Q3-2025 Earnings 

While the revenue and earnings came inline with estimates, the excitement was really in the ARR guidance for 2026, and the announcement of a new deal with Meta Platforms.

Here are the salient points and key takeaways:

• Deal signed with Meta for $3Bn over 5 years, it could have been larger except for capacity constraints.
• 2025 ARR $900Mn to 1.1Bn, 2025 Revenue $525Mn
• Forecast - $7-$9Bn in ARR for 2026. This is a huge number and looks interesting from the diversity of the client base. 2026 Revenue should come in around $3.2Bn to 4$Bn given the ramp of the Microsoft $17Bn and Meta Platforms $3Bn ramp starting in 2H of 2026.
• The first delivery of the $17Bn Microsoft was confirmed, indicating that the New Jersey facility in Vineland is on schedule.
• 25 MN shares ATM listing to ensure that capital is raised for 2026-2027.
  • Additional forms of capital raise will include corporate debt and asset backed financing.
• Connected power 220MW 2025, 800MW to 1GW 2026
• Contracted power 2.5GW 2026
• $4.3Bn of financing completed in Sept 2023, at very attractive terms and minimal dilution.
• Full capacity sold out, with launches in UK and Israel in November and October, respectively.
• Capex of $5Bn in 2026
• Depreciation of GPUs 4 years, a conservative approach, faster than the industry average of 5-6 years as pointed by Michael Burry.

Good execution

Nebius' execution has been good to great so far, and it has successfully raised capital to fund all these projects. In my opinion the funding terms have been very good, and the dilution for existing shareholders manageable. The tenor and mood of management was extremely optimistic, a contagious optimism reflecting that their inability to say no to customers, and the CRO gushing that he didn't have to work to market Nebius, instead learn how to manage customers expectations.

De risking customer concentration

Nebius has two mega deals with hyperscalers. Assuming the $19Bn signed with Microsoft through 2030, that would be $3.6Bn for Microsoft per year and $3Bn signed with Meta over 5 years or $600Mn per year, which totals $4.2Bn of ARR. Given the mid-point of $8Bn of management's forecast of $7-$9Bn in ARR, that leaves about $8Bn - 4.2Mn or 3.8Bn, almost 50% from non-hyperscalers, which is great. That is key to growing the market and Nebius showed growth from clients like Shopify, Black Forest Labs, World Labs and Cursor. They also spoke of increased interest from non hyperscalers for compute power from startups, enterprise software customers and platforms like Shopify, widening the market. It has slightly less customer concentration than Coreweave, which has 35% from one customer and about 25% from other hyperscalers.

Challenges

Coreweave dropped significantly because one of its suppliers is behind schedule, so capacity is a problem, and Nebius could have these problems as well. For example, Meta could have taken more compute, but Nebius could only promise and deliver the $3Bn

Investors will have to deal with lumpy revenue and earnings given the tight capacity constraints for power, GPU, and should be prepared for a bumpy ride, like Coreweave.

Interestingly Nebius also confirmed great pricing for even the Hoppers, which is surprising. Michael Burry had made the point that hyperscalers weren’t depreciating their GPUs fast enough and overstating profits - increasing their supposed life from 3 to 6 years. Burry does his homework and then some, so do expect this to resurface whether clouds and neoclouds change their depreciation methods or not. This discussion will not go away, and in my opinion it should not.

Nebius remains a strong buy:

Nebius has a lot of good things going for it and I will be adding on declines. I'll post another article outlining its strengths and competitive advantages, and financial forecasts highlighting why it remains a strong buy, in just its first innings.