Europe needs hyperscalers (lots of them) [LWN.net]

Europe needs hyperscalers (lots of them)

Posted Feb 10, 2026 21:56 UTC (Tue) by rbranco (subscriber, #129813)
Parent article: FOSS in times of war, scarcity, and AI

Europe needs startups (lots of them) and hyperscalers too. Also lots of them because competition is good. Otherwise it won't be a match for the US & China. This hipster mentality - buy indie & drink a homebrew afterwards - is detached from reality and it'll be discarded soon anyway. Big brands can coexist alongside craft beer.

And this "there is no Claude, only other people's code" is just another empty phrase that some people will print in t-shirts and that's it. But is it verifiable true? Otherwise it's just copium. We create new works by recombining words according to some rules and LLM's are large language models.

to post comments

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 1:05 UTC (Wed) by mirabilos (subscriber, #84359) [Link] (24 responses)

Perhaps we do *not* need “hyperscalers”, after all. Or to be “a match”. Just to do our own thing.

And no, we don’t need the eso-fascist planet-burning theft machine, either.

We “create new works” by acts of human creativity. This is what needs encouragement. I’ll add a call for UBI to that; after all, pilots were successful, and partial models are now permanent (as was just announced this week).

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 10:04 UTC (Wed) by rbranco (subscriber, #129813) [Link] (23 responses)

You need hyperscalers if you want European companies to have a presence in other countries.

And WTH is "eso-fascist". I tried a Google search and you seem to repeat this everywhere. Europe does have a strain of "eco-fascism" though, seen in the likes of Extinction Rebellion with a circled hourglass that resembles a swastika. An almond plantation in California consumes 4 times more water than all US datacenters combined. Avocado farming is drying up Spain. Thanks to "AI" nuclear energy is on the table again so enough with the secular doomerism. The whole thing is overblown.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 11:56 UTC (Wed) by nim-nim (subscriber, #34454) [Link] (16 responses)

But do we need European companies to have a presence in other countries? What we need is European companies serving European interests and values in Europe. Once *that* works, maybe exporting the result to other places where they have the same needs is valuable. Or maybe it will be taken as a way to assert a form of dominance in places that resent and have not asked for this dominance.

The things that work in Europe and that have been exported successfully somewhere else (for example good technical and sanitary standards) were done by Europeans for Europeans first. Because Europe has the money to serve its own needs when it wants to. Doing garbage because “we want a presence in other countries and those countries are more permissive than the European market” does not work for the European market and usually does not work for the export market either (who wants to import someone else’s garbage).

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 12:17 UTC (Wed) by rbranco (subscriber, #129813) [Link] (2 responses)

> But do we need European companies to have a presence in other countries?

Yes. Otherwise other companies will fill that vacuum. Autarchy doesn't work. Companies need to grow.

Europe lost a sugar daddy (the US) and got a delayed adolescence as a result. Switching to another (China) is not the solution. Europe needs to grow up.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 14:39 UTC (Wed) by nim-nim (subscriber, #34454) [Link] (1 responses)

Growing up does not mean starting with a presence in other countries. It means forgetting about other country wishes for a time and concentrating on its own needs first to get working products and services out of the door.

Maintaining a presence for presence sake gave us spectacular turds like MicroNokia. Extrinsic is no substitute for intrinsic growth, you need a robust intrinsic backbone to survive extrinsic expansion.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 15:37 UTC (Wed) by rbranco (subscriber, #129813) [Link]

Europe wouldn't start from scratch.

And MicroNokia wasn't the end of the world. We can't predict these outcomes and we don't need to micromanage them either.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 15:29 UTC (Wed) by mathstuf (subscriber, #69389) [Link]

That sounds an awful lot like an echo of the current US administration's "America first" rhetoric.

> Doing garbage because “we want a presence in other countries and those countries are more permissive than the European market”

Corner cutters will always exist. Alas, it has economic knock-on effects of the more upstanding entities…but the EU is "good" at regulation (certainly compared to others), no?

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 20:40 UTC (Wed) by khim (subscriber, #9252) [Link] (11 responses)

> But do we need European companies to have a presence in other countries?

There are, ultimately, only two choices:

Ensue that European companies have something worth selling for other countries in exchange for raw resources (e.g. uran that's needed to drive electricity generation in France).
Accept return to technologies of XIX or maybe even XVIII century that can be supported using resources that Europe have indigenously.

And if you rejecting choice #1 then you are getting choice #2 by default.

> Because Europe has the money to serve its own needs when it wants to.

Europe have papers that it calls “money”, but we are fast approaching point where the only way to ensure that something you call “money” is worth something would be raw military might and/or the things that you may exchange with others.

Europe doesn't have former (and don't have time to restore it) thus it's imperative to have latter.

> who wants to import someone else’s garbage

Well… Turkey was willing to do that, apparently — but, again, you need some money that are perceived by others as money for that.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 9:56 UTC (Thu) by kleptog (subscriber, #1183) [Link] (10 responses)

Europe's Achilles heel is that it imports ~50% of its energy needs. While we've made great strides in energy efficiency, that's not enough by itself.

> 1. Ensu[r]e that European companies have something worth selling for other countries in exchange for raw resources (e.g. uran[ium] that's needed to drive electricity generation in France).

That's not a problem. We spend ~375 billion on energy imports, while exporting ~4 trillion in goods & services, so a bit less than 10%. The problem is not being able to import energy at any price.

We don't need to open our data-centres to people from other countries.

> 2. Accept return to technologies of XIX or maybe even XVIII century that can be supported using resources that Europe have indigenously.

Or go straight to XXI century technology. Anything non-renewable is dead-end anyway, it's just a matter of time. Nuclear would be great, but right now we're installing renewables at a rate of about 1 nuclear reactor every 5 days. It's going to be some time before nuclear installations ramp up to meet that, if ever.

And of course, fusion power is only 20 years away /s

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 10:13 UTC (Thu) by taladar (subscriber, #68407) [Link] (8 responses)

Nuclear power isn't actually great at all, it needs fuel that needs to be imported from a small number of countries again.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 12:16 UTC (Thu) by anselm (subscriber, #2796) [Link] (7 responses)

Current nuclear reactors are very inefficient in the way they use their (imported) fuel. In principle, we know how to build nuclear reactors that would make more efficient use of their fuel, and/or would be able to use existing “nuclear waste” as fuel and, as a side benefit, make it less nasty in the process. The remaining challenge is to make these technologies competitive with renewables on a euros-per-megawatt-hour basis.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 13:37 UTC (Thu) by Wol (subscriber, #4433) [Link] (3 responses)

> and/or would be able to use existing “nuclear waste” as fuel and, as a side benefit, make it less nasty in the process.

Aka "fast breeders". I've also heard something about thorium burners that apparently burn up pretty much everything.

> The remaining challenge is to make these technologies competitive with renewables on a euros-per-megawatt-hour basis.

So using a fast breeder to turn U-238 into Pu-239 would be one obvious way. Processing (separating) waste into short- and long-lived radio-nuclides, and using the nasty stuff to bulk up new rods from old also seems an obvious way to me to reduce costs - store the nasty stuff inside the reactor, and break it up at the same time!

Of course, there's the political problems, and geo-stability ... the German nuclear industry as an example of the former, the tsunami and Japanese example of the latter. Then of course, we've got Dungeness (now shut down?), but the site is built on an area that was 5 miles out to sea in Roman times, and if we have global warming could very easily become 5 miles out to sea again in a pretty short time ...

Cheers,
Wol

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 14:59 UTC (Thu) by malmedal (subscriber, #56172) [Link] (2 responses)

> seems an obvious way to me to reduce costs

These things have been tried, they are not cheap. The reason they(I think France got the furthest) tried was to have the technology ready for when the uranium mines ran dry, not because it would immediately save money.

To run a reactor efficiently and safely you need to know exactly what the fuel is made of down to at least cubic decimeter resolution.

For instance Xenon-135 is a neutron absorber produced by nuclear reactions, when this builds up you need to increase neutron flux to compensate, however being a gas it can escape quickly if it gets the chance, so if the operators make a mistake the reactor can overheat and melt down.
(This is likely a major reason for Chernobyl)

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 15:38 UTC (Thu) by Wol (subscriber, #4433) [Link] (1 responses)

Don't remember the technology - was it AGR? I remember our AGR reactor technology got overwhelmed by the American PWR reactors, despite being much safer.

Iirc it used deuterium (heavy water) as the moderator, so if it got hot, the moderator boiled away and the fast neutrons escaped without being captured and driving fission.

(I thought the main reason for Chernobyl was the operators disabling the safety features "to see what will happen"!)

Cheers,
Wol

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 17:38 UTC (Thu) by malmedal (subscriber, #56172) [Link]

> I remember our AGR reactor technology got overwhelmed by the American PWR reactors, despite being much safer.

Don't know if they were safer, but it's clear they were more expensive.

> Iirc it used deuterium (heavy water) as the moderator

No, graphite for moderator, CO2 as coolant. They used control rods to control the reactivity.

> "to see what will happen"

There are very conflicting stories about the details, but it is clear that it was a pre-planned test, deemed necessary to verify correct operation. It was supposed to be done by the day-shift, but it was delayed, so the unprepared night-shift got the job.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 17:46 UTC (Thu) by rgmoore (✭ supporter ✭, #75) [Link] (2 responses)

The biggest problem with nuclear is that it's still a heat engine, and just building the heat engine parts of the power plant (i.e. the external steam loop, turbines, and cooling system) is more expensive than equivalent renewable energy. Even if the actual reactor were free- and they're obviously far from it- and it still wouldn't be able to compete on cost with renewable energy. That's today. Renewable energy (including battery storage) is still getting cheaper, so its competitive advantage over nuclear is only going to grow.

Of course renewable energy isn't perfect. It still has problems, and there are probably specific applications where other technologies have non-cost advantages that outweigh renewables' cost advantage. There's also no good reason to stop using nuclear plants that are already paid for and are running well. But we're now at the point that plans for new and replacement, non-renewable power plants need to include a justification for not using renewables instead.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 18:04 UTC (Thu) by Wol (subscriber, #4433) [Link] (1 responses)

> But we're now at the point that plans for new and replacement, non-renewable power plants need to include a justification for not using renewables instead.

The justification for nuclear is "it's not carbon". So much of this carbon-neutral is exactly that - crap. When you burn anything the question should be "how long ago was the sunlight I'm releasing locked up?". Anything more that 100 years or so should be a red flag (nuclear doesn't count here - that's releasing ancient starlight :-)

Renewables, it's possibly measured in hours, which is great. Wood, of course, while not a particularly good fuel in many ways, is measured mostly in a century or two (or less). Coal and oil are adding to the CO2 burden, and from my knowledge of what's going on, I think we passed the point of no return quite a while back. I suspect Khim may be right in saying Europe will be going back to the 17th century, but on current performance I suspect the rest of the world will be joining us!

Cheers,
Wol

Maybe that's enough

Posted Feb 12, 2026 19:41 UTC (Thu) by corbet (editor, #1) [Link]

Can we agree that this has wandered pretty far off-topic, even for an article like this one? Energy issues are certainly of great interest, but we'll not solve them on LWN.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 15:39 UTC (Thu) by khim (subscriber, #9252) [Link]

It's really funny how you may say two sentences that directly contradict each other in one, single, paragraph.

> That's not a problem.

Interesting… and why is that?

> We spend ~375 billion on energy imports, while exporting ~4 trillion in goods & services, so a bit less than 10%.

So that's not a problem because currently Europe does have things to sell… okay.

> The problem is not being able to import energy at any price.

IOW: it's is a problem. And a pretty big one.

We are entering times where money are almost entirely detached from the actual things that are needed: there are mountains of money that are exchanged in the virtual world of “services” and there are real things that one couldn't buy at any price except if you have the right connections.

Things like “rare earth”. There was lot of hoopla about these in the press, lately… what is the total market of these? Go google that and compare to market capitalization of AI companies (which are using these same rare earths)… you would be surprised.

And, again, Europe either have to ensure that Europe does have “good and services” to sell — or it would find a way to live without things Europe needs.

Chances are extremely high that “XXI century technologies”, in Europe, would be identical to XVIII century technology and not anything new and exciting.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 12:16 UTC (Wed) by ballombe (subscriber, #9523) [Link] (5 responses)

There will not be any European commercial hyperscaler, because if someone want to start a new ethically-challenged business, there are better places than the UE.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 19:55 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link] (4 responses)

What exactly is unethical in "hyperscalers"? If anything, they're about as ethical as any business is. They provide a service and actually bill you for your usage.

No ads or data mining. You put in money, you get services in exchange.

And just with all other businesses, there are economies of scale. Hyperscalers have lower fixed overhead per "unit of compute". This is a _good_ thing, we want society to be more efficient and wasting less resources.

Europe needs hyperscalers (lots of them)

Posted Feb 13, 2026 21:56 UTC (Fri) by leromarinvit (subscriber, #56850) [Link] (3 responses)

There's no doubt that hyperscalers are efficient on their end of the business, which is of course why they're making lots of money per unit of whatever they sell. But even though the prevailing wisdom these days is to go with them and just pay the bill (which can sometimes be a reasonable business decision), quite a few who have done the math have ended up with the conclusion that it's cheaper to run your own infrastructure, as long as it can be utilized fully and load isn't hugely variable (and unpredictably so).

I think the complaints about hyperscalers are twofold:

For one, they're (pretty much by definition) huge companies, so almost inevitably they end up doing huge company things like lobbying, tax "optimization", and so on. Exhibit A: "Do no evil."
They structure their pricing in a way that it's easy and cheap to get hitched, but very expensive compared to their actual costs once you scale up later. As you say, that's about as ethical (or not) as any other business trying to increase its profits - but it can be a valid argument against using their services.

Europe needs hyperscalers (lots of them)

Posted Feb 13, 2026 22:31 UTC (Fri) by malmedal (subscriber, #56172) [Link] (1 responses)

> cheaper to run your own infrastructure,

I think the real problem is that the people controlling the money don't understand the language of the technical people
they need to run the infrastructure. In particular, they have no real way of knowing if they even are competent at their job.

That said, there are a number of objective advantages to using something like a fully managed database where you just write the SQL and it just works the same whether the query needs a single worker for a second or ten thousand workers for an hour.

One thing the EU could do would be do mandate interoperability, force the vendors to agree on common SQL, common orchestration etc.

I wouldn't hold my breath waiting for this though, just mandating USB for phones took about fifteen years...

Europe needs hyperscalers (lots of them)

Posted Feb 16, 2026 9:46 UTC (Mon) by taladar (subscriber, #68407) [Link]

There are pretty much zero advantages to being able to scale up to tens of thousands of workers for an hour for 99% of all applications since those run on a single small-ish cloud server just fine.

Also, I used to think the problem was that the people controlling the money couldn't determine the competence of the technical people but I think the evidence becomes more and more clear that the people controlling the money can't even tell if they themselves or their peers doing their job are competent.

Europe needs hyperscalers (lots of them)

Posted Feb 13, 2026 23:25 UTC (Fri) by Cyberax (✭ supporter ✭, #52523) [Link]

> quite a few who have done the math have ended up with the conclusion that it's cheaper to run your own infrastructure, as long as it can be utilized fully and load isn't hugely variable (and unpredictably so).

Sure, and that's fine. This is just normal commercial activity, you check available options and decide which one is the best for you. This also happens all the time with "classic" businesses that need to do lease-vs-own calculations or find suppliers for some components.

> For one, they're (pretty much by definition) huge companies, so almost inevitably they end up doing huge company things like lobbying, tax "optimization", and so on. Exhibit A: "Do no evil."

I actually like that. Business interests of companies should influence the lawmakers. I just don't necessarily like the _level_ of their influence.

> They structure their pricing in a way that it's easy and cheap to get hitched, but very expensive compared to their actual costs once you scale up later. As you say, that's about as ethical (or not) as any other business trying to increase its profits - but it can be a valid argument against using their services.

It really depends. I know AWS from inside out, so I can easily keep my costs way below what I'd pay for a similar service if I were to build it myself. But I agree that AWS makes it very easy to build infrastructures that just burn through money. But it's also not necessarily a sign of "evilness", especially since AWS provides tons of tools to control the costs.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 9:00 UTC (Wed) by kleptog (subscriber, #1183) [Link] (8 responses)

We don't need hyperscalers, but we do need competition. The biggest issue currently is that if you have your infrastructure setup in (for example) AWS using Terraform, it's non-trivial task to convert that to any other cloud provider. It's basically vendor lock-in, cloud-style.

Until it's easy to transport your infra between cloud providers there can never be true competition. FOSS can help there.

The most portable setup I've come across so far is having your application run under K8s, and then using the cloud-providers to supply a K8s environment. Because it's all standard you can run the same application anywhere. But of course this ignores 90%+ of the services cloud providers provide.

Europe needs hyperscalers (lots of them)

Posted Feb 11, 2026 19:33 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link]

With K8s it's not that hard to do the transfer. In my experience, most of the complexity happens if you want to keep your service available during the migration. If you can afford ~8 hours of downtime, a lot of things become so much easier.

Other than that, it's always the case of complex systems growing implicit dependencies where you don't expect them.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 9:52 UTC (Thu) by taladar (subscriber, #68407) [Link] (6 responses)

I haven't worked with K8S that much but my impression is that it is literally the anti-thesis of standardization, every little subsystem has multiple options that are mutually completely incompatible and often proprietary and undocumented and there isn't even any sort of default.

Even the few subsystems like storage and ingress where some vague resemblance to standardization exists are often only maybe 90% compatible. And good luck getting any kind of tooling to replicate a setup that uses the same configuration format on another provider or locally.

Europe needs hyperscalers (lots of them)

Posted Feb 12, 2026 19:03 UTC (Thu) by Cyberax (✭ supporter ✭, #52523) [Link] (5 responses)

The problem with K8s is that it's extremely flexible. It can be used for anything from a single-node "cluster" to deployments with millions of nodes and multiple concurrent rolling deployments.

Europe needs hyperscalers (lots of them)

Posted Feb 16, 2026 9:42 UTC (Mon) by taladar (subscriber, #68407) [Link] (4 responses)

It feels like one of those technologies where those who need millions of nodes are pushing some of their costs onto those who do not even need HA in the first place via hype to be honest.

Europe needs hyperscalers (lots of them)

Posted Feb 16, 2026 20:32 UTC (Mon) by Cyberax (✭ supporter ✭, #52523) [Link] (3 responses)

To an extent. K8s is a bit over-engineered in that regard. For example, it uses etcd to be resilient against the database failures for its control plane. Hardly anybody really needs this.

But it's not _all_ over-engineering, a lot of K8s complexity is justified once you get to a ~10-20 node cluster. Things like resource-based scheduling, gradual rollouts, and automatic rollbacks become important.

And just like with Linux itself, the set of features that are important to you, is often very different from the set of features that are important to me. So it's hard to create one definitive "standard" minimalistic set of features that would satisfy everyone.

Europe needs hyperscalers (lots of them)

Posted Feb 17, 2026 6:42 UTC (Tue) by zdzichu (subscriber, #17118) [Link] (2 responses)

etcd is not mandatory. On my home cluster I'm using k3s, with single PostgreSQL as a data store. Works perfectly on trash old laptops I've used as nodes.

Europe needs hyperscalers (lots of them)

Posted Feb 17, 2026 18:55 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

Yes, but K3s is not K8s. And the design of K8s is severely affected by etcd limitations, even though all the large players have versions of K8s that use different storage backends.

Europe needs hyperscalers (lots of them)

Posted Feb 17, 2026 20:19 UTC (Tue) by zdzichu (subscriber, #17118) [Link]

K3s _is_ k8s – it passes conformance test suite and has CNCF's certificate. But I recognize this is a nitpicking not contributing to the discussion, so I'll yield here.