01
Notify & retain records
Report the event as the policy requires, retain incident and service records, and take reasonable steps to reduce loss.
Core protection · Compute operators & enterprise inference providers
Compute Outage Protection
Insurance payouts under agreed rules help ease the financial pressure of customer compensation and business continuity after an outage.
What is covered
First, choose what the cover will measure
01
Data center operations
For owners or operators bearing data center interruption risk, this measures power, temperature and other indicators at specified sites. A facility fault does not automatically mean a customer service outage.
02
The service your customers actually use
For operators or business users bearing compute service interruption risk, this measures actual availability for agreed regions, endpoints and request types, including the service status after backup takes over.
Each option requires its own insured parties, indicators, payout rules and limits to be confirmed. Choose according to your needs; selecting both does not mean one incident automatically triggers two payouts.
What is excluded
- Reduced customer usage, idle compute capacity or falling token prices.
- Unaccepted sites, services or metrics, and maintenance, customer-side failures or known events excluded by the policy.
- Model response quality, infringement, data breaches, unpaid debts and GPU price declines are not independent triggers for this outage protection.
When an event accepted for coverage meets the agreed severity and duration thresholds, payment is calculated using the payout schedule, subject to waiting periods, exclusions and limits.
Costs & Claim Examples
References & estimates · Not a formal quoteInsurance cost reference$2,400–6,000 / yearPublic premium reference calculation · 1 MW data center · 12 months · USD
Standalone service budget$3,200–4,800Optional standalone assessment budget
Assumed workload · 4–6 person-days × $800/person-day
Assumed workload · 4–6 person-days × $800/person-day
The insurance reference comes from Parametrix's public data center SLA pricing. Associated coverage limits and some conditions are not disclosed, and it does not price the payout assumptions below.
The formal proposal specifies whether standard monitoring and verification are included. Standalone services are not a mandatory cost of buying insurance, and included work is not charged twice. Taxes and third-party expenses are listed separately; the two figures do not form a package price.
View the insurance cost source and scope
A Parametrix article published on 21 September 2026 describes a data center SLA insurance structure at $0.20–$0.50 per kW per month. Multiplying by 1,000 kW × 12 months gives $2,400–$6,000.
The source does not disclose the corresponding coverage limit, deductible, waiting period, definition of billable capacity, region, minimum premium or taxes. It also does not provide a one-year quote for this example.
View the public pricing source ↗View the standalone service budget and scope
Scope assumed here: 1 site, 1 upstream agreement, 1 downstream contract template and 1 existing record source, with complete information and remote review.
Contract, backup information and record review, with assessment deliverables4–6 person-days$3,200–4,800
Estimated at 8 hours per person-day and an assumed rate of $800. This is for budget planning, not measured cost or a confirmed quote. Actual scope, workload and deliverables must be agreed in advance.
On-site validation, custom integration development and third-party professional services are excluded. Ongoing monitoring is priced for its separately commissioned scope and duration.
Public sources are provided for information only and do not imply a partnership with or endorsement of aimam.
A 60-minute outage: what would this example pay?
Assume a qualifying outage lasts 60 minutes. After a 15-minute waiting period, 45 minutes are payable at $500 per minute.
Assumed per-event limit$25,000
Loss left uncompensated in this example$7,500
Calculated payout in this example$22,500
Assuming $30,000 of related actual loss and a $22,500 payout, $7,500 remains uncompensated. The 15-minute waiting period is not payable. There is no additional monetary deductible in this example; $7,500 is not a fixed deductible. Premiums and additional service fees are separate.
Expand the calculation & conditions
- The triggers, per-minute amount, limits and losses below are illustrative assumptions, not accepted insurance terms. They are unrelated to the public price above.
- Assume the specified service is continuously unavailable for 60 minutes and meets the policy trigger. The event has been verified, and all required loss declarations and other insurance conditions have been met.
- The 15-minute waiting period runs from the start of the outage. It is neither payable nor retrospectively covered. Each qualifying minute thereafter is calculated at $500, with no further monetary deductible in this example.
- Assume related actual loss of $30,000, with no other insurance proceeds or recoveries. All amounts are in USD.
- Verified continuous outage
- 60 minutes
- Non-payable waiting period
- 15 minutes
- Qualifying time after the waiting period
- 45 minutes
- Agreed amount per minute
- 500 USD/minute
- Per-event payout limit
- 25000 USD
- Remaining annual limit at claim time
- 50000 USD
- Qualifying outage time
- 60 − 15 = 45 minutes
- Calculated at the agreed amount per minute
- 45 × $500 = $22,500
- Check per-event and remaining annual limits
- $22,500 is below both the $25,000 per-event limit and the $50,000 remaining annual limit
- Calculated payout
- $22,500
This payout is calculated from qualifying outage time, rather than reimbursement of itemized actual losses. Subtracting the $22,500 calculated payout from $30,000 of related loss leaves $7,500 uncompensated.
$7,500 is the difference between actual loss and the parametric payout in this example, not a separate fixed policy deductible. Premiums, service fees and other expenses are additional.
Loss declaration requirements, any relationship required between the payout and actual loss, and coordination with other insurance are determined by applicable law and the formal policy.
If customers simply reduce usage, or an outage does not meet the agreed threshold and duration, this parametric cover may not pay. Any other available coverage must be checked separately.
Parametric cover uses agreed events and a payout schedule, and the payment may differ from actual loss. Indemnity-based backup expense or contractual liability cover requires separate verification of actual costs or liability. Required declarations, proof of insurable interest and coordination with other insurance follow the law and policy. Unconditional automatic payment or a fixed payment time is not promised.
How claims work
02
Verify the outage conditions
Check affected assets or services, start and end times, and agreed metrics. Submit loss declarations and other documents as required.
03
Payment under the agreed rules
The insurer or its authorized claims handler applies the waiting period, payout schedule and remaining limits.
Customer compensation remains due under the service contract; do not assume it can wait for insurance proceeds. Payment timing and required declarations and documents follow the policy.
Detailed Terms & Extensions
Do slowdowns, rate limits or partial failures count as an outage?
Only indicators and thresholds accepted in advance are used. Regions, endpoints, model versions, request types and observation windows must be specified. Latency, throughput and partial failures require separate confirmation.
Fewer customer requests do not mean a supply failure. Scheduled maintenance, customer-side issues and model response quality do not automatically become covered outages based on request logs alone.
Who is insured, and who receives the payout?
Before cover begins, identify the insured parties, the risks they bear and the payment recipient. A supplier's insurance does not automatically give its customers a direct right to claim under that policy.
Direct payment to customers or financiers, assignment of rights and payment priority require separate confirmation. Operators must still compensate customers on time under their service contracts and cannot assume they may wait for insurance proceeds.
How do existing insurance and supplier credits work together?
Check supplier service credits, compensation owed to customers, your own expenses and existing insurance separately. A service credit may not be cash and may not cover the full business loss.
How parametric payouts coordinate with other insurance must be confirmed in advance. Do not assume that all payments can be added together or must all be deducted. The same actual expense is not reimbursed twice.
What if monitoring data is missing or the service changes?
Agree on raw records, alternative evidence and dispute review before connection. Missing data requires review: no record does not mean no outage, and not every log is automatically acceptable for calculating a payout.
Material changes to endpoints, models, dependencies or contractual commitments require version records and the agreed notification and acceptance process. Monitoring alerts do not replace reporting a claim through the designated channel.
How are multiple outages in the same year calculated?
First use the agreed cause, interval and time window to determine whether incidents form one event. Then apply event limits, sublimits and the remaining annual limit. A common failure affecting several services does not give each service a separate full limit.
Confirm separately whether limits can be reinstated and whether terms remain the same on renewal or after a change. Assess shared power, network, authentication and upstream dependencies between primary and backup resources.
How can backup procurement costs be covered separately?
The basic scope addresses eligible extra procurement costs caused directly by accidental failure of specified power equipment at listed sites, making primary equipment unavailable. Software, network and external grid failures are not automatically included; other causes require separate acceptance.
The waiting period follows continuous unavailability of the primary equipment. Backup restoring customer service does not automatically end eligible backup costs. The agreed cutoff is the earliest of actual restoration, when restoration could reasonably have been achieved, and the end of the maximum indemnity period.
Check procurement necessity, equivalent service and reasonable prices. Deduct actual cost savings and other compensation for the same expense, then apply deductibles and limits. Routine capacity reservation fees are not incident losses.
How can contractual customer compensation be covered separately?
Service contracts, compensation terms and covered causes must be accepted in advance. Refunds, service credits and agreed compensation are assessed against the full measurement period and established liability. A brief fault does not necessarily breach a monthly commitment.
Agree in advance how liability, settlement approval, notification and loss evidence are handled. Future service credits are not all treated as cash losses at face value. Unaccepted new commitments and unlimited liability are not automatically covered.
Calculate contractual compensation and backup costs separately, confirming sublimits, shared limits, retained shares and annual balances. The same actual loss is not reimbursed twice; parametric payments follow the agreed coordination arrangements.
What information should I prepare before applying?
Prepare operating entities and payment recipients, site and service lists, upstream and downstream contracts, existing coverage, backup arrangements, operating records and past incidents. Check each party's payment duties, entitlement to proceeds and notification requirements.
Insurance measurement rules, alternative evidence and scope must be confirmed by the authorized parties. Custom connections, on-site inspections and additional specialist support are priced as separate engagements.
Illustrative Coverage CalculationSee how backup costs, waiting periods, retained amounts and limits affect the estimate
Expense illustration
Assume listed primary equipment is down for 40 minutes following an agreed accidental power-equipment failure. Backup takes over at minute 5. Customers are affected for 5 minutes, but primary equipment remains unavailable, so necessary backup procurement continues.
- 14:00Primary equipment failsWaiting period begins
- 14:05Backup takes overCustomer service restored
- 14:15Waiting period endsSubsequent eligible costs begin accruing
- 14:40Primary equipment restoredExpense calculation ends in this example
The 15-minute waiting period here is based on continuous unavailability of the listed primary equipment. Customer service restored through backup is not the same as primary-equipment recovery. Costs in the first 15 minutes are not retrospectively reimbursed.
Estimate under these assumptions
$5,000
Expense estimate · Not a promise of payment
Backup costs after the waiting period$7,500
Less: costs actually saved−$1,000
Less: other compensation for the same expense−$500
Less: customer-retained amount−$1,000
Remaining limit appliesNo more than $20,000
This example does not reach the per-event or annual limit.
Actual payment also requires verification of the event cause, necessity of costs and all policy conditions, as determined by the insurer.
Four key points in assessing expenses
- Establish the cause: A monitoring anomaly is evidence, not a standalone payout trigger. The insured event affecting listed equipment must be linked to the backup expenditure.
- Define the period: Costs accrue after the waiting period and end at the earlier time specified by the agreed rules: restoration of the listed primary equipment's original service capacity, when restoration could reasonably have been achieved, or the maximum indemnity period.
- Avoid double recovery: Deduct only actual savings from the same period and compensation for the same loss. Recoveries received after a claim payment are settled under the policy to prevent duplicate compensation.
- Aggregate first: Where the same cause affects several devices or customers, the per-event limit cannot be reapplied for every customer. Event aggregation, intervals and allocation across policy years must be specified in advance.