Cumulus Labs

StartupHub.ai · Jun 2026

Long-form coverage of Cumulus Labs' Ion engine, GH200 optimization, benchmark claims, and the company's cloud-plus-software positioning.

Agent Wars · Mar 2026

Independent coverage of Cumulus Labs' IonRouter launch, describing a GH200/B200-focused inference service, model multiplexing on one GPU, and the company's throughput and cold-start claims.

Cumulus Labs Blog · Feb 2026

Benchmarks and pricing comparison showing multiple vision-language models running on a single GPU, with per-token and per-GPU-second pricing positions and scale-to-zero economics.

Cumulus Labs Blog · Feb 2026

Technical deep-dive on Ionattention, the C++ inference runtime built for Grace Hopper, including custom CUDA-kernel techniques and benchmark claims such as 588 tok/s on a single GH200 in one cited comparison.

Y Combinator on LinkedIn · Jan 2026

Launch post describing Cumulus Labs as a YC W26 GPU cloud that optimizes training and inference workloads, charges by physical GPU usage, and highlights the founders' backgrounds.