AI supply-chain thesis — mapping bottlenecks, focus companies, and supply-chain exposure for investors.
# Nebius Stock ($NBIS): The Vertically-Integrated Neocloud Scaling Into the GPU Shortage **Nebius ($NBIS) is a pure-play AI cloud that builds and operates its own NVIDIA GPU data centers and rents that compute to AI labs and enterprises.** Unlike an asset-light reseller, it owns the buildings, the racks, and the power deals. Spun out of the former Yandex group — it divested its Russian operations for ~$5.4B in July 2024, renamed to Nebius Group N.V., and resumed Nasdaq trading as $NBIS in October 2024 after a halt since early 2022 — and now carrying a ~$50B market cap, it is one of the few independent "neoclouds" with the capital, the NVIDIA relationship, and the multi-region pipeline to scale into the compute shortage rather than merely arbitrage it. ## The hardware A current-generation NVIDIA fleet on Nebius-designed racks: $NVDA H100/H200, HGX B200, GB200 Grace Blackwell, and Blackwell Ultra, with Vera Rubin capacity earmarked for future clusters. $AMD MI300X is evaluated as a secondary source to cut single-vendor risk. Host CPUs are $INTC Xeon (Sapphire/Emerald Rapids); racks come from $SMCI and Gigabyte; the silicon is fabbed by $TSM. Clusters run non-blocking NVIDIA InfiniBand with $ANET Ethernet and $JNPR routing around them, $MU memory, $PSTG all-flash storage, and $VRT plus $SBGSY power and cooling. ## The data-center footprint Operating and announced sites span **Finland** (Mäntsälä flagship ~75 MW, plus a 310 MW Lappeenranta factory), the **US** (Kansas City with Patmos, Vineland NJ 300 MW with DataOne, and a gigawatt-scale Independence, MO site), the **UK** (Ark Data Centres' Longcross Park, Surrey — an initial 4,000 Blackwell Ultra GPUs), **France** (Paris at Equinix PA10 and a 240 MW Lille factory), and **Iceland** (Keflavík, renewable). Owning the footprint is the structural difference from $CRWV-style resellers and from leasing-only peers. ## The demand side The backlog funds the capex. Anchor contracts reported in the tens of billions lead with Meta ($META) — up to ~$27B over five years (~$12B tied to NVIDIA Vera Rubin plus ~$15B flexible) — and a separate Microsoft ($MSFT) agreement reported around $17.4B. Enterprise and AI-lab customers include $SHOP, $NET, $PRX.AS, $PLTR (stack partner), Mistral, and a long tail of generative-AI startups on B200 clusters. Nebius also owns Toloka (data labeling), Avride (autonomous driving), and ~28% of ClickHouse. ## Risks - **Capex intensity** — gigawatt builds cost billions up front, funded with debt and equity; dilution and financing risk are part of the story. - **Customer concentration** — a few anchor deals carry much of the backlog. - **Competition & GPU allocation** — competes with $CRWV and the hyperscalers' own clouds; NVIDIA allocation gates growth. - **Legacy overhang** — the reported income statement still carries the old Yandex history; judge the new cloud business and the contract backlog, not the old consolidated numbers. ## How to play it $NBIS is the European-rooted, vertically-integrated neocloud in the [AI Cloud / Neoclouds](/baskets/ai-cloud-neoclouds) basket; $CRWV is the larger US pure-play, $IREN the miner-turned-HPC operator. The names are gated by megawatts, so the trade is tied to the data-center power and nuclear build-out as much as to GPUs. For the full deep-dive — the data-center footprint on an interactive map, the hardware stack, the anchor-contract backlog, and the risks — see the [Nebius ($NBIS) neocloud breakdown](/blog/nebius-stock). *This is not financial advice — for research and education. Neocloud equities are volatile and capital-intensive.*
# Nebius Stock ($NBIS): The Vertically-Integrated Neocloud Scaling Into the GPU Shortage **Nebius ($NBIS) is a pure-play AI cloud that builds and operates its own NVIDIA GPU data centers and rents that compute to AI labs and enterprises.** Unlike an asset-light reseller, it owns the buildings, the racks, and the power deals. Spun out of the former Yandex group — it divested its Russian operations for ~$5.4B in July 2024, renamed to Nebius Group N.V., and resumed Nasdaq trading as $NBIS in October 2024 after a halt since early 2022 — and now carrying a ~$50B market cap, it is one of the few independent "neoclouds" with the capital, the NVIDIA relationship, and the multi-region pipeline to scale into the compute shortage rather than merely arbitrage it. ## The hardware A current-generation NVIDIA fleet on Nebius-designed racks: $NVDA H100/H200, HGX B200, GB200 Grace Blackwell, and Blackwell Ultra, with Vera Rubin capacity earmarked for future clusters. $AMD MI300X is evaluated as a secondary source to cut single-vendor risk. Host CPUs are $INTC Xeon (Sapphire/Emerald Rapids); racks come from $SMCI and Gigabyte; the silicon is fabbed by $TSM. Clusters run non-blocking NVIDIA InfiniBand with $ANET Ethernet and $JNPR routing around them, $MU memory, $PSTG all-flash storage, and $VRT plus $SBGSY power and cooling. ## The data-center footprint Operating and announced sites span **Finland** (Mäntsälä flagship ~75 MW, plus a 310 MW Lappeenranta factory), the **US** (Kansas City with Patmos, Vineland NJ 300 MW with DataOne, and a gigawatt-scale Independence, MO site), the **UK** (Ark Data Centres' Longcross Park, Surrey — an initial 4,000 Blackwell Ultra GPUs), **France** (Paris at Equinix PA10 and a 240 MW Lille factory), and **Iceland** (Keflavík, renewable). Owning the footprint is the structural difference from $CRWV-style resellers and from leasing-only peers. ## The demand side The backlog funds the capex. Anchor contracts reported in the tens of billions lead with Meta ($META) — up to ~$27B over five years (~$12B tied to NVIDIA Vera Rubin plus ~$15B flexible) — and a separate Microsoft ($MSFT) agreement reported around $17.4B. Enterprise and AI-lab customers include $SHOP, $NET, $PRX.AS, $PLTR (stack partner), Mistral, and a long tail of generative-AI startups on B200 clusters. Nebius also owns Toloka (data labeling), Avride (autonomous driving), and ~28% of ClickHouse. ## Risks - **Capex intensity** — gigawatt builds cost billions up front, funded with debt and equity; dilution and financing risk are part of the story. - **Customer concentration** — a few anchor deals carry much of the backlog. - **Competition & GPU allocation** — competes with $CRWV and the hyperscalers' own clouds; NVIDIA allocation gates growth. - **Legacy overhang** — the reported income statement still carries the old Yandex history; judge the new cloud business and the contract backlog, not the old consolidated numbers. ## How to play it $NBIS is the European-rooted, vertically-integrated neocloud in the [AI Cloud / Neoclouds](/baskets/ai-cloud-neoclouds) basket; $CRWV is the larger US pure-play, $IREN the miner-turned-HPC operator. The names are gated by megawatts, so the trade is tied to the data-center power and nuclear build-out as much as to GPUs. For the full deep-dive — the data-center footprint on an interactive map, the hardware stack, the anchor-contract backlog, and the risks — see the [Nebius ($NBIS) neocloud breakdown](/blog/nebius-stock). *This is not financial advice — for research and education. Neocloud equities are volatile and capital-intensive.*
The Nebius Stock ($NBIS): The Vertically-Integrated Neocloud Scaling Into the GPU Shortage thesis on Macroplane focuses on Nebius Group N.V. (NBIS).
It covers Semiconductor Wafer Fabrication, AI GPUs, Memory Supercycle, Data Center Servers, Data Center Infrastructure, Cooling, Storage, AI Cloud / Neoclouds, Hyperscalers, Enterprise Software.
Browse all AI supply-chain theses · Macro trends · Industries · Product categories · Baskets