for use. It's not as if they aren't making any attempt at all with regard to emissions. They've been working on this stuff since the 90s. They're on 2nd and 3rd generation combustor technology now.
That said, a parking lot full of these things could eventually crank up the ambient NOx levels. The real problem is low stack height. The exhaust being close to the ground creates a hot spot that would have otherwise dispersed into a much larger volume of air. One generator running by itself is probably not an issue in any arrangement. 20 is a medium sized power plant.
the reason it sells it that some non small part of the investor marked expects there is a "explosion" in compute needs, cause by AI adoption
which will lead to an explosion of electrical need
except there is limited world wide production capacity for gas power plants and other power plants, and some (e.g. nuclear) are very slow to build
and Trump put a lot of extra artificial hurdles for expanding anything "renewable"
so basically if that happen we most likely will have
- exploding electrical bills, and if no intervention is done for private homes, too.
- air pollution (and other environmental pollution) which doesn't just give people asthma implicitly will kill them (like them ding years earlier, but also potentially through damage from asthma attacks etc.)[1]
- price explosion of any electric components which share production lines with GPU servers (which isn't just GPUs, but also RAM, CPUs, etc.), the predicted price increase of RAM by ~30% is partially due to people anticipating this and already buying capacities :/
Naturally there needs to be a sinner to put the blame one so expect claims that electric price increases are fully at fault of Wind Farms and similar :/
Oh also if the AI bubble goes pop it sadly will not just be all fine, because we now have a double bubble of the core AI bubble and the "speculative new data center investment" bubble. The first will majorly hit the top of SAP100 and with it a lot of founds and similar which try to be "stable"/"reliable" etc. The other is more fulled by private equity and it's (for me) very unclear what is linked to it. One way or another things look pretty bad.
[1]: If anyone is wondering if there will be interventions from the state weather it's the OP politico article or long term promulgated and reported issues with fracking causing mutated fishes and major increases in cancer down stream, or the long history with toxic wast dumping in the US the answer is, most likely, no. And that sadly isn't even Trump specific.
Makes a lot of sense as described, basically recycling using turbines retired for flight, but I wonder how efficient these retrofitted gas turbine generators are compared to turbines originally designed and built for this fuel and application.
Edit: Turns out GE makes these things called LM2500 gas turbines and they're pretty much the same thing, CF6 aero turbines set up for power generation. They're advertised as being 35-39% efficient, which makes them rather on the less efficient end of the gas turbine spectrum. But they're light and quick to start so they have seem a lot of use in ships and some peaker plants. They definitely aren't the best turbine to be running a data center with long term, but if they can make them cheap and available that might be good enough. Not a new idea though.
The LM6000 and its variants have been in operation since probably the 1980s. I can ask around at work. I used to develop the code for the LM6000.
You're spot on that people use them for peaking, but it's a big mix. Peaking, mid-merit, and sometimes base load. There are low emissions versions as well, that keep NOx to a handful of ppm without using extra water.
I bet the thermodynamic efficiency of an aircraft jet power plant would be far less than that of purpose built turbine.
An aviation jet is designed to efficiently generate thrust at minimal weight. A power plant turbine is designed to efficiently turn the generator, and weight is no object.
These things are engineering marvels as fuel-burning engines go. From what I know, there is actually not that much that can be done to make gas turbines more efficient if weight is no object. Fuel weight and cost are very important for commercial turbofans, and engine manufacturers have been doing a huge amount of R&D in efficiency and are using some very advanced manufacturing techniques for certain parts. Turbine inlet temperatures are well into the white-hot range. A mechanical engineer described it to me as "A hundred degrees hotter and they'd melt and spill their internals out the back" - that is with exotic metals and active cooling. The main problem with these turbines is that they are somewhat old and will probably run at less than maximum power for longevity reasons. Turbines are terrible at less than full power efficiency. They also won't have a steam turbine added to use the exhaust heat like with stationary engines.
According to this article, the best current commercial turbofans can reach over 50% thermodynamic efficiency. I remember thinking that 50% seem like an unreachable holy grail when first reading about thermodynamic efficiency of common engines.
https://nap.nationalacademies.org/read/23490/chapter/6#36
It does say they replace the power turbine part with one designed to extract shaft horsepower. Given that the hot end is the difficult and expensive part that probably has loads of useful life left long after it's aged out of being usable in an aircraft, sticking a new power turbine on isn't the worst idea you could have.
Of course "jet" aircraft are mostly using the engine to produce shaft power to run a ducted fan anyway.
True, the turbines they're retrofitting are cores from high bypass turbofans. As far as I've managed to figure, they output somewhere in the neighborhood of 75% of their power through the shaft when in a turbofan engine.
However, these cores are designed to spin very fast, not at grid speed, so these need some form of gearbox or electronic frequency conversion.
This isn't exactly novel. I was recently visiting a gas peaker plant in the EU, built 10-15 years ago. It uses two jet turbines as found on a 747 (didn't ask for the exact type).
I have a weird question. Wouldn't it be simpler just like not run AI when there is no green power available? Surely you could save running tasks on disk and then wake them up again when there is green power available?
You could, but an AI datacenter is awfully expensive, and every minute down costs a ton in unused capital so generally they'll do anything in their power to guarantee maximum utilization at all times
Gotta love this... The companies running these (Amazon, Google) push hard on the narrative of how green they are, but their executives fly private everywhere and their data centers run on gas jet engines.
if all cars where replaced with EVs power usage would go up ~20-50% (based on full usage, i.e. all cars including trackers, trucks but also stuff like gasoline power generators). Number is such a large range due to unclearity for total average full usage efficiency across all cars, unclearity how much fast and slow charging would be used etc.
but cars normally get replaces with EVs only if you need a new car anyway, so ignoring that different classes of cars are used for different number of years etc. we are speaking about something like (very roughly) 5% or so off the total increase coming per year if no one including truckers would buy non-EVs. So we have a upper ceiling of around 2.5% increase per year, but more realistically we are speaking about less then 1% increase per year (but also that much for the next 20-50 years).
And sure there are other things, like electric heating, full electric cooking etc. which I am fully ignoring.
But for AI needs alone we have a projected increases of 25% in the next 5 years or so, depending on who you ask. And even around 80% until 2050... which is a completely different scale then like 5% in the next 5 years and up to 50% in the next 50 years ...
Readit News
https://www.politico.com/news/2025/05/06/elon-musk-xai-memph...
Compared to full-size natural gas powerplants they're extremely awful, but they're a necessity when we otherwise wouldn't have reliable power. If it wasn't for NINBYism, we'd have plenty of hydro and nuclear power to meet all of our needs, but with NINBYism the only option is to build smaller, which is also dirtier and less efficient, but avoids the NIMBY blockades.
Hey now, give the environmentalists some credit too.
Even if the NIMBYs evaporated overnight these projects would still be hobbled by all manner of "you're gonna need an engineered stormwater solution" and "gonna need an impact study" type crap wherein firms and trades that manage to market what they do as being good for the environment (but at what cost?) are empowered sometimes by law, but more often by bureaucratic policy to leech large sums out of money at every step of the process.
That said, a parking lot full of these things could eventually crank up the ambient NOx levels. The real problem is low stack height. The exhaust being close to the ground creates a hot spot that would have otherwise dispersed into a much larger volume of air. One generator running by itself is probably not an issue in any arrangement. 20 is a medium sized power plant.
the reason it sells it that some non small part of the investor marked expects there is a "explosion" in compute needs, cause by AI adoption
which will lead to an explosion of electrical need
except there is limited world wide production capacity for gas power plants and other power plants, and some (e.g. nuclear) are very slow to build
and Trump put a lot of extra artificial hurdles for expanding anything "renewable"
so basically if that happen we most likely will have
- exploding electrical bills, and if no intervention is done for private homes, too.
- air pollution (and other environmental pollution) which doesn't just give people asthma implicitly will kill them (like them ding years earlier, but also potentially through damage from asthma attacks etc.)[1]
- price explosion of any electric components which share production lines with GPU servers (which isn't just GPUs, but also RAM, CPUs, etc.), the predicted price increase of RAM by ~30% is partially due to people anticipating this and already buying capacities :/
Naturally there needs to be a sinner to put the blame one so expect claims that electric price increases are fully at fault of Wind Farms and similar :/
Oh also if the AI bubble goes pop it sadly will not just be all fine, because we now have a double bubble of the core AI bubble and the "speculative new data center investment" bubble. The first will majorly hit the top of SAP100 and with it a lot of founds and similar which try to be "stable"/"reliable" etc. The other is more fulled by private equity and it's (for me) very unclear what is linked to it. One way or another things look pretty bad.
[1]: If anyone is wondering if there will be interventions from the state weather it's the OP politico article or long term promulgated and reported issues with fracking causing mutated fishes and major increases in cancer down stream, or the long history with toxic wast dumping in the US the answer is, most likely, no. And that sadly isn't even Trump specific.
Edit: Turns out GE makes these things called LM2500 gas turbines and they're pretty much the same thing, CF6 aero turbines set up for power generation. They're advertised as being 35-39% efficient, which makes them rather on the less efficient end of the gas turbine spectrum. But they're light and quick to start so they have seem a lot of use in ships and some peaker plants. They definitely aren't the best turbine to be running a data center with long term, but if they can make them cheap and available that might be good enough. Not a new idea though.
The LM6000 and its variants have been in operation since probably the 1980s. I can ask around at work. I used to develop the code for the LM6000.
You're spot on that people use them for peaking, but it's a big mix. Peaking, mid-merit, and sometimes base load. There are low emissions versions as well, that keep NOx to a handful of ppm without using extra water.
An aviation jet is designed to efficiently generate thrust at minimal weight. A power plant turbine is designed to efficiently turn the generator, and weight is no object.
Different trade-offs.
According to this article, the best current commercial turbofans can reach over 50% thermodynamic efficiency. I remember thinking that 50% seem like an unreachable holy grail when first reading about thermodynamic efficiency of common engines. https://nap.nationalacademies.org/read/23490/chapter/6#36
...more or less. There have been outliers.
Of course "jet" aircraft are mostly using the engine to produce shaft power to run a ducted fan anyway.
However, these cores are designed to spin very fast, not at grid speed, so these need some form of gearbox or electronic frequency conversion.
Deleted Comment
but cars normally get replaces with EVs only if you need a new car anyway, so ignoring that different classes of cars are used for different number of years etc. we are speaking about something like (very roughly) 5% or so off the total increase coming per year if no one including truckers would buy non-EVs. So we have a upper ceiling of around 2.5% increase per year, but more realistically we are speaking about less then 1% increase per year (but also that much for the next 20-50 years).
And sure there are other things, like electric heating, full electric cooking etc. which I am fully ignoring.
But for AI needs alone we have a projected increases of 25% in the next 5 years or so, depending on who you ask. And even around 80% until 2050... which is a completely different scale then like 5% in the next 5 years and up to 50% in the next 50 years ...