This SLA forms part of the agreement between GridWind Integrated Solutions Pte. Ltd. and Customer and applies to the covered services: Training Clusters, Inference Endpoints, Burst Capacity reservations, and the Console & API (each a “Covered Service”).
For context: 99.95% allows at most 21 minutes 54 seconds of downtime in a 730-hour month; 99.9% allows 43 minutes 48 seconds. Order Forms may commit uplifted targets for specific deployments.
If Monthly Uptime falls below target and you meet the claim conditions in Section 8, we credit a percentage of the month’s fees for the affected Covered Service in the affected region or cluster:
Credits apply to future invoices within 60 days of approval, are capped at 50% of the month’s fees for the affected Covered Service, and — except where liability cannot be limited by law — are your sole and exclusive remedy for missed service levels. If the agreement ends before a credit is consumed, we pay out its remaining value.
For Training Clusters we additionally commit that:
P1/P2 are answered by a named infrastructure engineer with access to your topology — not a triage queue. Escalation contacts are listed in your Order Form.
Planned maintenance is announced on the status page and to registered contacts at least 72 hours in advance, is scheduled in low-impact windows for the affected region, and targets less than 2 hours of customer-visible impact per Covered Service per quarter. Cluster maintenance uses checkpoint-aware draining (Section 4). Emergency maintenance necessary to protect the platform may proceed on shorter notice; we notify as early as practicable. Announced maintenance performed within these bounds is excluded from Downtime; maintenance that exceeds its announced window is not.
Availability is measured by GridWind’s monitoring: 5-minute probes per component and region — the same feed that drives status.gridwind-solutions.com — plus per-cluster fabric telemetry. Downtime accrues from when our monitoring detects the condition or you report it, whichever is earlier, and ends on restoration or in-kind replacement. Monthly service reports are available to committed customers. Your own logs and monitoring are admissible evidence in a claim; discrepancies are reconciled in good faith against raw probe data, which we retain for 13 months.
Submit credit claims to your account team or noc@gridwind-solutions.com within 30 days of the end of the month in which the target was missed, identifying the Covered Service, region or cluster, and the Downtime periods (with logs if available). We respond within 10 business days; approved credits appear on the next invoice. Claims submitted after the window are forfeited.
Downtime does not include unavailability caused by:
For clarity, two things are not excluded: unavailability of a confirmed Burst Capacity reservation during its window, and planned maintenance that overruns its announced window. Both count as Downtime.
If an Order Form commits different targets, tiers, or credit ladders for a deployment, the Order Form prevails for that deployment. We may update this SLA no more than once per calendar year for new orders; for in-flight committed terms, the version in force at execution continues to apply. Material changes are notified to account contacts 30 days in advance.