How Exelerit compares
An honest comparison, including where the others win. Cells we have not sourced are marked, not invented.
| Axis | Exelerit | RunPod | Vast | Lambda | Modal |
|---|---|---|---|---|---|
| Price transparency | Rate in the calculator = rate you pay; per-second; spend caps. source ↗ | Clear on-demand rates; community-tier variance. (unsourced) | Bid/auction; cheapest but variable. (unsourced) | Clear list; no per-second granularity. (unsourced) | Usage-based; harder to predict at spiky volume. (unsourced) |
| Reliability | Owned metal + public SLOs + status page. source ↗ | Mixed (community vs secure cloud). | Borrowed consumer GPUs; a host can vanish. | High (datacenter). | High (managed). |
| L40S availability | Core tier — our focus. core tier | Available. | Spotty / consumer-card heavy. | H100-centric; L40S secondary. | Abstracted; you don’t pick the card. |
| Egress | Flat, explicit. | Varies. | Varies by host. | Varies. | Bundled / opaque. |
| Developer experience | One-click + API/CLI, web-shell. source ↗ | Strong; large template/community library. | Thin; DIY. | Solid; cluster-oriented. | Excellent serverless abstractions (proprietary). |
Where they win
- RunPod — Template & community breadth.
- Vast — The absolute price floor.
- Lambda — H100 frontier-scale clusters.
- Modal — Managed-inference polish.
- Exelerit — We own the L40S tier — that is the lane we play.
Where we don’t play
We don’t chase the H100 trophy/frontier-training market, and we are not the absolute consumer-GPU price floor. Stating this is the point.
Last updated 2026-06-23. Figures we cannot source yet are marked rather than guessed.