← Data Center Market Intel·Risk
Insurance · Non-damage business interruption

NDBI

SLA risk analytics on the reviewed agreements and industry estimates in the Colocation SLA benchmark, read twice: once for an insurer (what the SLA pays, and where the limits sit) and once for a lender (what reaches the borrower’s contracted revenue, and how collateral ages). Outage frequency and restore time are assumptions you set, because no incident data is held yet.

Concept page
Contract terms come from reviewed agreements; the AI-HPC and Real-Time rows are industry estimates. Sections and bars tagged Concept depend on assumptions you set, and nothing tagged Concept is a measured loss rate.
Confidence A Directly reported figure B Estimated / Derived C Regional inference / indirect estimate D Not found / insufficient data

Largest month of credits
SLAs with a stated remedy
Colour key Insurance view: payouts, tail, limits Lender view: revenue, leakage, recourse, collateral
SLA stack

What each SLA promises and pays

Routine vs tail

Credit exposure under an assumed outage pattern

Concept
Risk layers

What credits cover and what they leave

Early warning

Drift against the downtime budget

Concept

Assumes the outage pattern above as the baseline, plus how far through the rolling year the site is and how much downtime has been counted.

How far through the 12-month window the site is.
Electrical downtime minutes recorded in the window.
Collateral and obsolescence

How chip collateral could age

ConceptLender view

Assumes how long the chips stay economically useful and how hard they are run. No resale or depreciation data is connected.

Years until value reaches its floor. A 2-year compressed life is always drawn for comparison.
Above 1 means the chips run harder than design, which shortens their useful life.
What the monitoring sample supports
A hardware partner has shared a confidential 30-day monitoring sample for field review. It is not loaded on this page.
  • Can inform: a usage and thermal stress profile, from temperatures, power draw and hardware status over 30 days. That is the idea behind the stress input above.
  • Cannot give: depreciation, resale outcomes or failure rates. An internal review put the minimum at 20 to 50 adverse events, which 30 days of monitoring does not provide (to be confirmed).
  • Needed to calibrate: resale prices linked to usage, hardware invoice prices, depreciation cycles, and maintenance and hardware return (RMA) history.
  • Not connected: any market price series for chips.
Type compare

SLA stringency across types

All types