Listen to this article
Introduction
Traditional warehouses and lakehouses force you to choose between performance, cost, and simplicity. Firebolt delivers all three on a single platform—with the best price-performance in the market so you can ship faster without compromises.
We're built for companies that need their data platforms to do more — run AI workloads, power sub-second customer-facing analytics at scale, or execute ELT jobs faster at a fraction of the cost.
ClickHouse recently published a couple of articles titled "How the 5 major cloud data warehouses really bill you" & “How the 5 major cloud data warehouses compare in cost-performance” introducing their Bench2Cost framework for translating benchmark runtimes into real costs. We agree with the core premise:
Price lists don't tell you real costs. Compute models do.
That's why we ran Bench2Cost on Firebolt Cloud. We ran the complete 43-query ClickBench suite at 10 billion and 100 billion rows with comparable Firebolt compute clusters. We compared our results against the results of all cloud data warehouses in the original article. We also agree that, as data and workloads grow, architectural differences become undeniable.
This blog post walks you through Firebolt’s compute and billing model, the results of the benchmark and a full review of where we win and where we lose.
VIDEO HERE
TL;DR: Firebolt delivers better performance and significantly lower cost than ClickHouse, Snowflake, BigQuery, Redshift, and Databricks.
Key takeaways:
- Firebolt is 43% faster than the runner up (Clickhouse)
- Firebolt is 68% cheaper than the runner up (Clickhouse)
- Firebolt dominates the field on individual query performance.
Firebolt's Compute and Billing Model
Before diving into results, and in the spirit of transparency, let's explain how Firebolt bills compute.
Firebolt Units (FBUs)
Firebolt uses Firebolt Units (FBUs) as its billing currency. It is a hardware-grounded compute unit similar in spirit to ClickHouse's 8 GiB memory increments, but with a key difference: Firebolt offers two engine families optimized for different workloads.
Compute-Optimized engines deliver more CPU per FBU by trading off local NVMe storage—ideal for analytical workloads where data resides in object storage and compute is the bottleneck.
Pricing (AWS us-east-1, December 2025)
Different pricing tiers are included here for full transparency. The benchmark was executed entirely on our Standard tier.
How Bench2Cost Computes Firebolt Costs
Our enrichment script calculates costs using this formula:
Compute Cost = query_time_seconds × (fbu_rate_per_hour / 3600) × total_fbuWhere total_fbu = fbu_per_node × cluster_size.
For a 20 node Large Compute-Optimized cluster:
- FBU per node: 16
- Total FBU: 320
- Cost per second (Standard): 320 × ($0.23 / 3600) = $0.0204/sec
The Benchmarks: 10B and 100B Rows
We ran the complete ClickBench suite (43 analytical queries) on a (20n) Firebolt cluster and used the results from other systems with comparable configurations:
Firebolt test ran on AWS us-east-1 with per-second billing. Each query was executed 3 times and following the ClickBench approach, we use the best run for comparison.
Results: 10 Billion Rows
Relative Comparison (vs Best)
We calculated the ratio of total runtime between each system and the fastest system as well as the ratio of cost between each system and the cheapest system at both standard and enterprise pricing.
Example: Firebolt’s total runtime was the best at 22.4s, ClickHouse was 66.7s, so the Runtime ratio is 66.7 / 22.4 = 2.98 which means Firebolt is almost 3 times faster than Clickhouse at this scale.
Summary
At 10 billion rows Firebolt is the leader:
- Fastest: Firebolt 20n (22.40s)
- Most Queries Won: Firebolt 20n (36 of 43 = 84%)
- Cheapest (Standard): Firebolt 20n ($0.46)
- Cheapest (Enterprise): Firebolt 20n ($0.60)
Results: 100 Billion Rows
At 100B rows, the differences grow:
Relative Comparison (vs Best)
Summary
At 100 billion rows Firebolt continues to lead:
- Fastest: Firebolt 20n (156.11s)
- Most Queries Won: Firebolt 20n (29 of 43 = 67%)
- Cheapest (Standard): Firebolt 20n ($3.19)
- Cheapest (Enterprise): Firebolt 20n ($4.16)
Query Level Breakdown at 100 Billion rows
Firebolt (29 queries won)
Firebolt dominates in the most demanding analytical query types:
Key Strength: Firebolt excels at the hardest analytical queries—those involving high-cardinality grouping and filtered scans on large datasets. These queries stress hash aggregation and data pruning, where Firebolt's architecture provides significant advantages.
ClickHouse (12 queries won)
ClickHouse wins on simpler, full-scan workloads:
Key Strength: ClickHouse excels at full-table scans, string operations, and queries that don't require complex aggregation state management. Its vectorized execution shines on straightforward column scans.
Snowflake (1 query won)
Snowflake's optimizer handles this simple metadata query efficiently but struggles with more complex analytical workloads at 100B scale.
Redshift Serverless (1 query won)
Redshift wins this single filtered aggregation but is significantly slower on all other query types.
Key Insight
The query breakdown reveals a clear pattern:
- Firebolt wins the complex queries that define real-world analytics: high-cardinality grouping, filtered aggregations, and multi-column operations
- ClickHouse wins the simple queries: basic counts, full-table scans, and string searches
- Traditional DWs (Snowflake, Redshift, Databricks, BigQuery) win almost nothing at this scale
This suggests that as analytical workloads become more sophisticated (beyond simple scans), Firebolt's architecture provides increasingly significant advantages. For analytical workloads at enterprise scale, the numbers are clear: faster queries compound into dramatically lower costs.
Clickhouse’s article includes a visualization of performance/cost comparing results among cloud data warehouses and shows that it is faster and more cost efficient than the others. We replicated the same chart and included Firebolt in the mix.
At the 10 billion row scale, using the 20-node results, Firebolt is roughly 3x faster and 5x less expensive than Clickhouse. As the data volume grows tenfold to 100 billion rows, both engines increase in cost and slow down in query performance. It is, afterall, an order of magnitude more data so this is expected for many of the benchmark queries. The rate of change in performance and cost as the data grows, however, is different across systems. As data volume grows, Firebolt scales more efficiently than any other system tested. At 100 billion rows, Firebolt is still roughly 2x faster and 3x less expensive than ClickHouse; the runner up. The differences with other systems are much larger.
Conclusion
ClickHouse's Bench2Cost framework is valuable as it brings transparency to cloud data warehouse pricing. When we include Firebolt in the picture and use the same metrics to compare them, Firebolt is the clear winner.
At scale (10B-100B rows), Firebolt consistently delivers:
- 40-66% faster query performance
- 68-81% lower compute costs
- More query wins across diverse workload patterns
Where ClickHouse excels:
- Simple LIKE pattern matching (Q21-Q23)
- Trivial queries under 10ms (Q1)
For analytical workloads at enterprise scale, the numbers are clear: faster queries compound into dramatically lower costs.
All benchmarks were run using the open-source Bench2Cost framework and publicly documented pricing. The full results are reproducible and available in the bench-2-cost public repository.
Get started with Firebolt Cloud today. Visit firebolt.io for a free trial.
%20logo.png){kind=icon}
