Nebius Group N.V.

Jason support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E296 · 2026-11-07 ▶ 1:12:48

running local models first and then falling back to them because we don't want to be throttled. We want to have unmetered intelligence.

Chamath support ×3 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E296 · 2026-11-07 ▶ 1:10:41

If anything, I think what's going to happen is, all the renter classes will get co-opted. For example, I don't know if you guys saw today, but Google just took over a Crusoe data center. They co-opted the renter. They're like, you're gone, we're going to take it over.

Chamath support ×1 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E295 · 2026-10-30 ▶ 46:41

If that's true, then Friedberg's chart is also true, which is that those rates cascade through the ecosystem. The neoclouds will have to pay more. All these other folks have to pay more. What that means is that you increase and you pull forward the risk of defaults, which we talked about last time.

Friedberg support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E295 · 2026-10-30 ▶ 15:51

I've talked a lot about moving to open source, and we have our own cluster installed with our own kind of open source now.

The challenge we have is, it's very hard to find a global, resilient, highly available, inference provider outside of the hyperscale community right now.

we preference open-source, open-weight, because it allows us to modify the weights and train them. Something which most people don't talk about is, it allows us to interrogate the chain of thought. Because, over time, what's going to happen is, all of those enterprises are going to need explainability, auditability, and verifiability.

Jason support ×1 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E294 · 2026-10-23 ▶ 6:39

Wow, we're back to the future. People staying at their own data centers, yeah.

Chamath support ×3 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E294 · 2026-10-23 ▶ 4:50

Ever since Alex Carp's commentary on CNBC about data sovereignty, buying patterns have changed. And I think people are being a lot more methodical. They are building out their own data centers or at least moving into VPCs, virtual public clouds.

Jason support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E293 · 2026-10-17 ▶ 1:15:46

That you own and that you have AI sovereignty, you're not educating a frontier model company.

Sacks support ×3 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E293 · 2026-10-17 ▶ 1:15:12

And so what you're seeing in the enterprise right now is that for a lot of enterprise workloads, you don't need a general superintelligence. You don't need a frontier model. You need a small specialist model that's been trained on your own data.

Sacks oppose ×3 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E293 · 2026-10-17 ▶ 40:25

So, in any event, payback period is really critical and as long as the payback periods are really short, I don't see how these companies can lose

Jason oppose ×2 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E293 · 2026-10-17 ▶ 18:30

15% chance at the end of 27 next year, 23% at the end of 2028 seems a little far-fetched. Given the fact that they have a massive demand for their equipment, as Ben pointed out.

It's why, by the way, we're not invested in any of these neoclouds either, for that reason, because that whole thing scares me, having been through it.

Sacks support ×1 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E290 · 2026-09-26 ▶ 45:21

I fully understand that the majority of tokens will go to open models or commodity intelligence.

Jason support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E290 · 2026-09-26 ▶ 44:31

We need people to use those models so they have sovereignty, so you're not giving your data to the frontier model so that they don't get that tip of the spear superintelligence edge, which is why people are disengaging.

Friedberg support ×3 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E290 · 2026-09-26 ▶ 38:08

Separately, I think most of the enterprise applications are going to be run using these open weight models.

Chamath support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E290 · 2026-09-26 ▶ 26:10

Now, you may choose to go down to cheaper and cheaper Anth models, so then Anth keeps the revenue, but there's also a risk that you may do what Friedberg just said and just go to an open source model and just host it yourself.

Jason support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E289 · 2026-09-11 ▶ 1:15:55

They're going to need to have rack servers. They're going to stack rack servers themselves.

Friedberg support ×3 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E289 · 2026-09-11 ▶ 1:13:17

This is why I care a lot about open source, because I don't want them having my chat logs, because they can use it for training to create an IP advantage that is now diffused to the rest of the market.

Sacks support ×2 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E289 · 2026-09-11 ▶ 1:10:41

I think we need much stronger data privacy laws around the data that you have with AI services.

Chamath oppose ×1 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E289 · 2026-09-11 ▶ 1:07:43

You can work with folks like Amazon or Nebius or Fireworks.

Chamath support ×3 ▲ on 🤖 Enterprise AI sovereignty: workloads leave frontier APIs for private stacks E289 · 2026-09-11 ▶ 1:03:56

Will a handful of CIOs get very publicly flogged and fired in the next year because they accidentally didn't understand this and just did an API deal because they wanted to feel popular and leaked data into these models? Guaranteed.

Gavin Baker oppose ×3 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E221 · 2025-03-29 ▶ 12:47

I think what may be underappreciated about CoreWeave is it's actually really hard to run these big training clusters. Everybody thinks it's easy, but to synchronize tens of thousands of GPUs where they're melting, or cables are being unplugged, and you lose a bunch of training data when it happens. I just think it may not be the commodity that everyone thinks it is, and it may turn out that it's much harder to do than people think.

Friedberg support ×2 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E218 · 2025-03-08 ▶ 59:13

And so I worry a little bit about a business like this, where there's four or five companies that are each doing 80 billion dollars of CapEx this year to create infrastructure that effectively starts to replace what these guys are effectively offering out as a service. ... If I was to do diligence on this business, that's where I would spend a lot of my time is like, guys, what's the capacity going to be in a year or two? Sort of like when broadband hit the Internet, you didn't need speed doublers anymore.

Joe Lonsdale oppose ×2 ▼ on 🤖 Neoclouds are a transitory GPU arbitrage, not a durable business E218 · 2025-03-08 ▶ 57:28

It's really interesting what they did is they locked down the full supply of tons of data centers, tons of the power they needed, tons of the chips. They're very, very thoughtful. I mean, they're effectively, they're traders and they're very economic. And I think they have a lot of the stuff well.