Nscale
Vertically integrated B2B AI infrastructure: owns/operates power + greenfield data centers + NVIDIA GPU superclusters; monetizes via (1) reserved private cloud (multi-year contracted GPU clusters) and (2) consumption/serverless inference. Heavy capex funded by equity + GPU-backed debt.
Nscale spun out of Arkon Energy in May 2024 and raised a ~$155M Series A in Dec 2024, followed by a ~$1.1B Series B in Oct 2025 and the $2B Series C in Mar 2026 (~$6.29B total raised to date incl. GPU-backed debt). Only the $14.6B Series C mark is firm; the Series B valuation is approximate (post-money not consistently disclosed) and the Series A valuation is not public.
Earnings, margins, COGS & capex
Pre-scale, capex-heavy AI-infra builder. The story is contracted backlog, not reported revenue: Nscale has signed very large multi-year GPU-capacity contracts (Microsoft ~200,000 GB300 GPUs across Texas/Norway/UK/Portugal, reported at ~$14B and up to ~$24B in aggregate program value; OpenAI Stargate Norway/UK) but is still building the data centers that will generate the cash. The only public P&L point is a ~$24M operating loss over the 7 months to Dec 2024. Funded by ~$6.29B raised to date (equity + GPU-backed debt).
Revenue trend
Margins
build-out phase; path to positive depends on filling contracted capacity
structurally advantaged by owning power + full stack vs. leasing-model neoclouds, per management
remains negative through the capex super-cycle
COGS structure
Dominated by GPU depreciation, power/electricity, data-center opex, and networking. Vertical integration (owning renewable/low-cost power in Norway/Iceland and building its own DCs) is the claimed cost edge vs. neoclouds that lease colo + power.
Capex
Capex-led model: greenfield 'AI factory' data centers + large NVIDIA GB300 fleets. Program scale in the billions; e.g. ~230MW initial capacity for Stargate Norway (Narvik, targeting ~100k GPUs by end 2026, +290MW expansion), a Texas site (~104k GB300s), plus UK (Loughton), Portugal (Sines), Iceland. Partly debt-financed ($1.4B GPU-backed loan; $790M Norway facility).
Latest earnings
n/a
No public financial guidance. CEO Josh Payne has signaled the company is exploring a 2026 IPO.
- Latest valuation
- ~$14.6B (Series C, Mar 2026)
- Total raised to date
- ~$6.29B since 2023 (equity + GPU-backed debt); $2B Series C led by Aker ASA + 8090 Industries
- Microsoft contract
- ~200,000 NVIDIA GB300 GPUs; ~$14B (up to ~$24B aggregate program value, per CNBC/FT)
- Op loss (7mo to Dec 2024)
- ~$24M
Growth drivers
- Multi-year anchor contracts with Microsoft (~200k GB300 GPUs, up to ~$24B) and OpenAI (Stargate Norway/UK) providing contracted demand ahead of build
- Sovereign-AI tailwind in Europe (UK, Norway) — governments and hyperscalers wanting in-region, low-carbon compute
- Access to cheap, abundant renewable power in the Nordics (hydro) as the scarce input for AI compute
- NVIDIA relationship - preferential GPU allocation + NVIDIA as investor, in a supply-constrained market
- Vertical integration lowering unit cost of compute vs. lease-model competitors
- Ability to raise large equity + GPU-collateralized debt to fund the capex cycle
Bull & bear
Nscale is a pole-position, power-first European AI hyperscaler with signed multi-year contracts from the most creditworthy AI buyers on earth (Microsoft, OpenAI) and NVIDIA in its cap table - buying the scarce inputs (power, land, GPU allocation) before demand fully arrives, with a 2026 IPO as the liquidity catalyst.
- Contracted backlog (Microsoft ~200k GB300s, ~$14B and up to ~$24B; OpenAI Stargate Norway) converts to revenue as sites energize in 2026-2027 - the demand is pre-signed, not speculative
- Vertical integration + cheap Nordic hydro power is a real, durable cost moat vs. lease-model neoclouds; power is the true bottleneck of AI, and Nscale owns it
- NVIDIA as investor + supplier gives preferential GPU allocation in a supply-constrained market - a scarce advantage competitors can't easily replicate
- Sovereign-AI politics in the UK/EU favor an in-region, low-carbon operator; regulatory tailwind, not just commercial
- Demonstrated capital-markets access ($2B Series C, GPU-backed debt) means it can fund the capex super-cycle; a 2026 IPO would re-rate it against public comps CoreWeave (CRWV) and Nebius (NBIS)
A barely-two-year-old, loss-making company has raised billions and taken on GPU-backed debt to build capacity whose revenue is undisclosed, dangerously concentrated in two customers, and already slipping - a classic 'fast to build, slow to monetize' setup exposed to any AI-capex air-pocket.
- No disclosed revenue against a reported operating loss and multi-billion capex - the entire thesis rests on backlog converting on time and at expected margin, which is unproven
- Customer concentration is existential: Microsoft/OpenAI dominate the book, and OpenAI already paused Stargate UK over energy/regulatory costs - one renegotiation reprices the company
- Debt-funded GPU fleet ($1.4B GPU-backed loan + $790M Norway facility) carries depreciation/obsolescence and rate risk; if utilization disappoints, the leverage bites
- Execution is already slipping (Loughton delayed to Q2 2027; the Guardian found no construction at the site) amid 7-13-year grid queues and local power-allocation opposition
- Sits inside the scrutinized NVIDIA-neocloud-hyperscaler circular-financing web; a broader AI-infra oversupply or sentiment reversal would compress private marks and jeopardize the IPO
- The ~$14.6B valuation prices in near-flawless execution of a build-out that has barely begun to generate cash
What it is worth
Private, last-priced-round + public-comp cross-check. Anchor: $14.6B Series C (Mar 2026, $2B raised, led by Aker ASA and 8090 Industries). Cross-checked against public neocloud comps CoreWeave (CRWV) and Nebius (NBIS) on EV/contracted-backlog and forward EV/revenue, given Nscale's own revenue is undisclosed.
Down-round / delayed IPO if Stargate demand softens further, MSFT/OpenAI concentration bites, execution keeps slipping, or an AI-capex air-pocket + circular-financing scrutiny compresses private marks - the ~$14.6B prices in near-flawless execution that hasn't been demonstrated.
Holds ~$14.6B private mark into a 2026 IPO; value tracks backlog conversion and neocloud comp multiples, with concentration/execution risk capping the premium.
IPO/re-rate above $14.6B if sites energize on schedule, utilization is high, and revenue inflects to a CoreWeave-like run-rate while sovereign-AI demand holds - contracted backlog converting cleanly justifies premium neocloud multiples.
No public price - value rests on the last primary round and the size/credit-quality of contracted backlog (Microsoft ~$14B, up to ~$24B; OpenAI Stargate Norway) rather than reported financials. A 2026 IPO would mark it against CRWV/NBIS multiples. The mark is highly sensitive to backlog conversion, utilization, and AI-capex sentiment.
SWOT
Strengths
- Marquee anchor customers (Microsoft, OpenAI) and NVIDIA as both supplier and investor - rare validation for a company barely two years old
- Vertically integrated, power-first model — owns/operates DCs + secures low-cost renewable power (Nordic hydro), a structural cost edge
- Very large contracted backlog signed ahead of build, de-risking initial demand
- Proven ability to raise capital at scale ($2B Series C, largest in European history; plus GPU-backed debt)
- Sovereign-AI positioning in the UK/EU where in-region, low-carbon compute is politically favored
Weaknesses
- No disclosed revenue and a reported operating loss; monetization lags the build
- Extreme customer concentration — Microsoft/OpenAI dominate the book; loss or renegotiation of one contract is material
- Capex-heavy, debt-reliant model exposed to GPU price/depreciation and rate risk
- Very young operating company scaling faster than its execution track record
- Execution slippage already visible (Loughton UK site delayed to Q2 2027; sites still under construction)
Opportunities
- Ride the multi-trillion-dollar AI-infrastructure build-out and sovereign-compute demand across Europe, US, Middle East, Asia
- IPO in 2026 to fund further expansion and provide liquidity
- Expand serverless inference (higher-margin, consumption-based) as an enterprise AI-cloud layer atop raw GPU capacity
- Broaden customer base beyond MSFT/OpenAI (Google already took over some UK capacity) to reduce concentration
- Monetize owned power + land as a durable, scarce asset independent of any single GPU generation
Threats
- 'Fast to build, slow to monetize' - AI-infra oversupply / a demand air-pocket would strand capacity and debt
- Circular-financing dynamics across NVIDIA-neocloud-hyperscaler deals draw scrutiny and could unwind
- OpenAI paused the Stargate UK initiative (energy costs, regulatory uncertainty) in Apr 2026 - anchor-demand fragility
- Grid-connection queues (7-13 years) and local opposition (Norway hydro allocation) delay/limit build-out
- Hyperscalers (Azure, AWS, GCP, Oracle) building their own capacity could insource demand
- GPU depreciation / next-gen obsolescence risk on a debt-funded fleet
Moats, dependencies & bottlenecks
Moats
Owned low-cost renewable power + greenfield DC footprint (Nordic hydro, Iceland) Moderate-strong if power/land locked long-term Power is the binding constraint on AI compute; owning it is the most durable edge Nscale has.
Real advantage while GPUs are scarce; erodes if supply loosens or NVIDIA spreads allocation.
Multi-year contracts are sticky but concentrated; a moat and a single-point-of-failure at once.
Lowers unit cost vs. lease-model neoclouds, but hyperscalers integrate too.
A capability, not a structural moat; contingent on continued risk appetite for AI infra.
Dependencies
Supplier + investor GPU supply, allocation, roadmap and pricing all flow through NVIDIA; also an equity holder, tightening the coupling.
~200k GB300 GPU program (~$14B, up to ~$24B) is the backbone of the book; concentration risk.
Stargate Norway/UK demand; already paused the UK initiative over energy/regulatory costs (Apr 2026).
$1.4B GPU-backed loan + $790M Norway facility; capex model depends on continued cheap leverage.
Grid-connection queues (7-13 yrs) and local hydro-allocation opposition gate the build-out.
Nordic 50/50 JV (Stargate Norway) provides local footprint, power and capital access; also led the Series C.
Advantages
- Power-first, vertically integrated model with owned low-cost renewable energy
- NVIDIA as supplier and investor - scarce GPU allocation
- Blue-chip anchor demand (Microsoft, OpenAI) signed pre-build
- Sovereign-AI positioning in politically favorable UK/EU jurisdictions
- Demonstrated large-scale capital raising (equity + debt)
Weaknesses
- Undisclosed revenue + operating losses; unproven monetization
- Severe customer concentration (MSFT/OpenAI)
- High leverage on a depreciating GPU fleet
- Very short operating history relative to build ambition
- Execution slippage and external (grid/regulatory) dependencies
Bottlenecks
- Grid connection and power availability - multi-year interconnection queues and local opposition
- Data-center construction timelines slipping (Loughton UK now Q2 2027)
- GPU supply/allocation timing from NVIDIA
- Monetization/utilization lag - filling contracted capacity fast enough to service debt
- Skilled DC build/operations talent
Top signals & trends
Top signals
Deep investor conviction and capital access; blue-chip board additions (Sandberg, Decker, Clegg).
Backbone of contracted backlog; also the concentration risk.
Anchor-demand fragility; the flagship-project narrative wobbles.
Execution slipping against an aggressive timeline.
Enables the build but adds leverage/obsolescence risk.
Preferential allocation + validation; deepens dependency.
Liquidity catalyst, but exposes undisclosed financials to public scrutiny.
Trends
Multi-trillion-dollar projected spend underpins demand for GPU capacity.
Regulatory and political tailwind for local, low-carbon operators.
Favors operators that own cheap renewable power - Nscale's core pitch.
Raises questions about demand durability and mark quality.
Overbuild risk if utilization lags the capex.
Azure/AWS/GCP/Oracle building own DCs could absorb demand.
Pressures returns on debt-funded fleets.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GB300 GPUs + networking; also an investor. The critical supplier.
Server/systems integration; investor in the Series C.
Networking; strategic investor in the Series C.
Systems/hardware; Series C participant (HKG:0992 / OTC: LNVGY).
Anchor customer; ~200k GB300 GPU program across Texas/Norway/UK/Portugal.
Stargate Norway/UK workloads (sovereign focus); paused the UK initiative in Apr 2026.
Took over some UK capacity previously allocated to OpenAI.
Closest public pure-play neocloud comparable; US-listed, NVIDIA-backed, similar contracted-backlog + leverage model. The default valuation benchmark for a Nscale IPO.
European-rooted (ex-Yandex) AI neocloud, NVIDIA-backed, directly overlapping European AI-cloud demand.
OCI is a major GPU-cloud supplier to OpenAI/Stargate; competes for the same anchor workloads at hyperscaler scale.
Both Nscale's largest customer and a hyperscaler that could insource capacity - customer and latent competitor.
Hyperscaler GPU capacity (Trainium + NVIDIA); competes for enterprise AI compute.
GCP + TPUs; already took over some UK capacity Nscale had allocated to OpenAI.
US power-first AI data-center developer; similar 'own the power + DC' thesis.
Crypto-miner-turned-AI-cloud with renewable power - a direct analog to Nscale's Arkon origin.
Private US AI-infra / Stargate builder; power-first, NVIDIA-aligned - a direct private peer.
Private US GPU cloud; competes for GPU-cloud workloads (more enterprise/dev than sovereign-scale).