Cohere brings confidential computing to Model Vault with NVIDIA GPU isolation
cohere · x · 2026-09-17
Cohere announced that Confidential Computing is now available in its Model Vault dedicated inference platform, with early access open to a limited beta group.
- End-to-end encrypted inference keeps data from being tracked or used
- Hardware-enforced isolation extends to the GPU via NVIDIA Confidential Computing
- Customers can request a signed attestation token, independently verifiable against the hardware vendor's public keys at any time
- Aimed at enterprise and sovereignty-sensitive deployment scenarios: "Your data under your control"
Related event: Cohere Brings Confidential Computing to Model Vault(4 posts)→
More from Infra
- AWS shows NVRx fault-tolerant FSDP training on EKS: sync checkpointing ate up to 40% of wall time — AWS ML Blog · 2026-09-17
- Qwen 3.8 27B NVFP4 benchmarked on 2xV100 with ~400k context — jjusko20 · 2026-09-17
- NVIDIA launches CUDA Rust: two paths to write GPU kernels natively in Rust — blelbach · 2026-09-17
- AI agent ports and optimizes CUDA attention kernel to CuTeDSL in an afternoon, 1.2x faster — knowrohit07 · 2026-09-17
- llama.cpp PR Enables CUDA Graphs for MTP Draft, Delivering Another Inference Speedup — jacek2023 · 2026-09-17
- Open-Source Maestro: One-Click Local AI Creative Studio for Music, Video and More — cocktailpeanut · 2026-09-17