Spot Strategy 2026

Spot saves 60-90%. The architecture that fits.

Overview

Spot strategy 2026 uses AWS spot instances correctly with diversified fleets. Picking spot is the easy decision; the architecture that survives interruption is the discipline that turns the headline 60-90% savings into actual operational savings.

• Spot saves 60-90%. Headline savings versus on-demand; the savings only land when the workload tolerates interruption.
• Per-fleet diversification. Mixed instance types per fleet; reduces simultaneous interruption risk across the diversified set.
• Capacity-optimized strategy. Allocation strategy that picks deep-capacity pools; modern default; replaces lowest-price chasing.
• Per-AZ spread plus per-quarter review. Diversify across AZs; quarterly spot review catches pool drift before it causes interruption spikes.

The approach

The practical approach: diversify per fleet, use capacity-optimized allocation, spread across AZs, review quarterly, document the per-fleet rationale. The team’s discipline produces reliable spot capacity instead of cheap-then-suddenly-gone.

• Per-fleet diversification. Mixed instance types per fleet; the diversified set absorbs single-pool interruption without dropping capacity.
• Capacity-optimized strategy. Allocation by deep-capacity pool; trades a small price uplift for materially lower interruption rates.
• Per-AZ diversification. Spread across AZs; AZ-level capacity events do not take the entire fleet at once.
• Per-quarter review plus documented policy. Quarterly spot review catches drift; per-fleet rationale committed to the repo for operational reviews.

Why this compounds

Spot strategy discipline compounds across fleets. Each diversified fleet preserves capacity; the team’s spot expertise grows; new fleets inherit the patterns from the previous round.

• Better cost efficiency. Spot delivers 60-90% savings; the savings actually land because interruption rates stay low.
• Better spot reliability. Diversification reduces interruption; the workload sees occasional churn, not capacity collapse.
• Better operational fit. Right strategy matches workload; batch absorbs interruption naturally, user-facing needs more careful design.
• Institutional knowledge. Each fleet teaches spot patterns; the team’s compute engineering muscle grows.

Spot strategy discipline is an operational discipline that pays off across years. Nova AI Ops integrates with spot telemetry, surfaces patterns, and supports the team’s compute discipline.