Section A · Defining

Defining 'Neocloud'

'Neocloud' is a category invented in 2023-2024 to describe a then-emerging class of AI-focused GPU clouds. The term is now widely used but its definition has shifted as the category matured.

The term

"Neocloud" emerged as analyst shorthand for the wave of GPU clouds that grew up alongside the LLM boom — companies whose product was GPU compute specifically for AI / ML workloads, distinct from both the hyperscalers (AWS, Azure, GCP) and from generic cloud providers.

The term has stuck partly because it captures something real: these companies share strategic features that distinguish them from older cloud categories. They:

  • Focus on AI / ML workloads, not general-purpose compute.
  • Specialize in GPU rental, often at the expense of broader cloud services.
  • Have business models structured around the AI demand surge.
  • Often rely on NVIDIA strategic relationships.
  • Operate at different scales than either hyperscalers or traditional hosting providers.

Origin of the term

The earliest popular uses of "neocloud" appeared in analyst commentary and infrastructure-focused reporting around 2023. By 2024, the term was being used in earnings calls, investor presentations, and competitive comparisons.

The neocloud-vs-hyperscaler framing was particularly useful for investors trying to understand whether companies like CoreWeave were competing with or supplying to the established cloud giants.

Who counts as a neocloud

The boundaries are fuzzy. The clearest cases:

  • CoreWeave, Crusoe, Lambda, Nebius — universally counted.
  • Together.AI, Hyperbolic — sometimes counted, sometimes treated as managed-inference platforms.
  • RunPod, TensorDock, Vast — sometimes counted as marketplace neoclouds, sometimes treated as a separate category.
  • Hyperscalers (AWS / Azure / GCP) — not counted, although they offer comparable products.
  • Traditional cloud providers (DigitalOcean, OVH, etc.) — not counted unless AI-specialized.

The boundary that probably matters most: AI-focused intent. Companies whose product strategy and customer focus are explicitly about AI / ML workloads count; companies whose AI offering is one of many products don't.

ClusterMAX framework

SemiAnalysis published the ClusterMAX rating system to formalize neocloud comparison. ClusterMAX 2.0 (2024-2026) categorizes providers into tiers based on:

  • Cluster fabric quality.
  • Operational maturity.
  • NVIDIA reference design adherence.
  • Customer commitments and track record.
  • Specific infrastructure features.

ClusterMAX is the closest thing the industry has to a standardized rating. It's not perfect but provides a common vocabulary for discussing tiers.

Tier emergence

The 2024-2025 tier hierarchy:

  • Platinum: CoreWeave (sole occupant in ClusterMAX 2.0).
  • Gold: Crusoe, Nebius, Lambda, others operating at scale with proven cluster fabric.
  • Silver: Smaller dedicated neoclouds.
  • Below tier: Marketplaces and specialty platforms (often outside ClusterMAX framework entirely).

The tier hierarchy isn't static. Companies move up or down as their infrastructure and operational sophistication evolves.

Takeaway

The neocloud category is a 2023-2026 phenomenon — a real strategic class of company with shared features but fuzzy boundaries. The next chapter traces the pre-AI history.