The NeoClouds – Game of Numbers

The numbers don’t add up yet.

• The two best-known NeoClouds reported results this week that demonstrate solid progress, but due to the long-term nature of much of their business, they are not yet making the returns that will drive a recovery in their valuation.
• The market remains very edgy and marked Nebius down 16.8% for a small delay caused by one of its power suppliers failing to obtain its permits and having to switch off its temporary power before it was previously expected to.
• Environments where the market is skittish like this are always where the best opportunities are to be found, which is why CoreWeave and especially, Nebius are worth a closer look.
• Both Nebius and CoreWeave reported results where management discussed the continuing strong environment for compute pricing but failed to show hard evidence of this in their income statements.
• Both companies confirmed RFM’s contention that pricing in the term market for 1 to 5 years is at $20bn -$25bn / GW and the spot market is at $40bn per GW.
• Furthermore, both reported rising prices throughout 2026 and that the level of demand that they have outstrips their ability to supply it in 2027 and most likely 2028 as well.
• With the change of guard that is coming in the US Congress, the scope for more delays and curtailments of data centre build-outs is likely to mean that prices will remain elevated for longer, playing directly into the hands of those who have capacity either online now or nearing the final stages of completion.
• This is Nebius and CoreWeave as pure plays, but also Meta, SpaceX, Google, Amazon and Microsoft.
• At the end of Q3 26, I estimate that Nebius had something like 500MW of active power, while CoreWeave had 1.7GW at the end of Q3 26, with an average of 1.6GW during the quarter itself.
• Based on the term contract price, Nebius should have generated $2.8bn in revenues with very high operating margins on a GAAP basis, even taking into account depreciation.
• By contrast, it reported $981m in revenue and operating margins of 5%
• On the same basis, CoreWeave should have generated $9bn in revenues with fat margins, but instead it reported $3.6bn in revenues and operating margins of 6%.
• This translates into $9bn per GW for CoreWeave and $7.8bn per GW for Nebius, which explains why they are not as profitable as they should be.
• At $25bn per GW both companies should be comfortably into 30% operating margin territory and earning more than 30% annually in terms of return on invested capital.
• I suspect that this is due to contracts that they signed a year or more ago when pricing was much lower, and the shortage was much less acute, and this is going to take time to correct itself.
• I expect that this is going to begin properly in 2027, and it is not until then that the doubts and uncertainties around their business models and their financing are going to be alleviated.
• Both Nebius and CoreWeave have reserved some of their capacity for the spot market, and Nebius has already launched online auctions where clients can bid for capacity.
• Both companies have also moved towards financing their capex bills via customer prepayments, which is an excellent method of financing.
• This is because demand is currently so high that neither company needs to give a discount for prepayment, making this much better than raising debt.
• Nebius expects that 75% – 85% of its 2027 capex to be financed in this way, which should be comforting to skittish investors.
• The net result is that the market remains very nervous about both of these companies and is not prepared to give them any credit for the contracts that they have already signed and will be delivering in the next two years or so.
• It is also not giving them any credit for being beneficiaries of the crazy pricing in the spot market for compute, nor for having capacity to sell should compute become even scarcer as a result of roll-out delays.
• This means that there is a substantial upside opportunity in both of these companies, but I don’t really want to pull the trigger yet, as the upside catalyst is unlikely to occur before 2027.
• This catalyst will be improving revenue per GW and higher margins as new contracts and spot market sales start appearing in the financial statements.
• Furthermore, with Nvidia not being given any credit for gross margins normalising in 2027 or having a life after the sad departure of Jensen Huang, there is as much upside in this as in the neo-clouds with a much lower risk profile.
• This is why I am putting my extra capital into Nvidia for the moment.