Research

Neocloud Business Model Advisory

Capital deployment data. Operator consolidation. Unit economics.

[01]

Capital Deployment & Market Consolidation

Neocloud operators collectively raised and deployed $60B+ in capital across 2024-2025. The operator landscape contracted from 150+ viable platforms in early 2024 to approximately 40-50 sustainability-viable operators by Q1 2026.

This consolidation reflects the harsh reality of capex leverage: any operator missing scale targets faced immediate cash-flow pressure as depreciation consumed operational margin. Winners accumulated market share through three mechanics: (1) hyperscaler partnerships providing preferential hardware allocation and customer leads, (2) geographic diversification reducing regional pricing pressure, (3) vertical integration capturing both infrastructure and applications revenue. Smaller regional operators (sub-100-exaflop capacity) found themselves unable to amortise infrastructure capex across sufficient customer base, forcing either exit, acquisition, or pivot to vertical specialisation.

[02]

Pricing Dynamics & Margin Compression

On-demand GPU pricing declined approximately 40% from peak (November 2024) by March 2026, driven by supply expansion and softening demand from model-training applications reaching saturation. H100 pricing fell from $3.40/hour to $2.10/hour; Blackwell on-demand ranges $4.80-$7.50/hr depending on provider and commitment length (pricing data current as of Q1 2026; verify current rates with providers directly). The market increasingly bifurcates between committed workloads on longer-term contracts and opportunistic training negotiating flexible on-demand access.

Infrastructure-only gross margins compressed to 35-50% by Q1 2026 from 62-75% in 2023, reflecting pricing compression and underutilisation penalties from customer churn. Managed services (custom orchestration, optimization, support) maintain 55-70% gross margins but require significant engineering headcount, limiting scalability. Profitability now divides cleanly: operators with >70% utilisation achieve 12-18% net margin; those below 65% face negative or single-digit margins.

[03]

Customer Cohort Dynamics & Retention

Neocloud customer base segments distinctly: research institutions and startups (15-20% of capacity, price-sensitive, high churn); established AI labs at tech firms (40-45% of capacity, committed long-term); inference clusters for commercial applications (35-45% of capacity, growing, locked via custom integration). Monthly customer churn at opportunistic-tier operators reaches 35-48%, necessitating constant acquisition spending at approximately $8,400-$14,200 per retained customer.

Sticky customers (research groups with multi-month projects, production inference workloads) show 8-12 month retention, yielding lifetime value of $34,000-$78,000 per customer. Operators with substantial captive workloads (internal applications, partnerships) realise 72-78% annual retention versus market average of 38-42% retention. Contract stickiness inversely correlates with price sensitivity: customers locked into integration (Kubernetes, data pipeline custom builds) accept 5-12% annual price increases without defection.

[04]

Service Bundling & Revenue Diversification

Pure infrastructure margin compression drove operators toward bundled services generating incremental 18-28% revenue. Storage, networking, and egress bandwidth services attached to GPU infrastructure add 8-12% to gross revenue with 68-74% gross margin.

Managed orchestration platforms (Kubernetes, Ray, distributed training frameworks) command 12-18% premium over bare infrastructure with 62-71% gross margin. Data egress and cross-zone networking cost operators $0.03-$0.08 per gigabyte but command list pricing of $0.12-$0.20 per gigabyte, creating 58-75% gross margin on network services.

The most successful operators generate 55-68% revenue from infrastructure, 18-24% from managed services, and 14-22% from advanced services. Pure infrastructure players remain stuck at 92-98% infrastructure revenue. This diversification sustains operator viability as raw GPU pricing compresses further.

Key Takeaways
01

Neocloud raised $60B+ 2024-2025; operator base consolidated from 150+ to 40-50 sustainability-viable platforms

02

On-demand pricing declined ~40% from peak; H100 fell to $2.10/hour, Blackwell ranges $4.80-$7.50/hr depending on provider and commitment length (on-demand; pricing data current as of Q1 2026;verify current rates with providers directly)

03

Infrastructure gross margins compressed to 35-50% by Q1 2026; profitable operators maintain >70% utilisation generating 12-18% net margin

04

Diversified operators mix 55-68% infrastructure revenue with managed services (18-24%) and advanced services (14-22%) to sustain viability

Next Steps

This analysis is produced by Disintermediate, drawing on data from The GPU intelligence platform - tracking 2,800+ companies across 72 categories, real-time GPU pricing from 70+ providers, and advisory engagement experience across the GPU infrastructure value chain.