Spot Market Intelligence

Compare regional arbitrage, hardware families, and historical volatility.

{{ updateMessage }}

Arbitrage Matrix

Click headers to sort
Region {{ sortAsc ? '↑' : '↓' }} Instance {{ sortAsc ? '↑' : '↓' }} Avg Price {{ sortAsc ? '↑' : '↓' }} Std Dev {{ sortAsc ? '↑' : '↓' }} Suggested Max Price (P95) {{ sortAsc ? '↑' : '↓' }} Volatility {{ sortAsc ? '↑' : '↓' }} Total ({{ computeHours }}h) {{ sortAsc ? '↑' : '↓' }}
{{ est.region }} {{ est.instance }} ${{ est.hourly_estimate }} ±${{ est.std_dev }} ${{ est.recommended_max_bid }} {{ est.volatility_risk }}% ${{ est.total_estimate }} (Best)
Select regions and instances to build the matrix.

How to Interpret This Data

Avoid common Spot Market traps with these strategic guidelines.

1. The "Cheap but Volatile" Trap

If a region is the absolute cheapest but has a Volatility Risk over 20%, you will be interrupted.

Gotcha: Only use highly volatile regions for stateless web servers, containerized tasks, or ML training jobs that aggressively checkpoint (save) their progress every few minutes.

2. The "Max Bid" Reality

We recommend setting your Maximum Bid to the P95 Price (the 95th percentile). This prevents your instance from being killed during standard daily price fluctuations.

Gotcha: Setting a massive bid (e.g., $100/hr) will not protect you from eviction. If AWS physically runs out of servers for On-Demand customers, they will terminate your Spot instance regardless of your bid.

3. "Inverted" Hardware Markets

Don't blindly assume a .xlarge is always cheaper than a .2xlarge. "Herd behavior" often drives up the price of smaller instances.

Gotcha: Always check the next size up in your region. You will frequently find scenarios where you can get double the compute power for less money because nobody else is requesting that specific size.

Pricing History (Region + Instance)