Methodology

Where the data comes from

We read each company’s own pricing page, docs, SEC filings, investor releases and press releases. Ratings come from SemiAnalysis ClusterMAX. Every record lists its sources with the date we read them, and the small number beside a figure opens the source behind it. A figure with no public source shows as a blank cell and “Not published” on its page, so every number you see has a source.

How prices are read

• On-demand list price for one GPU, in US dollars per GPU-hour. H100 means SXM unless the company page says otherwise; Lightning AI and OVHcloud list H100 without naming the form factor.
• Multi-GPU instances are divided down. CoreWeave sells H100 only as 8-GPU instances at $49.24 per hour, which we show as $6.16 per GPU-hour.
• Every price carries its date. Prices change: Lambda’s $4.89 comes from the price list taking effect 16 Nov 2026, and Crusoe’s H100 rises from $3.90 to $4.50 on 1 Dec 2026. Both rows are re-checked after those dates.
• What stays out of the price column: quote-only prices (Fluidstack, TensorWave, Nscale, IREN, Firmus, Radiant and others), host-set marketplace prices (Vast.ai, TensorDock), “as low as” term prices (Cirrascale) and prices published only in another currency (Scaleway in EUR, Yotta in INR). Where one of these is published, it is spelled out under “Pricing detail” on the company’s page.
• Reserved, spot and committed-use rates are often lower. Where a company publishes them, they are in the pricing detail too, for example DigitalOcean’s H100 at $3.26 on a 12-month term against $4.41 on demand.

How ClusterMAX ratings are cited

ClusterMAX™ is a rating system and trademark of SemiAnalysis. It tiers GPU clouds from hands-on cluster testing of security, networking, reliability, orchestration, storage and more. We quote each company’s tier from ClusterMAX 3.0 (published September 2026) and link the SemiAnalysis page on every rating. ClusterMAX is SemiAnalysis’s independent rating; their review has the full findings.

• Tiers, top to bottom: Platinum, Gold, Silver, Bronze, Underperforming. Edition 3.0 adds “Participation” for companies that took part without earning a medal.
• Companies ClusterMAX lists as “Unavailable”, and companies outside the 3.0 rankings, show blank.
• We store the 3.0 tier only. Earlier editions differ: in 2.1 (April 2026) Fluidstack rated Gold and Voltage Park Silver.

Funding, debt and capacity

The funding column is the total raised when a source states one; otherwise it is the latest sourced equity round or debt facility. Equity and debt are different money, so read the funding note before ranking by it. When a report is disputed by the company, as with press reports of Fluidstack rounds at $18B and $40B valuations, we use the company’s own figure.

Stated capacity is copied as the company states it, with its date: GPU counts, megawatts or gigawatts of power, or revenue backlog. The units differ, so capacity reads as text, exactly as stated.

Ownership changes

Mergers and renames are shown on the company page with the date and the source: Voltage Park merged with Lightning AI (January 2026), Ori now runs as Radiant with Brookfield, Northern Data’s Taiga Cloud sits in Rumble’s Quake AI, and Nscale has filed for a NYSE IPO. Merged companies keep their own rows while both brands sell capacity.

Who is in the index

Companies that rent GPU compute as a cloud: on demand, reserved or as dedicated clusters, including marketplaces and AI platforms that rent raw GPUs. Data-centre landlords that lease only space and power sit outside the index. Next in line: ChronoScale, SF Compute, Fluence, Jarvislabs, Gcore, E2E Networks, FPT AI Factory and more regional clouds.

Corrections

Companies and readers can send corrections to hello@neocloudcompanies.com with a public source. We update the value and its date. Featured placement is paid and labelled; every row’s data and its place in a sorted table stay exactly as sourced.