Provider record · updated 2026-09-10

Prime Intellect — GPU pricing & deliverability

5 tracked offerings · grade mix A:0 · B:0 · C:0 · D:0 · U:5. Aggregator across '50+ providers'; real-time offers are behind the authenticated app, so homepage widget prices are the only public numbers. Homepage claims 'Infiniband Networking - high-bandwidth interconnects across nodes' for cluster products, but publishes no per-offer topology, provider identity, region, or node size. DATA QUALITY CAVEAT: the same widget shows four rows labeled 'H200 / 80 GB VRAM / 184 GB RAM / 32 vCPU' at $1.99, $1.80, $1.23 and $0.47/hr - internally inconsistent (H200 is 141GB), likely an illustrative/mock element; those rows were EXCLUDED. A reserved-cluster calculator shows an example 'B300 SXM6 x 512, 3-year reserved, $5.00/HR/GPU' ($2,560/hr total) - illustrative, quote-based. GH200 ($3.14), RTX Pro 6000 ($3.14), A100 ($3.14), A40 ($3.14) also listed; the identical $3.14 across four different GPUs further suggests partly illustrative pricing. No MI300X or RTX 4090/5090 shown.

Tracked offerings

Grades measure what sellers publish — deliverability of the written claim, not measured performance. No fleet holds a measured (Verified) badge yet; that tier arrives with the verification harness. Source classes: API = provider's live pricing API · vendor = provider's published page · mkt API = marketplace API · 3rd-party = aggregator (provider publishes no price on its own pages) · quote = quote-gated or "starting at", not a walk-up price. Quote-class prices are shown in tables but excluded from headline ranges, spreads, and fair-price bands.

Terms & trust (as published, 2026-09-14)

Notes: Marketplace/aggregator, not an infrastructure owner: pods deploy onto ~20+ underlying providers (API enum includes runpod, lambdalabs, fluidstack, hyperstack, cudocompute, scaleway, tensordock, datacrunch, crusoecloud, nebius, akash, vultr, etc.); ToS pushes provider terms onto the customer ('you explicitly agree to the terms and conditions of any cloud service providers we may utilize'). Egress pricing not published anywhere found. No SOC 2/ISO 27001/HIPAA claims — the /security page is a vulnerability-disclosure policy only. Tier distinction exposed as an API/filter parameter: security = 'secure_cloud' | 'community_cloud'; Multi-Node clusters deploy on 'Secure Cloud infrastructure', but published criteria for secure vs community qualification were not found. SLA posture explicitly deferred to underlying provider SLAs.

Sources: ↗ ↗ ↗ ↗ ↗ · Fields marked "not published" are exactly that — the seller does not state them anywhere we could find. Absence is a finding.

Reliability record (public status feed, last 90 days)

5 incidents on Prime Intellect's own status page in the trailing 90 days, collected 2026-09-15. Self-reported by the provider — a floor on true incident count, not a ceiling; it still beats the war-story thread you were going to read instead. High counts often reflect granular per-rack reporting, not worse infrastructure: read the incidents, not just the number.

What Prime Intellect does not publish

• underlying provider per offer
• form_factor
• gpus_per_node
• intra_node
• inter_node_fabric per offer
• max_contiguous_gpus
• regions
• whether homepage widget prices are live or illustrative

Unpublished fields are recorded as unknown and grade conservatively. Sellers improve their grades by disclosing — that is the point.