CryptoLite_247

Here are the main takeaways:
1) Demand keeps accelerating
Nebius made it clear that AI infrastructure demand is still materially ahead of supply, and importantly, that visibility is extending further out.
A few months ago, management was talking about roughly 18 months of strong demand visibility. Now Arkady says that has stretched to 24 months or potentially longer, with customers already asking for capacity in Q1 and Q2 2028.
He also said customers are currently requesting tens of thousands of Vera Rubin GPUs.
Marc reinforced the point by saying demand isn't merely above supply, it's actually growing faster than supply.
2) Nebius $NBIS x Palantir explanation $PLTR
The underlying thesis is centered around enterprise data.
Arkady’s view is that companies will increasingly want to run open-weight models on their own proprietary data, repeatedly improve those models inside their own domain, and retain control over the resulting data and intelligence rather than continuously feeding it back into external commercial models.
Palantir already has the enterprise software layer and the tooling required to orchestrate those workflows. What it needed was the infrastructure underneath it. That's where Nebius comes in.
“We provide the whole stack up to our Token Factory. They take it, add their tools on top of it, and their customer base. For us, it is an excellent channel to enterprises.”
This potentially gives Nebius access to an enterprise customer base that would have taken considerably longer to penetrate organically.
3) Enterprise demand appears to be inflecting
The Palantir deal is part of a broader trend. Marc said Nebius has seen a “significant uptick” in its pipeline with platform companies that themselves serve enterprise customers.
Some of these companies are looking at Nebius as their first non-hyperscaler supplier. They're not only looking for training capacity, but a partner that can support training, post-training and inference across the full AI lifecycle.
The customer progression is becoming increasingly clear: AI natives → larger AI natives → digital natives → enterprise platforms → enterprises themselves.
Management still believes enterprise will represent the majority of the market opportunity.
4) xAI and Meta selling excess capacity
Goldman asked what happens if companies like xAI and Meta build huge amounts of infrastructure for themselves and eventually start selling unused capacity into the market.
Arkady’s answer was that this is mostly bare metal capacity, not a full AI cloud. Someone still needs to take that infrastructure, add the cloud and software layers, and distribute it to end customers. Nebius could potentially be one of those companies.
“If they could put more capacity on the market, it’s good for the industry, it’s good for us. We could take it and repackage and resell it to the end users.”
5) The four landmark customers are already moving toward the next clusters
The four large customer wins announced last quarter were not one-off capacity deals. Marc said these customers already have additional requirements and are discussing their next deployments with Nebius.
In some cases they're asking for more GB200 capacity, while all four are already discussing Vera Rubin platform requirements. Nebius also has other similar large opportunities in the pipeline.
Importantly, Marc reiterated that these wins weren't driven by price, but by the reliability and performance Nebius demonstrated during the POCs.
6) Older GPUs are holding up for a reason
This is exactly what I’ve been emphasizing for a long time.
Marc explained that older GPUs aren't simply staying busy because customers cannot get newer chips. Many workloads, including RAG infrastructure, text prediction and image generation, are simply better matched economically to older hardware, where customers already understand the TCO, reliability and performance profile.
“We actually have a list of new customers that are looking for any of the older generation chips that come available.”
That's precisely the dynamic I've been arguing for: the highest-end GPUs should increasingly migrate toward the most compute-intensive workloads, while older generations remain economically relevant for workloads where their performance is already more than sufficient.
7) The asset-light model is progressing
The asset-light strategy is designed to attack what Arkady sees as Nebius’ two main constraints: how fast it can physically build capacity and how much of that build it can finance itself.
The partners Nebius is talking to are typically electricity or data center companies that already have land, power and, in many cases, access to cheap financing, but lack the expertise to move higher up the AI infrastructure stack. Nebius can bring the data center know-how, racks, software stack and the customer demand to monetize that infrastructure.
Management says it has a long list of companies interested in the model, with Arkady describing each line in their internal pipeline as a separate large project. Nebius is already working on several of them, with capacity expected to come online from 2027 onward.