The Honeymoon Phase of Capacity

Although I am yet to write any Deep Dive on any neocloud, I still like to pay close attention to what they are saying during their earnings calls. Both CoreWeave and Nebius had some interesting nuggets in their 1Q’26 call.

Let’s start with CoreWeave.

Just like the hyperscalers, CoreWeave’s backlog continues to grow at a rapid pace. Their backlog now reached almost $100 Billion in 1Q’26 (vs just $25.9 Billion in 1Q’25). While their backlog likely remains concentrated, they did mention that they now have 10 customers committed to spending at least $1 Billion with CoreWeave.

CoreWeave indicated that “materially in excess of 50%” of their compute is currently being used for inference. They slightly raised the low end of the capex guide from $30-35 Billion to $31-35 Billion for 2026. The increase on the low end is primarily due to memory price increases.

What is Nebius saying?

Nebius raised their capex guide too, but unlike CoreWeave, they mentioned the impetus for capex increase was primarily higher demand. From the call:

Demand is high. Everything we build with is sold. That is what is driving us to build more and to raise our 2026 CapEx guidance to between $20 billion and $25 billion, which is up from our prior range of $16 billion to $20 billion.

Perhaps the more interesting tidbit from the Nebius call is how Nebius management elaborated on their deal with Meta Platforms (emphasis mine):

“First, I want to say that we love working with Meta, and we’re excited that they chose to buy more capacity from us. This expanded new agreement is to make sure that we all understand this, a 5-year contract for a total of $27 billion, and it is structured in 2 parts. First, there’s a $12 billion commitment to dedicated compute capacity with delivery starting in early ‘27. And then second, as you pointed out, there’s another $15 billion of additional capacity that we, at our discretion, can either allocate to Meta or sell to our AI cloud customers as it comes online for the duration of the 5-year contract.

Let me explain this in a bit more detail. Meta is committed to buy up to $15 billion of any capacity in these clusters at our option during the entire 5-year contract. This commitment will likely allow us to finance the clusters with asset-based -- asset-backed financing at attractive terms, while selling them to, as I think you pointed out, to our AI cloud customers at potentially higher market prices.

The unique combination of being able to sell at a premium, along with the commitment by Meta to purchase any capacity during the contract should provide us with higher margins, less risk and more visibility in our revenue. If the market remains strong, we should generate more than $27 billion in revenue from this great agreement.”

Meta has effectively written Nebius a put on $15 Billion of capacity. If the market stays hot and Nebius can sell to higher-margin customers, Meta may get none of that $15 Billion capacity. If the market softens and Nebius can't place it at premium prices, Nebius puts it to Meta at the contracted price. Now what is Meta’s incentive for being so charitable here?

We cannot be sure for Meta’s motivations but we can make some pretty good guesses. The most likely explanation is Meta is getting the first $12 Billion tranche of capacity at a noticeably below current market rates. But the overall structure of the deal makes me think Meta may be fairly confident of generating decent return on the additional $15 Billion capacity at current contracted rates in case Nebius is unable to sell capacity to customers at more attractive prices.

Overall, both companies sound as good as you would expect in an environment compute demand continues to outstrip supply. The real question, of course, is whether these business will prove to be durable or are these just “stop-gap” solutions for the major hyperscalers. Gavin Baker in his recent Sohn interview made the case that some of these neoclouds will be durable business model. Here’s an excerpt from Gavin Baker’s interview:

…the hyperscalers for a long time were stuck in a cost mentality. The hyperscalers were competing with people running these Formula 1 cars and they were like, you know, doing like overnight shifts and 18-wheelers trying to stay awake, deliver the lowest cost. And that's not what AI is about. Now, I thin they're making this mental and cultural shift and they've made it, but I think some of these Neoclouds have a very durable business model.

Baker also pointed out that Amazon’s Trainium is likely to be very underestimated by the market. As I have alluded before, in a compute constrained environment, the market signals can get so weak and hazy that makes it very challenging to assess the longer term competitive dynamics. In current environment, everyone’s capacity is getting sold out and if you have a chip, give some of the hyperscalers a call because they may buy whatever you’re selling. I have laid out more qualitative concerns (see here, and here) before about the long-term economics of 3P hyperscalers. I will highlight a few data points behind the paywall that continue to keep me unenthusiastic about 3P hyperscaler business model even though investors are currently showing zero inclination of concern right now.

Subscribers get the daily journal and five+ years of Deep Dives, i.e. full-length analyses with financial models on 65+ companies. The daily is just how I think out loud between the Deep Dives!