NVIDIA Aerial CUDA-Accelerated RAN (open source)

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

The readiness gap, in one scan

AI-assisted assembly · derived results

Evidence strength Strong

Decision

Human editorial judgment · 2026-06-28

Track; not yet

Why: Genuinely open (Apache 2.0) and well-resourced with academic traction (MIT, Northeastern, Virginia Tech) and Sionna lineage (200k+ downloads), but it is a research/SDK substrate requiring NVIDIA hardware, not a drop-in production RAN, and the openness of the core PHY modules is unconfirmed
Next: Clone the repo, run pyAerial + Sionna in the prebuilt NGC container on an available NVIDIA GPU; evaluate cuPHY/cuMAC for AI-channel-estimation research; do not assume modifiable PHY until the blob question (issue #15) is resolved

Constraints

No named blocker is present in the current public projection.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Initial public reading

This is the initial public reading. No earlier readiness change is recorded.