Key Takeaways
- The competitive advantage for neocloud providers is not secured through hardware acquisition or energy efficiency, but through the sophisticated software and service layers built atop the physical infrastructure.
- Differentiation stems from proprietary orchestration tools that optimise GPU workloads, bespoke developer environments that accelerate AI model deployment, and expert-led operational support.
- While hyperscalers like AWS and Azure compete on scale, neoclouds compete on specialisation, offering performance and customisation for intensive AI tasks that generalist platforms cannot easily match.
- The primary long-term risk is the commoditisation of compute, making a sticky software ecosystem the most critical defence and the foundation of a durable business model.
- Valuations for leading neoclouds should be assessed using software-as-a-service (SaaS) metrics, focusing on revenue quality and margins, rather than treating them as low-margin hardware resellers.
The prevailing narrative surrounding the ‘neocloud’ sector often frames it as a simple arbitrage play: acquiring scarce GPUs and reselling their compute cycles in the most energy-efficient manner possible. This view, however, misses the fundamental value proposition and the actual basis of competition. As observed by analyst M. V. Cunha, the true differentiation from an end-user’s perspective emerges not from the silicon itself, but from the innovation occurring in the software and service strata above the hardware. This shifts the investment thesis away from a pure infrastructure story towards one centred on specialised platforms, operational expertise, and defensible software moats.
Hardware as the Ante, Not the Ace
In the high-stakes game of artificial intelligence infrastructure, securing a sufficient allocation of NVIDIA’s latest GPUs is merely the cost of entry—the ante required to sit at the table. It is not, however, the winning hand. While initial hardware scarcity created a temporary moat for early movers, the stabilising supply chain is eroding this advantage. The competitive battleground is therefore moving up the stack, away from procurement and towards operational excellence.1
Metrics such as Power Usage Effectiveness (PUE), once a key marketing point, are now considered table stakes. Achieving a PUE below 1.2 is the expected baseline for any serious operator, not a mark of distinction.2 The more pertinent questions clients ask relate to performance and workload optimisation. Can a provider architect a bare-metal environment that minimises latency for inference tasks? Can they guarantee sustained throughput for training colossal large language models? The answers lie less in the data centre’s cooling system and more in the intelligence of its management layer.
The Software and Services Moat
The most successful neoclouds operate less like infrastructure resellers and more like specialised Platform-as-a-Service (PaaS) providers for AI development. Their value is created through proprietary software that addresses the unique challenges of GPU-accelerated computing. This includes sophisticated orchestration engines that can dynamically schedule and prioritise workloads, ensuring that eye-wateringly expensive hardware is never idle. For clients, this translates directly into a higher return on their compute spend.
Beyond resource management, these platforms offer a superior developer experience. This can include pre-configured software environments, managed Kubernetes services tailored for GPUs, and APIs that simplify the deployment of complex machine learning pipelines. By abstracting away the underlying infrastructure complexities, they enable data science teams to focus on building models, not managing servers. This acceleration of the development lifecycle is a powerful selling point that generalist cloud providers often struggle to replicate at the same level of performance and support.3
| Feature | Hyperscaler (e.g., AWS, GCP, Azure) | Specialist Neocloud |
|---|---|---|
| Core Offering | Broad portfolio of hundreds of services; a “one-stop shop”. | Highly specialised, performance-tuned compute for AI/ML. |
| Performance | Generalised, often involves performance trade-offs due to virtualisation overhead. | Optimised for maximum throughput on bare-metal or near-bare-metal setups. |
| Cost Structure | Complex pricing with significant egress fees and charges for ancillary services. | Often simpler, more transparent pricing focused on compute time. |
| Support | Tiered support model; deep technical expertise requires premium plans. | Direct access to engineers and AI infrastructure specialists. |
Navigating the Threat of Commoditisation
The most significant long-term risk facing the neocloud sector is commoditisation. As GPU supply becomes more readily available and hyperscalers improve their own AI offerings, the space could become crowded, leading to intense price competition and margin erosion.4 If a neocloud’s only value is providing raw compute at a slight discount to AWS, its business model is indefensible.
The strategic imperative, therefore, is to build a sticky ecosystem that locks in customers through software and service, not just price. By becoming deeply integrated into a client’s development workflow, a neocloud can make the cost of switching prohibitively high. This is the classic SaaS playbook, applied to the world of high-performance computing. The goal is to evolve from being a utility provider into an indispensable technology partner.
Conclusion: More PaaS than IaaS
For investors, analysing a neocloud requires a new lens. To value a company like CoreWeave—which raised $7.5 billion at a $19 billion valuation in May 2024—as a simple hardware reseller would be a profound misjudgement.5 Such valuations are predicated on a business model that exhibits software-like characteristics: high gross margins (after accounting for power and depreciation), strong net revenue retention, and a defensible technological edge. The neoclouds that succeed will not be the ones with the most GPUs, but the ones with the most intelligent software managing them.
A speculative but logical hypothesis for the sector’s evolution is a move towards vertical specialisation. Rather than competing head-on for every AI workload, the most durable neoclouds may carve out niches by becoming the undisputed platform for specific industries, such as drug discovery, financial modelling, or autonomous vehicle simulation. By tailoring their software stack and regulatory compliance to the unique needs of a single vertical, they can create moats that even the largest hyperscalers will find difficult and uneconomical to breach.
References
- SemiAnalysis. (2023). AI Neocloud Playbook and Anatomy. Retrieved from SemiAnalysis.
- The Fast Mode. (2024). Rethinking Data Center Cooling: Meeting the Demands of AI, Edge, and High Performance Computing. Retrieved from The Fast Mode.
- DriveNets. (n.d.). What are Neocloud Providers? Retrieved from DriveNets Education Center.
- Futuriom. (2024). Could Neoclouds Become Commoditized? Retrieved from Futuriom.
- Bloomberg. (2024, May 1). CoreWeave Raises $7.5 Billion in Debt Deal Led by Blackstone. Retrieved from Bloomberg News.
- @mvcinvesting. (2024, August 1). [The real differentiation, from the end-user’s perspective, comes from what happens above the hardware layer]. Retrieved from https://x.com/mvcinvesting/status/1920491727724445791
