DeAI Daily Brief — 13 November 2026

Today in DeAI: Nvidia installs Colette Kress as interim CEO after Jensen Huang's resignation, a viral report claims Blackwell fingerprints in six Chinese open-weights models, Cohere opens confidential-computing beta, OpenAI's docs confirm a 10% data-residency premium, and Novita AI retires hosted models on November 26.

Nvidia names CFO Colette Kress interim CEO after Jensen Huang resigns

Nvidia's board announced on November 12 that Jensen Huang resigned as President, CEO, and board member effective immediately, ending a 33-year run, and that CFO Colette Kress will serve as interim CEO while a permanent search runs. Harvey C. Jones was appointed the company's first-ever board chair, and Q3 earnings stay scheduled for November 17. Why it matters: Nvidia supplies effectively all GPU capacity open-weights inference runs on, and its open-weights commitments — the $12.93 billion Hugging Face acquisition, the Nemotron line — were Huang's personally. (NVIDIA Newsroom) — our full coverage

Viral report claims six Chinese models carry Blackwell fingerprints

Will DePue's Unemployment Labs report, published November 12, claims that recovered PyTorch init seeds and streaming-multiprocessor layouts match DeepSeek-V4, Kimi K3, Kimi K3.1-Pro, MiniMax-M3, and Ling-3.1 to Blackwell-class GPUs, implying export-controlled hardware despite US restrictions. The report says its own evidence "supports, without proving" Blackwell use, publishes no code, and describes a single analysis pass not yet re-run. Why it matters: if checkpoint forensics becomes reproducible, weight provenance becomes a real due-diligence tool — until then, every claim in it is just that, a claim. (Unemployment Labs) — our PULSE coverage

Cohere opens confidential-computing beta on Model Vault

Cohere opened a public beta of Confidential Computing on Model Vault, with customer-side remote attestation performed before data is released to the inference stack. The move lands in the same week as Google Cloud's Confidential G4 general availability and NVIDIA's confidential bare-metal extension across Grace Blackwell and Rubin. Why it matters: hardware-attested private inference is becoming a default offering rather than a specialist deployment, which changes the compliance math for regulated workloads. (Cohere)

OpenAI pricing docs confirm 10% uplift on data-residency endpoints

OpenAI's API pricing documentation states that regional processing (data-residency) endpoints, along with FedRAMP endpoints, carry a 10% price uplift for models released on or after March 5, 2026. Why it matters: keeping data in-region on a frontier closed API is now a metered premium — a concrete price signal for builders comparing hosted residency against self-hosting open weights, where the region is wherever the GPUs are. (OpenAI)

Novita AI retires hosted models November 26

Novita AI's changelog, dated November 5, lists multiple hosted models retiring at 16:00 UTC on November 26 — about two weeks' notice for anyone pointing production traffic at them. Why it matters: third-party inference hosts can pull models on short notice, which is a recurring pattern this fall; pin the endpoints you depend on and mirror the weights you cannot afford to lose. (Novita AI)

Watching tomorrow

Nvidia's Q3 fiscal 2027 earnings call on November 17 is the first test of the Kress interim tenure — listen for how the open-weights and Hugging Face commitments are described when Huang is not the one describing them.