AI in Linux
drewdevault.com
drewdevault.com
I don't see evidence this is true, and the critical mass of data centers is making electricity cheaper. Economies of scale -> more energy = bigger electric plants = cheaper energy.
https://netchoice.org/new-study-confirms-what-weve-been-sayi...
Problem is; power plants are not something that can be spun up with a kubectl invocation.
Power plants need to be built. This process takes YEARS, and in some cases over a decade. And that's just the plants themselves. The grid also has to be built out to be able to handle the new load.
And while this potentially +10year process is ongoing, the current supply is gobbled up, and prices for consumers skyrocket.
Yes, eventuelly, market forces may solve the issue. By then however, it will already be too late for many areas, with residents being priced out of a basic utility.
Or worse, the markets may not solve the issue at all.
Because; all this building out of electrical gear and power plants, requires very specific resources and talent. Whta if the markets decide that these resources are much better allocated (at least from the PoV of company shareholders), in building power generation capacity exclusively for datacenters, and the consumer market is simply ignored?
Yes and... it's still a projected ~50/50 split of future datacenter energy growth being met by renewables or coal (China) + natural gas (US). https://www.iea.org/reports/energy-and-ai/energy-supply-for-...
This isn't true. Certainly not globally. Because electricity in 2026 is converting finite and scares resources. Plants need fuel to burn, water to cool, material to build, land to sit on, etc. Yes, even Nuclear plants¹ need increasingly more scarce resources like "coolant".
What has been bringing electricity prices down is renewables: insane amounts of solar and wind being installed hourly. But even with that, we are still lagging, globally: energy prices are -on average- rising, projects halted or stalled because of lack of electrical capacity and ever increasing amounts of CO2 pumped into the atmosphere. Not less. Not cheaper. Not more supply.
¹ https://www.france24.com/en/france/20260712-france-temporari...
We should be cognizant of the environmental impact of producing more energy, but it still remains true that making more energy means cheaper energy across the resources we have today.
Energy prices, and electricity prices are rising, indexed for inflation. For households and for commerce.
One does not need to look very far.
https://www.theguardian.com/us-news/2026/may/13/utah-approve...
> The facility will require about 9GW of power, which is more than the entire state of Utah currently consumes
Or because Chinese AI has enabled a complete crackdown on dissent?
Hmm, imagine that!
All technologies have political, social, ethical, and economic aspects, and not all technologies are the same. AI is not, in reality, powered by CO2-neutral electricity. As the article mentions, AI is the fastest-growing energy user in the world, so it probably merits a deeper look than aluminum cans.
And while it is 'fastest-growing', it is small as an absolute share - around 0.2% of total world CO2 emissions.
Or if truly all jobs can be done by machine, we retire and go live a life of fully automated luxury space communism.
From what I can tell your argument works out to be as simple as we can’t do this because we want to keep people working, which the vast majority of people hate doing.
There is a small minority of people who enjoy work. It seems cruel to enslave a people in mass simply to keep them busy.
I'm not arguing for forcing people to work. If we get to a point where no work is economically feasible for humans to do I would very much like some form of UBI or similar to take care of people.
Speaking of enslaving people, the billionaire class has started speaking of removing peoples right to vote if they do not pay taxes (e.g. if they do not work.)
https://www.yahoo.com/news/politics/articles/why-did-154-bil...
I very much wish that people would no longer have to do busy work and that they would have time for family and hobbies, but being the pessimist that I am, I'm not very hopeful we'll get there any time soon.
Someone suggested that pensioners shouldn't get a vote, and the Shopify founder responded with "good system".
That's it. That's the whole story. Somehow you turned that into "billionaires are trying to enslave us", probably because you already believe that anyway.
Now, I'm fully aware that it's just one guy. But technically, under the proposed system it would be more like 5 guys.
- I now regularly receive AI-written spam emails (marketing pitches, "vulnerability reports").
- I was receiving AI-written garbage quality pull requests. It got so annoying I closed pull requests on my project's repo.
- When I try to contact some company's customer support these days there's a good chance I will be fed slop by an agent posing as a customer support person, this is absolutely infuriating.
- When I search for information on the internet, for certain queries, 90% of the results is slop.
- AI crawlers are DDoSing websites.
- The already mentioned things – hardware getting more expensive, climate impact, unethically sourced training data.
True. AI is actually useful.
It is more like leaded fuel or DDT.
Is the assumption that maintainers did not consider those things?
Of course Linus realizes linux development is political, because he's lived that, however he might phrase it.
What I think both are actually saying:
Drew Devault (blogger): Linux development should place political/moral/ethical arguments in primacy (or at least more important?)
Linus: Technical reality should always come first, with other considerations secondary
If the above read is correct, I'm sympathetic to Linus' position. If the linux kernel doesn't work and improve, technically, then nothing else about the project matters.
Bikeshedding about political/moral/ethical qualms impairs the technical work.
Ergo, a strong lighthouse statement from Linus that the technical work always comes first, even if other considerations are made when appropriate.
The AI genie (referring broadly to 'any LLM-based tool that performs a useful development function') isn't going back in the bottle, and linux sticking its head in the sand for ethical reasons will harm linux.
And despite the author’s logical leaps; no it's not impossible to fork the kernel and go another route. The fact it doesn't happen is because its stewardship so far has been exceptional.
It's not a political issue. It's a religious one with the anti AI people raising pitchforks.
That doesn't mean I _should_ be given access to a nuclear missile.
Share of global CO2:
China 33%
US 12%
India 8%
Russia 5%
Japan 3%
History didn't start in the last 12 months.
AI is at the top of an industrialization iceberg going back a century, with uneven countries as beneficiaries and externalities.
It seems like for much of Western world, Indians aren't really humans but a giant undifferentiated mass of temporarily animated bodies that are fit only as a canvas for one's own political paintwork.
Instead, it's:
> how many tons of CO₂ added to the atmosphere or liters of fresh water supplies disrupted is a tolerable price for a better code review?
The answer isn't zero, is it?
This paragraph links to the IEA report which does indeed call datacenters the fastest growing energy consumer, but also says:
> Despite rapid growth, data centres remain a relatively small part of the overall power system, rising from about 1% of global electricity generation today to 3% in 2030, accounting for less than 1% of total global CO2 emissions.
For scale, beef production is about 9% of total greenhouse gas emission.
It feels wrong to me, though, to see AI being used in kernel development, as I see the trouble it causes on regular projects.
Fewer than if a human had done it without AI: https://news.ycombinator.com/item?id=48941143
I guess that means he should advocate for more use of AI for Linux dev! ;-)
And they're driven by fear--fear of being replaced. This is one reason why open source projects are limiting use of LLMs. LLMs finally make software free! Or at least, much cheaper.
In reality, these reactionary anti-AI policies mean that users are private forking OSS projects, improving them with AI, and not contributing the changes back, making the original OSS project even more irrelevant over time.
The strong argument to make is that AI code review sucks. It does a bad job. It denies there are bugs that you have personally verified exists, questions the reason for the patch existing based on this denial, and then demands you create a giant scaffolding of tests to prove to the AI that the bug exists and that your patch will fix it. It is rude to do that to someone contributing their free time to help your project. And yes, I have forked one project as a result of this already.
AI code review will atomize your community with great speed. Beware.
But if you want to use a non-AI OS, NetBSD banned the use of AI. Maybe some of the talent will move over there. They can sure use the help.
As for AI in Linux, I am on the fence about that, but since Linux is pretty much funded/owned by large corporations, AI being forced on Linux is no surprise at all.
Because if so, please, please elaborate on
> I do not see how AI could be legally used on a GPL project.
Too many people "believe" or "are convinced" about matters with AI and Open Source without any backup or sources.
when you license code under the GNU GPLv2 (as you are doing whenever you contribute to the kernel), you are making an overt intellectual property claim. Unless you include attribution back to another GNU GPLv2 (or compatible) source that you are copying from, you are asserting that all the code you've just contributed is work that you have a the full legal copy-right to distribute and (re-)license.
However, if non-trivial code contributions are made by LLMs, that cannot be ensured (even if the user intends for it to be the case), as LLMs will never be able to consistently and correctly attribute the source of their code (not to mention a lot of it was illegal to train on to begin with but that's besides the point).
An LLM could very easily reproduce code from a project which is incompatible with the GNU GPLv2, and not only will it be illegally integrated into the kernel source, but the contributor is actually falsely claiming it as their own original work.
> Are you a lawyer or an intellectual-property legal expert?
Appeal to authority - we are free to comment on this matter irrespective of our vocation/education.
<https://www.copyright.gov/ai/>
<https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...>
The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.
> As described above, in many circumstances these outputs will be copyrightable in whole or in part—where AI is used as a tool, and where a human has been able to determine the expressive elements they contain. Prompts alone, however, at this stage are unlikely to satisfy those requirements.
Then, since this stuff is changing a lot, they hedge with:
> The Office continues to monitor technological and legal developments to evaluate any need for a different approach.
Linus himself was made to sign the corporation-enforced CoC and "retreat" for "counseling".
He came back a changed man. Since then I've only see him a) use the CoC as a weapon for the old boys and b) parrot any current mainstream or corporate line.
That includes COVID vaccines, overt anti-Russian statements (when not warranted) and now AI. Would he still give Ndivia the finger? I doubt it.
These Gen-X millionaires cling to their positions and I'm sure the Linux Foundation has a well oiled HR department. Ted Tso also already interfered pro AI in the Debian discussion, where the pro-AI resolution is also championed by Canonical shills.
You can't fork, but you can use BSD.
Also, the open models are a great alternative to the datacenter ones.
I imagine AI will go towards the local models more when the bubble pops.
I didn't see any objection to LLM assisted development in the abstract. The objection is to relying on proprietary AI providers in general, and Gemini in particular, via Shashiko. The problem is that kernel contributions could become gated by hyperscalers who can draw on compute power normal people can't access.
Devault mentions data centers' carbon emissions and water usage. I find both of these objections to be distractions. It's not that carbon emissions don't matter. Keeping greenhouse gasses in check is incredibly important. But there are far more productive ways to cut carbon emissions than to block data centers. Heavy industry consumes far more electricity than data centers, for less economic value. This is even truer for water use. If carbon emissions is a major concern (as it should be), there are better targets to go after than data centers.
A more compelling argument is the danger that the Linux kernel becomes something that's effectively controlled by proprietary AI providers: Google, Anthropic, OpenAI, Microsoft, etc. The photo of tech company leaders at Trump's inauguration underscores their lack of any moral compass.
But if AI-driven kernel development is actually useful and productive, that's going to be hard to turn down. If you believe that LLM use is technically counter-productive, then this whole AI thing is a bubble and you can just wait for it to pop. If you believe that LLMs can be technically productive if used correctly, and the drawbacks are entirely social, then it means giving up something technically useful. That is a real cost.
I think Linus is right here: it would be hard for regular kernel developers to coordinate a fork of the project. It might not be that hard for the hyperscalers to simply fork Linux. They would still release the source code under the GPL, but nobody would be able to keep up with it. Ultimately, this could doom the human-written version of the kernel to irrelevance as most people switch to the faster/stronger/harder LLM-derived fork.
Maybe a reasonable medium would be to insist on using an open model for code review instead of depending on Gemini or any proprietary model.
Repeat after me: you can't put the cat back in the bag. Pandora's box is open. Sucks, but you have to learn to live with the reality that most developers will use AI to write code from now on.
Your suggestion that "anything popular cannot be morally wrong" is proposterous. It feels like you're trying to bait people into arguing with you in a way I associate with bot spam.
But on the off change you're a real person - I will politely say - just because something is useful and popular doesn't meant it doesn't have negative externalities and doesn't mean people shouldn't fight to make it better.
Improving emissions standards is important even though lots of people drive cars. The same can absolutely be true of AI.
Presumably, I'm a bot because you disagree with me.
> Your suggestion that "anything popular cannot be morally wrong" is proposterous. > just because something is useful and popular doesn't meant it doesn't have negative externalities and doesn't mean people shouldn't fight to make it better.
PREposterous. And I said nothing of the sort. It feels like you're hallucinating things nobody said in a way I strongly associate with bot spam.
> Improving emissions standards is important even though lots of people drive cars. The same can absolutely be true of AI.
The article did not at any point suggest that we need to improve and optimize AI. There is no magic wand that Linus can wave to measure how much water the incoming AI-assisted contributions are using. The guy should just go protest at a data center construction site or something; the article is pointless and completely fails to diagnose the actual problems with AI or propose viable solutions.
It’s not about being popular, it’s about facing a new economic reality. Once machines were invented there was no going back to human labor, no matter how bad it was for the displaced workers or the environment.
Technology has always been a relentless economic force. And this is largely because despite the myriad externalities, it’s blindingly obvious that we as a species are, on average, much better off with technology than without.
if you're just saying it as an ad hominem, that is pretty lame, and you can make a stronger argument without it.
No I don't. I can urge them to actually do their jobs and not just turn out slop, while refusing to use the slop machine myself.
Furthermore your comment ignores key arguments in the article, particularly the argument that contributors should not be forced to participate in AI-based ceremonies such as LLM code review that they may object to on personal ethical grounds.
Remind me please, how much exactly do data centers contribute to greenhouse emissions? And how much something like, I don't know, meat production contribute?
If my Googling is correct, apparently a single ChatGPT query produces ~0.004 kg CO2e/kg. Beef production contributes ~60 kg CO2e/kg. So you'd need to make, err... over 15k ChatGPT queries to produce as much emissions as eating a single kilogram of beef.
I've seen anti-AI protests. Where are all the anti-meat protests? I don't mean to engage in whataboutism, I'm just saying -- if your goal is to reduce CO2 in the atmosphere then there are bigger fish to fry, instead of just virtue signaling because there's a new trendy bogeyman in town.
I've seen anti-AI protests. Where are all the anti-meat protests?
Fucking everywhere bro.
29.08.2026 end-of-speciesism.org/en/
1.10.2026 wikipedia.org/wiki/World_Vegetarian_Day
2.10.2026 dayforanimals.com (birthday of Gandhi)
1.11.2026 wikipedia.org/wiki/World_Vegan_DayThat means you are likely responsible for more water wastage than a datacenter
And it's not Linux' task to close that, even if it could, which it cannot. How is that disingenuous?
I agree that we should keep discussing the merits and dangers of AI-aided software development. Broad, so indeed also CO2 etc. But that discussion is not something that Linux should push forward. It's a discussion that should take place outside the context of specific projects and in the context of the AI tools and providers, IMO.
The real argument was that AI has many externalities that should be considered (climate change, water use, energy prices, putting people out of jobs) and that Linux, as a relevant and influential political actor, should engage with these topics. It's not even giving a final determination about what Linux should do, it's just disagreeing with Torvalds' total rejection of ethical arguments about AI.
Even if you believe that the technical merits outweigh the other costs, it seems at least reasonable to engage on those grounds. But Torvalds isn't doing that.
Also, simply stating that the "cat is out of the bag" and "Pandora's box is open" isn't really an argument. These are thought-terminating cliches; not logical arguments
Consider sometimes terminating some thoughts and replacing them with actually useful ones. It’s good for you and your goals regardless of what they are.
People burning thought-hours on pretending that we can make AI go away are basically doing nothing except emitting CO2 and consuming water. If you want change, start with yourself!
Anyway, so far zero actionable suggestions for Torvalds in the article or in this thread. Great job so far.
That's not social warrior, that's simple facts. We don't need to try and guess motives or reasoning for people's contributions - that doesn't matter. What we have is an explicit declaration by virtue of the license, that the code being provided can legally be included in the GNU GPLv2 licensed "Linux" kernel. But with LLM generated code, this legal claim is dubious at best. For trivial (!= simple) code, one can probably argue that the LLM code will not be violating anyone's copyright, however for anything remotely novel, unless you are 100% certain that all of the training data was GPLv2 or compatible, it is not legally safe to use - and aside from that, even if all the training data was compatible with the license, the LLM will virtually always fail to properly attribute the source of its output, which is a requirement of - as far as I can tell - every license compatible with the GNU GPLv2. I cannot share GPL'd code and misrepresent it as my original work, I still have to make clear that it was someone else's creation.
LLM generated code is a ticking time bomb, and is going to poison projects.
Yes, it is possible to do this already without LLMs, but not by accident. A developer would have to make a conscious decision to violate someone's license in the past. Now it can happen without anyone being the wiser.
And as someone who very much uses GNU GPL licensed software because of its license, this worries me for the future of software. On technical merits, it doesn't matter, but if you care at all about the artful expression of software, as well as honesty about who created what, this should concern you.
Given the recent Anthropic ruling and OAI settlements, and the general stance of courts to look the other way when it comes to LLM use, I think an exemption for LLM-based software will be handwaved into existence (to my regret.)
I don't want to accuse the courts of intentionally siding with big corporations (I have no evidence to support such a claim anyway), but the only other option I see is that the courts are uninformed on either (A) the technology at play, or (B) the requirements of many software licenses.
Though, this should also apply to works licensed under variants of the Creative Commons (with Attribution), since once again, you can end up using parts of a work that requires attribution without even knowing that it's someone else's work (or at least without knowing its license requirements). There might be a better argument for generative AI images being transformative (though I think still a weak argument since you could almost certainly produce a 1:1 reproduction of a piece of training data, much like LLMs have already been pushed to doing with snippets from books).