What it is
A “neocloud” is a cloud computing company built around a single product: renting out graphics processing units (GPUs) for AI training and inference. Unlike hyperscalers such as AWS, Microsoft Azure, or Google Cloud — which offer GPUs alongside a broad portfolio of storage, databases, and enterprise software — a neocloud’s business is narrower. It buys large quantities of Nvidia or AMD chips, houses them in data centers (often leased rather than owned), and sells access to that compute by the hour or under multi-year contracts.
The term was coined by the chip-industry research firm SemiAnalysis, which published its “AI Neocloud Playbook and Anatomy” report on October 3, 2024, identifying a small group of “Neocloud Giants” — CoreWeave, Crusoe, Lambda, and Nebius — alongside a longer tail of smaller “emerging neoclouds.” CoreWeave, which began as a cryptocurrency-mining operation before pivoting to GPU rental, is generally described as the largest and best-capitalized of the group, with multibillion-dollar contracts with Microsoft, OpenAI, and Meta.
What distinguishes neoclouds economically is leverage. Because GPUs depreciate quickly and cost tens of thousands of dollars each at scale, neoclouds typically finance their hardware purchases with debt — asset-backed loans, special-purpose financing vehicles, and private credit — rather than paying cash. A neocloud’s margin is the spread between what it charges customers per GPU-hour and its costs of ownership: depreciation, power, colocation, and debt service. This structure makes neoclouds unusually sensitive to two things: whether AI labs keep signing long-term compute contracts, and whether credit markets remain willing to lend against chips as collateral.
Why it matters for AI governance and narratives
The neocloud category has become a flashpoint in the broader argument over whether AI capital expenditure is sound investment or bubble financing. Through 2025 the public debate centered on whether end-user demand justified the hyperscalers’ capex; by late 2026, attention had shifted toward the financing structures underneath that spending — who is actually lending the money, and what happens if contracted revenue doesn’t materialize fast enough to service the debt.
This is the shift the observatory’s editorial is tracking. Anthropic’s IPO registration statement is one data point in that shift: reporting on the filing describes roughly $900 billion in future commitments for servers, buildings, and power, financed in part through structures like the reported $63 billion “Project North Star” deal, in which banks and Blackstone fund a vehicle that buys Broadcom chips and leases them back to Anthropic. Commentators have flagged the resemblance to “circular financing” arrangements — where a chipmaker or financier takes a stake in, or extends credit to, a customer that then commits to buying that same financier’s hardware or compute — a structure that has drawn scrutiny for potentially inflating reported demand. Separately, reporting in late October 2026 (including from Fortune and industry newsletters) described tightening credit conditions for neoclouds specifically, with Elon Musk predicting consolidation among weaker players and ratings agencies such as Blackstone’s analysts signaling that even well-funded AI labs may not achieve investment-grade credit soon after going public.
For the observatory’s purposes, the notable pattern is not the existence of the neocloud business model itself — it’s two years old — but how quickly the financing anxiety around it has moved from specialist chip-industry newsletters into mainstream business press and company-filing scrutiny once a marquee AI lab’s IPO disclosures gave journalists a concrete document to read against.
Key facts and dates
- October 3, 2024: SemiAnalysis (researchers Dylan Patel and Daniel Nishball) publishes “AI Neocloud Playbook and Anatomy,” coining the term and naming CoreWeave, Crusoe, Lambda, and Nebius as the leading “Neocloud Giants.”
- 2025–2026: The category expands; industry trackers list dozens of smaller neoclouds (Nscale, Civo, Vultr, and others) competing on price and GPU generation (H100, H200, GB200/300 clusters).
- Late October 2026: Reporting on Anthropic’s IPO registration statement surfaces large forward debt commitments, intensifying scrutiny of financing arrangements between AI labs, chipmakers, and neoclouds.
- October–November 2026: Multiple outlets report tightening credit availability for neoclouds and warn of coming consolidation, alongside concern that some financing structures blur the line between genuine demand and vendor-financed revenue.
I could not independently verify the specific dollar figures attributed to Anthropic’s filing (the ~$900 billion and $63 billion figures) beyond secondary reporting; readers should treat those as reported, not confirmed from the primary filing itself.
Where to learn more
- AI Neocloud Playbook and Anatomy — SemiAnalysis, October 2024 (the originating report that coined and defined the term)
- The neocloud shakeout is coming, and even the neoclouds seem to know it — Fortune, October 29, 2026
- What is circular financing in AI infrastructure? — Capacity Media
- Anthropic IPO Filing Reveals Nearly $900 Billion in Future Debt — Bloomberg Tax