Independent verification · Large-scale compute

Compute became collateral faster than anyone learned to verify it.

Collatr is an independent verification and intelligence firm for large-scale compute. We bring the operator’s knowledge of GPU infrastructure, cluster systems, and hardware economics to the lenders, investors, insurers, and rating agencies that now depend on them.

$20B+
in announced GPU-backed credit facilities, an asset class roughly three years old (analyst estimates)
~$800B
of private credit projected into data-center debt, the broader category this asset class sits inside (Morgan Stanley, 2025)
~68%
peak-to-trough decline in H100 rental rates, 2023 to mid-2026, before spot rates firmed this year (our read of public data)
Capabilities

We understand compute infrastructure the way its operators do.

Large fleets are systems, not line items. Our work rests on the disciplines it takes to run them.

01

Fleet telemetry

Modern fleets emit device-level data as a matter of basic operations: utilization, power draw, thermals, error activity. We know what these signals mean, what they miss, and how they can mislead.

02

Cluster systems

A GPU’s productive value depends on what surrounds it: interconnect fabric, power contracts, cooling, and the software that keeps utilization high. We evaluate clusters as operating systems of hardware, not inventories.

03

Hardware lifecycle economics

Depreciation, refresh cycles, generational transitions, and residual value. The distinction between replacement cost and value in use, measured rather than assumed.

04

Physical verification

Serial-level sampling, site inspection standards, and evidence protocols designed for third-party reliance. Independent measurement points, cross-checked.

05

Market data

Daily capture of rental rates from public GPU marketplaces, begun July 2026, with secondary-market coverage in development. Observable series for assets that are usually marked by negotiated schedules.

06

Counterparty analytics

Demand decomposed by counterparty: whose workloads run on the fleet, whose capital funds the purchases, and how much of the order book stands at arm’s length.

Services

For the institutions on the other side of the hardware.

Built for lenders, structured-credit investors, insurers, and rating agencies.

Assessment

Deal-time evaluation of compute assets and the systems around them, delivered under a versioned methodology and built for the credit file.

Monitoring

Ongoing reporting through the life of a facility. Independent measurement, on a schedule, with defined escalation.

Data & research

Price series, datasets, and published analysis of the compute economy, built to be maintained continuously rather than assembled per engagement.

Approach

Independence is a method, not a claim.

Independent measurement

We rely on cross-checked signals from sources no single party controls. No one’s self-reported data is ever load-bearing on its own.

Versioned methodology

Our methodology is versioned and will be published. Findings state what agreed with what, under which version. Precision about what was checked is the product.

No conflicting positions

We work for the party relying on the answer. We do not trade, broker, or take positions in the assets we assess.

Research

Published analysis of the compute economy.

We publish our research openly. The first note appears this month; the datasets behind it are being built now.

Technical note · July 2026 · Publishing this week

GPU Collateral: Depreciation, Utilization, and What Lenders Can’t Currently Verify

The operator’s view of the fastest-growing collateral class in credit. Request an advance copy →

Data · Series begun July 2026

GPU rental price tracking

Daily capture from public marketplaces, with secondary-market coverage in development. A public data layer is in preparation. Ask about early access →

About

Built by operators.

Collatr was founded on a simple premise: the expertise that runs the world’s largest compute fleets should also serve the institutions that finance them.

Our founders built and ran machine-learning infrastructure at hyperscale: the data platforms and telemetry systems the world’s largest training fleets run on. Collatr’s methods come from that side of the industry: measurement-first, systems-level, and skeptical of any number that arrives without instrumentation behind it.

For engagements, research access, or an advance copy of the current note:

info@collatr.com