We delivered $529.8 M of revenue for 2025, up roughly 480%, and our core Nebius cloud segment turned adjusted‑EBITDA‑positive at about +$59 M, from a $(128.5) M loss the year before. The group’s adjusted EBITDA remained negative at about $(64.9) M, driven by Avride (~$(82.7) M) and TripleTen (~$(41.2) M). Adjusted EBITDA excludes GPU‑fleet depreciation, so the core is still loss‑making on a bottom‑line basis; Q1 2026 net loss was $(113.5) M. Still the very early days.
We're building an AI-native hyperscaler, and I think about it in four dimensions - capacity, product, demand, and capital. On the first three, everything we build is already sold: last quarter that was $399 million of revenue, up 684%, with the core cloud finally adjusted-EBITDA-positive even as the group still runs a net loss. The fourth dimension is where the honesty sits. Our 2026 build is guided at $20 to $25 billion against roughly $3 billion of revenue - seven to eight times over, spent years ahead of the demand it's meant to serve; our own filing calls the core business capital-intensive and not yet profitable. What I'd rather own than rent is the whole stack, and what I'd rather not do is over-borrow to get there - so we came in with about $9.3 billion of cash and cheaper convertible debt, not loans secured against the chips. Still the very early days.
On the product dimension, everything we build is already sold: 2025 revenue $529.8M up ~480%, core cloud adjusted-EBITDA-positive at ~$59M from $(128.5)M loss. Group adjusted EBITDA negative ~$(64.9)M, dragged by Avride ~$(82.7)M and TripleTen ~$(41.2)M. Adjusted EBITDA excludes GPU depreciation, so core still loss-making bottom-line; Q1 2026 net loss $(113.5)M. Still the very early days.
We posted $529.8M revenue for 2025, up ~480%. Core cloud turned adjusted-EBITDA-positive at ~$59M from a $(128.5)M loss. Group adjusted EBITDA remained negative at ~$(64.9)M, pulled down by Avride (~$(82.7)M) and TripleTen (~$(41.2)M); Toloka was deconsolidated mid-2025. Adjusted EBITDA excludes GPU depreciation, so core stays loss-making on a bottom-line basis; Q1 2026 net loss was $(113.5)M. The AI-cloud story reads stronger than consolidated numbers, but group losses are real. Still the very early days.
Nebius entered its build with a large liquidity buffer: about $9.3 billion of cash at March 31 2026, up from $3.7 billion at year‑end, built from early‑2026 financings. Those financings included roughly $2 billion from NVIDIA—recorded as proceeds plus a pre‑funded Class A warrant of about 21 million shares—and convertible debt. The CFO says incremental capacity will be funded through further financing, flagging asset‑backed debt against the Microsoft contract, so the model is debt raised against contracted demand, carried for now on the cash cushion. Still the very early days.
Meta has committed to take up to $15 billion of capacity, at our option, over a five‑year contract; a separate five‑year agreement framed at more than $27 billion is also referenced, and which anchor it belongs to is still being confirmed against the filing. An outside read puts Microsoft and Meta together at more than 90 % of backlog, but no exact split is disclosed and the precise share remains an open question until the next annual filing. Still the very early days.
On the capital dimension: we entered the build with a large liquidity buffer — about $9.3B cash at March 31 2026, up from $3.7B at year-end, from early-2026 financings. That included roughly $2B from NVIDIA (proceeds plus a pre-funded warrant) alongside convertible debt; the supplier-funds-customer circularity is in our filing. For the incremental capacity in our raised guidance, the CFO flags further financing — specifically asset-backed against the Microsoft contract. So the model is debt raised against contracted demand, carried for now on the cash cushion. Still the very early days.
On the demand dimension: Meta committed to take up to $15B of capacity at our option over five years; a separate five-year agreement framed at more than $27B is also referenced, and which of the two anchors it belongs to is still being confirmed against the filing. An outside read puts Microsoft and Meta together at more than 90% of backlog - that figure is an inference, no exact split disclosed. The precise share remains an open question until the next annual filing. Everything we build is already sold. Still the very early days.
On the capital dimension: we raised 2026 capex guidance to $20-25B from $16-20B — seven to eight times the roughly $3B revenue we guide. Q1 capex was $2.47B; property and equipment grew to $7.13B with $212M D&A. We frame the increase as funding 2027 capacity coming online early next year, expected to contribute revenue in H1 2027 against customer commitments already in place. Our filing states the core business is capital-intensive and not yet profitable; the spend deploys years ahead of the revenue it's meant to generate. Still the very early days.
A reading of the quarterly filing shows a net loss of about $(113.5) million. A separate read of the same quarter shows positive net income near $621 million, flattered by a roughly $780.6 million one-time, non-cash gain from revaluing a ClickHouse stake — stripping that out leaves an adjusted loss near $(100.3) million. What the reported net-income line includes or excludes is a to-verify item for the next filing. Either way the shape is the same: core cloud adjusted-EBITDA-positive inside a group that still loses money once GPU-fleet depreciation is counted. Still the very early days.
We delivered $529.8M revenue for 2025, up ~480%. Core cloud turned adjusted-EBITDA-positive at ~$59M, from a $(128.5)M loss. Group adjusted EBITDA stayed negative at ~$(64.9)M, dragged by Avride (~$(82.7)M) and TripleTen (~$(41.2)M). Adjusted EBITDA excludes GPU depreciation, so core remains loss-making on a bottom-line basis; Q1 2026 group net loss was $(113.5)M. Still the very early days.