I know this is not a great application, but the Brayton cycle really is quite amazing from an engineering perspective. The amount of power we can generate per unit of volume or mass is incredible. That's why they keep using these damn things. To get a sense of scale, the new GE helicopter engine can produce 3000 shaft horsepower (~2.2 MW) and it only weighs around 200kg. A commercial truck in the US can haul 20,000kg worth of mass before requiring a babysitter.
Huge 2-stroke Diesel engines designed for ships would be much more efficient for electric generators than jet engines, but they also would need much more space.
Nonetheless, I doubt that the xAI datacenter had such footprint constraints that it really needed the smaller jet engines. Perhaps the time-to-delivery for really big Diesel generators was too long, so they opted for the already available repurposed jet engines.
The efficiency of the generators with jet engines could be increased a lot if the heat from the exhaust gases would be recovered and used to generate extra electric power, with a steam turbine or with another kind of turbine with Rankine cycle, e.g. with organic vapor.
> efficiency of the generators with jet engines could be increased a lot if the heat from the exhaust gases would be recovered and used to generate extra electric power, with a steam turbine
I'm guessing most gas power plants use the combined cycle, otherwise you'd be wasting the energy of the exhaust gases. That uses a steam turbine heated by exhaust gases from the gas turbine.
https://en.wikipedia.org/wiki/Combined-cycle_power_plant
Yeah, the combined cycle gas turbines are the most efficient known method of generating power. That’s been proven for a long time.
But they do have certain drawbacks, and may not be the most EFFECTIVE for a given application.
You do have to look at a long list of constraints and use cases.
And sometimes, time to market is the most critical factor so you take whatever you can get your hands on. Which xAI has repeatedly proven that they will do, regardless of the environmental consequences.
Are you sure about this? AFAIK turbine generators are more efficient than diesel engines at large diameter and high RPM, for a given amount of power, and the tradeoff is usually that ships/vehicles require lower shaft speeds.
I remember doing the math for small power generators and the math didn't work out for small turbines until the diameter of turbine went up in size, before which a diesel engine was better. But when scaled up, turbine generators were rapidly making up and surpassing diesel generators.
Makes it pretty clear generation capacity was never a real argument against EV adoption. We'll recycle rusty jet engines to get it done when there's real money on the line.
Utility monopolies have been raising rates, and profits, for a long while now, to the point where making you own power is cheaper. On average, 15 cents per dollar directly into profit [1].
And, ironically, that means self generation of power (solar, maybe wind if you have lots of land) becomes even more attractive. Even if it's only a portion of your usage.
Nah, just attack the transformers. All large transformers worldwide are on a multi-year backlog, and almost all of them are made overseas. Many in countries that aren’t so friendly right now.
If you get a large transformer that didn’t make you wait multiple years, then you stole it, or bought it out from underneath someone else. Or, it was damaged somehow and has supposedly been rebuilt perhaps with questionable materials and reliability.
This is off-topic and I'm no fan of xAI or any company circumventing rules to allow them to use dirty energy sources.
That said.
> Normally, that would be enough to keep around 15,000 homes up and running.
Am I the only one who finds this kind of comparison meaningless? Like it's supposed invoke a feeling of "15K homes worth of power! How absurd that a data center use that much!" but that kind of how all centralized things work. That datacenter does work for, one would assume, way more than just 15K homes and if those homes wanted the same "work" output they'd use the same, or most likely, more energy to accomplish the same thing.
In many/most cases they aren't using electricity just to use it, they are using it because people are paying them for the output. With economies of scale I would guess a datacenter is many times more efficient than if we were all running local models at home.
Again, not an argument for data centers as much as I find this comparison always pointless.
It's like using bathtubs as a unit of measure for large volumes of water. Yes it communicates the scale, but abjectly fails to communicate the relative in-cohort comparison i.e. is this a lot of power for an industrial user? Is it wasting power /generating more emissions relative to other similar applications or not?
Industrial scale is difficult for most people to understand, and therefore easy to use to generate outrage. I used to work at a pulp mill, one that would be fairly small by American standards. Some of the numbers were nonetheless crazy by domestic comparisons. We used 90 million litres of water per day. That one sticks with me.
Mainstream news is pitched at an audience that hasn't got a clue what is a joule or a kilogram. You must've seen Sam Altman make the efficiency argument for data centers. But it deliberately bypasses the point, which is that data centers and frontier AI companies are systemically important and must be supported for the good of the nation.
I mean, they need to give an intuition of the magnitude. Tell me, what do you think when I say “a GPU consumes 1 kW of power”? What does that even mean to the regular Joe?
It is pretty easy to compare against homes because they consume on average 1 kWh of electricity (in the U.S.). It could be something else, but 99% of people have homes they live in, so this is something directly linked to their experiences. So it is not meaningless at all.
GPUs have power ratings: RTX 5090 575W, PRO 6000 Blackwell Max-Q 300W. Datacenter GPUs are more efficient still--one B200 is 1000W. Use during training consumes more power than inference.
As always, remind yourself: "the outrage is the product". The entire point of slotting something like that in is to get eyeballs and produce outrage on tap. It's the new dopamine. Justified or not is irrelevant, what matters is to have a talking point you can use to rile things up, no matter how pointless it is.
Do you think the author realizes that gas turbine makers already sell aeroderivative turbines because the core power plant of a gas turbine generator and a jet engine are very similar?
Pretty clever to use the engines to generate power at scale - as a plus, if there are any efficiency gains (whether in power production or pollution terms) they may make their way back to the aerospace industry eventually. Complaining about carbon monoxide production and formaldehyde in these things suggests to me that the cons aren’t as serious as the article suggests, as carbon monoxide is produced by every combustion engine on the planet and formaldehyde remains used in a great number of things. If that’s the worst outputs from these engines then I’m sure there are many far more harmful methods of generating power.
It is not clever. Jet engines have already been used as "peaker plants" since the sixties. They can ramp up quickly but have much higher cost per MWh than other power sources. They make sense to deal with high power demand for a short time, and to keep around for their black start capability. But not to run 24/7.
I'm pretty sure they don't plan on powering these datacenters in this manner indefinitely. Everyone on the pro-power-gen side wants proper build outs. But America needs a cultural shift before it regains the ability to build efficient, reliable and green power generation capacity at scale and at speed. Right now it's busy worrying about fake-news like water consumption and NIMBYism while China takes over.
I know this is not a great application, but the Brayton cycle really is quite amazing from an engineering perspective. The amount of power we can generate per unit of volume or mass is incredible. That's why they keep using these damn things. To get a sense of scale, the new GE helicopter engine can produce 3000 shaft horsepower (~2.2 MW) and it only weighs around 200kg. A commercial truck in the US can haul 20,000kg worth of mass before requiring a babysitter.
Huge 2-stroke Diesel engines designed for ships would be much more efficient for electric generators than jet engines, but they also would need much more space.
Nonetheless, I doubt that the xAI datacenter had such footprint constraints that it really needed the smaller jet engines. Perhaps the time-to-delivery for really big Diesel generators was too long, so they opted for the already available repurposed jet engines.
The efficiency of the generators with jet engines could be increased a lot if the heat from the exhaust gases would be recovered and used to generate extra electric power, with a steam turbine or with another kind of turbine with Rankine cycle, e.g. with organic vapor.
> efficiency of the generators with jet engines could be increased a lot if the heat from the exhaust gases would be recovered and used to generate extra electric power, with a steam turbine
I'm guessing most gas power plants use the combined cycle, otherwise you'd be wasting the energy of the exhaust gases. That uses a steam turbine heated by exhaust gases from the gas turbine. https://en.wikipedia.org/wiki/Combined-cycle_power_plant
Yeah, the combined cycle gas turbines are the most efficient known method of generating power. That’s been proven for a long time.
But they do have certain drawbacks, and may not be the most EFFECTIVE for a given application.
You do have to look at a long list of constraints and use cases.
And sometimes, time to market is the most critical factor so you take whatever you can get your hands on. Which xAI has repeatedly proven that they will do, regardless of the environmental consequences.
Are you sure about this? AFAIK turbine generators are more efficient than diesel engines at large diameter and high RPM, for a given amount of power, and the tradeoff is usually that ships/vehicles require lower shaft speeds.
I remember doing the math for small power generators and the math didn't work out for small turbines until the diameter of turbine went up in size, before which a diesel engine was better. But when scaled up, turbine generators were rapidly making up and surpassing diesel generators.
Makes it pretty clear generation capacity was never a real argument against EV adoption. We'll recycle rusty jet engines to get it done when there's real money on the line.
Utility monopolies have been raising rates, and profits, for a long while now, to the point where making you own power is cheaper. On average, 15 cents per dollar directly into profit [1].
[1] https://energyandpolicy.org/wp-content/uploads/2026/03/Payin...
And, ironically, that means self generation of power (solar, maybe wind if you have lots of land) becomes even more attractive. Even if it's only a portion of your usage.
With that portion being somewhat compensated for by the rising "connection fees", being over $150/month where I am.
The difference is this doesn't tie into the grid.
Good thing jet engines are famously quiet. I'm sure that will get all the people protesting against data centers on side.
These seem like soft targets for an adversary, whether that be a competitor, a goverment, or non-goverment organization.
I wonder how well a LM6000 would handle multiple drone ingestions.
Likely the PE6000 in this case...
They need to get through a lot of filtration and potentially evap coolers, or that stuff could just become FOD.
Nah, just attack the transformers. All large transformers worldwide are on a multi-year backlog, and almost all of them are made overseas. Many in countries that aren’t so friendly right now.
If you get a large transformer that didn’t make you wait multiple years, then you stole it, or bought it out from underneath someone else. Or, it was damaged somehow and has supposedly been rebuilt perhaps with questionable materials and reliability.
And transformers are surprisingly easy to damage.
This is off-topic and I'm no fan of xAI or any company circumventing rules to allow them to use dirty energy sources.
That said.
> Normally, that would be enough to keep around 15,000 homes up and running.
Am I the only one who finds this kind of comparison meaningless? Like it's supposed invoke a feeling of "15K homes worth of power! How absurd that a data center use that much!" but that kind of how all centralized things work. That datacenter does work for, one would assume, way more than just 15K homes and if those homes wanted the same "work" output they'd use the same, or most likely, more energy to accomplish the same thing.
In many/most cases they aren't using electricity just to use it, they are using it because people are paying them for the output. With economies of scale I would guess a datacenter is many times more efficient than if we were all running local models at home.
Again, not an argument for data centers as much as I find this comparison always pointless.
It's like using bathtubs as a unit of measure for large volumes of water. Yes it communicates the scale, but abjectly fails to communicate the relative in-cohort comparison i.e. is this a lot of power for an industrial user? Is it wasting power /generating more emissions relative to other similar applications or not?
Agree. Comparing it to other industrial users (X car factories) might be more useful.
Industrial scale is difficult for most people to understand, and therefore easy to use to generate outrage. I used to work at a pulp mill, one that would be fairly small by American standards. Some of the numbers were nonetheless crazy by domestic comparisons. We used 90 million litres of water per day. That one sticks with me.
Mainstream news is pitched at an audience that hasn't got a clue what is a joule or a kilogram. You must've seen Sam Altman make the efficiency argument for data centers. But it deliberately bypasses the point, which is that data centers and frontier AI companies are systemically important and must be supported for the good of the nation.
I mean, they need to give an intuition of the magnitude. Tell me, what do you think when I say “a GPU consumes 1 kW of power”? What does that even mean to the regular Joe?
It is pretty easy to compare against homes because they consume on average 1 kWh of electricity (in the U.S.). It could be something else, but 99% of people have homes they live in, so this is something directly linked to their experiences. So it is not meaningless at all.
GPUs have power ratings: RTX 5090 575W, PRO 6000 Blackwell Max-Q 300W. Datacenter GPUs are more efficient still--one B200 is 1000W. Use during training consumes more power than inference.
As always, remind yourself: "the outrage is the product". The entire point of slotting something like that in is to get eyeballs and produce outrage on tap. It's the new dopamine. Justified or not is irrelevant, what matters is to have a talking point you can use to rile things up, no matter how pointless it is.
Do you think the author realizes that gas turbine makers already sell aeroderivative turbines because the core power plant of a gas turbine generator and a jet engine are very similar?
Pretty clever to use the engines to generate power at scale - as a plus, if there are any efficiency gains (whether in power production or pollution terms) they may make their way back to the aerospace industry eventually. Complaining about carbon monoxide production and formaldehyde in these things suggests to me that the cons aren’t as serious as the article suggests, as carbon monoxide is produced by every combustion engine on the planet and formaldehyde remains used in a great number of things. If that’s the worst outputs from these engines then I’m sure there are many far more harmful methods of generating power.
It is not clever. Jet engines have already been used as "peaker plants" since the sixties. They can ramp up quickly but have much higher cost per MWh than other power sources. They make sense to deal with high power demand for a short time, and to keep around for their black start capability. But not to run 24/7.
I'm pretty sure they don't plan on powering these datacenters in this manner indefinitely. Everyone on the pro-power-gen side wants proper build outs. But America needs a cultural shift before it regains the ability to build efficient, reliable and green power generation capacity at scale and at speed. Right now it's busy worrying about fake-news like water consumption and NIMBYism while China takes over.
Related NYT piece: https://www.nytimes.com/2026/11/03/climate/jet-engines-ai-da... (https://news.ycombinator.com/item?id=50352584)