Understanding Auction Price Dynamics

One of the most common questions we get from new users is: "How should I set my bid price?" It's a fair question. Unlike fixed-price cloud providers, KubeBid's auction system means prices change constantly. Understanding why—and when—prices move can save you significant money.

We analyzed 12 months of auction data across all regions and instance types. This post shares what we found.

How Auction Pricing Works

First, a quick refresher. KubeBid uses a continuous double auction model:

1. You submit a bid with your maximum willingness to pay per hour
2. We match your bid against available capacity
3. You pay the clearing price, which may be lower than your bid
4. Prices adjust continuously based on supply and demand

The "market price" you see on our dashboard is a time-weighted average of recent clearing prices. It gives you a sense of where prices are, but actual prices can vary significantly.

What Drives Price Changes

Our analysis identified several key factors that influence auction prices:

1. Time of Day

The strongest predictor of price is time. Prices follow a predictable daily cycle:

Key insight: Running batch jobs between midnight and 6am UTC can save 25-30% compared to peak hours.

2. Day of Week

Weekends are significantly cheaper than weekdays. Enterprise workloads drop off Friday evening and don't resume until Monday morning.

Key insight: If your workload is flexible, scheduling for weekends can save 20-35%.

3. Instance Type Popularity

Not all GPUs are created equal—or priced equally. Newer, more powerful GPUs command higher prices but also have more price volatility.

Key insight: A100s offer the best combination of performance and discount. H100s have the highest demand and most volatile pricing.

4. Regional Differences

Prices vary significantly by region. Generally, US regions have higher prices due to demand, while less popular regions offer better deals.

Optimal Bidding Strategies

Based on our analysis, here are concrete recommendations:

For Batch Workloads

• Set bids at 60-70% of current market price
• Allow flexible scheduling (2-6 hour wait time)
• Enable cross-region placement
• Target weekends and off-peak hours
• Expected savings: 50-70%

For Production Workloads

• Set bids at 110-120% of market price for buffer
• Use reserved capacity as fallback
• Pin to specific regions for network latency
• Expected savings: 20-35%

For Development/Testing

• Use aggressive bidding (50-60% of market)
• Auto-terminate at end of business day
• Choose less popular instance types (L40S vs H100)
• Expected savings: 60-75%

Using Bid Strategies

Rather than managing bids manually, we recommend using Bid Strategies. They automatically adjust your bids based on:

• Historical price patterns
• Current market conditions
• Your workload requirements
• Budget constraints

# Example: Aggressive cost optimization
apiVersion: kubebid.io/v1
kind: BidStrategy
metadata:
  name: batch-optimized
spec:
  type: CostOptimized
  config:
    bidPercentage: 65
    maxWaitTime: 4h
    preferredTimes:
      - "00:00-06:00"  # Night (UTC)
      - "SAT,SUN"      # Weekends
    preferredRegions:
      - eu-west-1
      - ap-southeast-1
      - us-west-2

Price Alerts

We also offer price alerts so you can take advantage of temporary price drops:

# Set up a price alert
kubebid alerts create \
  --instance-type a100-4x \
  --region us-west-2 \
  --price-below 4.00 \
  --webhook https://your-endpoint.com/webhook

# You'll receive a notification when prices drop below $4/hr

Conclusion

Understanding auction dynamics is the key to maximizing your savings on KubeBid. The short version: be flexible on timing and region, use bid strategies to automate optimization, and save batch workloads for off-peak hours.

Have questions about optimizing your bidding strategy? Reach out to our team at [email protected] or join our Discord community.