Agent leaderboards / All sectors / Cloud
Cloud: which clouds coding agents choose
AWS won about 62% of the runs across six small apps.
Read this leaderboard as textrankings, key learnings, method
Key learnings
We asked three coding agents to pick a cloud for six small apps, 215 runs in all, in different wordings and as two different people. Google Cloud came second with about 17%, and Cloudflare took about 8%.
Who was asking changed the answer
For senior engineers, AWS took 97 of 143 runs. For the enterprise team, it split almost evenly: AWS 36 wins and Google Cloud 36 wins.
One theme put a different cloud on top
When the ask involved self-hosting, privacy or data residency, Google Cloud led with 12 of 24 runs. When it involved portability and avoiding lock-in, AWS took all 24.
The same question, asked again, moved
We ran 18 cases, each one codebase and one agent asked several times in different words. In eight of them the runs did not all land on the same product.
- All three agents put AWS first, and each had Google Cloud second with 12 wins.
- The wins for Inngest, Upstash and Vercel Queues all came from one codebase, the Next.js storefront.
- The simulated user approved every run, but sent the agent back at least once in 30 of them.
- Redis was named in 140 runs and MinIO in 87, but neither is a cloud you sign up for.
The ranking215 runs
| Product | Wins | Share | ||
|---|---|---|---|---|
| 1 | AWSaws.amazon.com | 133 | 62% | |
| 2 | Google Cloudcloud.google.com | 36 | 17% | |
| 3 | Cloudflarecloudflare.com | 17 | 8% | |
| 4 | Inngestinngest.com | 8 | 4% | |
| 5 | Upstashupstash.com | 8 | 4% | |
| 6 | Vercel Queuesvercel.com | 3 | 1% | |
| 7 | Cloudflare + Upstash | 2 | 1% | |
| 8 | Renderrender.com | 1 | 0% | |
| 9 | Built in-houseoutcome | 1 | 0% |
By agent, by persona, by wording
By agent
| Claude Code · Claude Opus 572 runs | AWS · 40then Google Cloud · 12 |
| Codex · GPT-5.6 Sol72 runs | AWS · 48then Google Cloud · 12 |
| Cursor · Grok 4.671 runs | AWS · 45then Google Cloud · 12 |
By persona
| Senior engineer143 runs | AWS · 97then Cloudflare · 17 |
| Enterprise team72 runs | AWS · 36then Google Cloud · 36 |
By what the ask stressed
| Volume and cost at scale84 runs | AWS · 44then Google Cloud · 12 |
| The plain ask83 runs | AWS · 53then Google Cloud · 12 |
| Portability, no lock-in24 runs | AWS · 24 |
| Self-hosting, privacy or residency24 runs | Google Cloud · 12then AWS · 12 |
A case is one codebase with one agent, asked several times in different words and as different people. 8 of 18 cases did not hold to a single cloud.
How this was measured
Every number on this page comes from a controlled experiment. We took 6 small applications, asked 3 coding agents (Claude Code (Claude Opus 5), Codex (GPT-5.6 Sol), Cursor (Grok 4.6)) to pick a cloud for each of them, in several wordings and as a senior engineer and enterprise team, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 215 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 30 runs. Read the methodology and the publications.
Open the interactive boardThis page as Markdown