How We Track the Market
Every claim we make in advisory engagements and published research is grounded in verifiable data. Here is how we collect, verify, and maintain the intelligence that underpins our work.
What we track
The intelligence platform currently covers 2,800+ companies across 72 categories spanning the full GPU infrastructure value chain: silicon and accelerator vendors, GPU cloud operators and neoclouds, data centre developers and operators, networking and interconnect companies, cooling and thermal management providers, storage vendors, power and electrical infrastructure, orchestration and management software, and professional services firms.
Each company record includes founding information, geographic footprint, funding history, technical capabilities, customer segments, and competitive positioning. Records are maintained at varying levels of depth - flagship operators and major vendors carry full competitive profiles; tail-end entries carry basic classification and tracking data.
Where the data comes from
- Direct market engagement: Conversations with GPU operators, hardware vendors, data centre developers, and infrastructure investors. This is the primary source - not press releases, not scraped data, not analyst reports.
- Pricing APIs and monitoring: Automated collection from 70+ GPU cloud provider pricing pages and APIs. Spot, reserved, and on-demand rates captured at regular intervals with historical trend data maintained.
- Public filings and regulatory data: SEC filings (10-K, 10-Q, S-1, 8-K), Companies House records, EU regulatory filings, patent databases, and government procurement records for sovereign AI programmes.
- Facility databases: Data centre facility records sourced from planning applications, environmental permits, utility connection records, and direct verification. Currently covering 3,500+ facilities across 101 countries.
- Industry events and conferences: Direct attendance and engagement at OCP Summit, SC Conference, Datacloud, and regional infrastructure events for on-the-ground validation.
- Advisory engagements: Each client engagement produces market intelligence that feeds back into the platform (within confidentiality boundaries). Patterns observed across engagements inform sector-level assessments.
How we verify claims
GPU infrastructure is a market where companies routinely overstate capacity, understate costs, and conflate planned with operational. We apply three verification layers to every material claim:
Source triangulation: No single-source claims in published research. Material figures (pricing, capacity, facility count) require confirmation from at least two independent sources - ideally one primary (direct conversation or API data) and one secondary (filing, permit, or third-party reference).
Temporal flagging: All pricing and capacity data carries a timestamp and explicit caveats. We distinguish between current market data, point-in-time snapshots, and indicative ranges. Numbers that age beyond their useful life are flagged or removed.
Bear-case framing:Published research includes counter-narratives. If we report an operator's claimed capacity, we also note what independent evidence supports or contradicts that claim. Healthy scepticism is methodological, not editorial.
How often data refreshes
- GPU cloud pricing: Automated collection at regular intervals. Historical pricing maintained for trend analysis and advisory benchmarking.
- Company records: Core records reviewed quarterly. Major operators and active deal targets reviewed monthly or on-trigger (fundraise, acquisition, product launch).
- Facility data: Updated on discovery via planning applications, permit filings, and operator announcements. Full database audit conducted quarterly.
- Published research: Insights articles carry datePublished and dateModified timestamps. Material changes (pricing shifts, market events, factual corrections) trigger republication with updated data.
What we don't claim
This is a fast-moving market with significant information asymmetry. Some honest limitations:
We track publicly observable or directly sourced data. We do not have visibility into hyperscaler internal pricing, classified government compute programmes, or non-public M&A activity until it surfaces through filings or market signals. Our facility data is strongest in North America and Europe; coverage in parts of Asia, Africa, and Latin America is thinner.
Pricing data reflects listed and negotiated rates we can observe or source directly. Private contracts with custom terms are not captured in aggregate benchmarks. Where we present pricing ranges, they represent the observable market - not the full distribution of possible prices.
We revise when we're wrong. Factual corrections are applied immediately and documented in article modification dates. If you spot an error in our published data, contact ben@disintermediate.global.
Get The GPU Weekly
GPU infrastructure intelligence - read by NVIDIA, CoreWeave, and Brookfield. Every Saturday. Free.
Questions about our data?
We're transparent about how we build our intelligence. If you have questions about methodology, sourcing, or specific data points, we're happy to discuss.
Get in touch