On demand
Standard capacity
More predictable access, usually at a higher price.
← Back to Compute College
Compute College
What are spot prices? AI GPU spot market explained
How interruptible GPU capacity prices expose marginal AI compute supply.
Interruptible capacityProduct
The discount exists because access can be reclaimed.
Marginal supplySignal
Spot depth and price can show spare capacity at the edge of the market.
Plain-English definition
Spot prices are rates for short-term, interruptible compute capacity sold from spare supply. They matter because they can reveal whether usable GPU capacity is tightening or loosening before standard contract prices visibly move.
Memory trick: Spot compute is an empty seat sold cheaply just before departure; it may disappear when the regular passenger needs it.
Why it matters
- Providers sometimes have unused capacity they would rather monetize than leave idle.
- Buyers with flexible workloads may accept interruption in exchange for lower prices.
- Spot markets help reveal the value of spare capacity at a given moment.
- The available pool can vary by region, chip type, and time.
Simple example
Interruptible
Spot capacity
Lower-cost access to spare capacity, but with the risk of interruption.
The discount is the trade-off buyers receive for accepting less certainty. Any figures shown are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
- Whether spare compute capacity appears abundant or scarce.
- Whether buyer demand is pressing into short-term supply.
- Whether a chip type is easy or difficult to access at the margin.
- Whether the market is loosening even before headline contract prices change.
Market read: spot prices and depth can move before standard contract pricing because they expose spare capacity directly. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
Spot capacity is cheaper because the buyer accepts a different product: lower certainty, possible interruption, and less suitability for workloads that cannot easily stop and restart.
Cost
Price
Usually lower than standard access.
Access
Reliability
Lower because the provider may reclaim capacity.
Fit
Workload fit
Best for jobs that can tolerate interruption.
Practical takeaway
What you can do with this
Compare spot capacity only for workloads able to tolerate interruption or checkpoint and restart. Record discount, reclaim risk, region, GPU class, and availability together.
- Buyers: reserve reliable capacity for deadlines and test flexible workloads against spot savings.
- Analysts: watch spot availability as a signal of spare capacity at the margin.
- A checkpointable experiment may tolerate reclaimed capacity, while a customer-facing service or deadline-bound training run may lose more value from interruption than it saves.
- Compare spot observations only within the same hardware class, region, and service terms, and note how much capacity can actually be obtained at the displayed level.
- A temporary spot discount should not be presented as a durable market price unless availability and comparable observations support that conclusion.
Decision check: a spot discount is valuable only when an interruption does not erase the savings or miss the deadline.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
GPU Pricing & Capacity
Step 7 of 8: What are spot prices