This brief covers AI news from 2026-11-02 UTC.
Saudi money reshapes a national AI champion, US regulators circle Nvidia’s biggest acquisition, and DeepSeek hands builders a new agent harness.
Aramco closes deal for 49% of Saudi AI firm HUMAIN
What happened: Aramco and Saudi Arabia’s Public Investment Fund said on November 1 they closed a deal on HUMAIN. Aramco takes 49% through a subsidiary and puts in about 9.9 billion riyals ($2.6 billion) in cash during 2026 plus its own AI, computing and digital assets; PIF keeps 51%.
Why it matters: HUMAIN says it plans 3 gigawatts of AI data centers by 2030, so builders selling compute, inference or tooling into the region now have one state-backed buyer instead of two.
DOJ asks how Nvidia’s Hugging Face buy affects competition
What happened: Reuters reported on October 28, citing two people familiar with the matter, that the Department of Justice has made inquiries about how Nvidia’s proposed $12.93 billion acquisition of Hugging Face could affect AI competition. The report describes inquiries, not a lawsuit or a decision to challenge the deal.
Why it matters: Hugging Face is where most developers pull weights, so any conditions on the deal could change how models get distributed across chips and clouds.
DeepSeek team open-sources a proactive coding agent
What happened: DeepSeek’s Harness team released dsh-superagent under the MIT license. The prototype takes development requests through GitHub and optionally Feishu, gives each task its own working directory and agent session, returns pull requests to the original conversation, and shows a web panel of items needing human attention.
Why it matters: It is a working reference for long-running agents that file their own PRs, so teams can copy the session and review model instead of building it from scratch.
OpenAI merges Chat and Work modes into one token budget
What happened: The Decoder, citing an updated OpenAI support page, reports that from December ChatGPT Plus, Pro and Business plans share one token budget across ChatGPT and Codex. Simon Willison writes that OpenAI now gives $100, $200 and $500 ChatGPT subscribers matching API credits.
Why it matters: If Codex and ChatGPT draw on the same quota, agent runs eat into chat headroom, so check plan terms before budgeting coding agents on a subscription.
Nvidia claims over 1, 000 tokens a second per user on 700B+ models
What happened: A new Nvidia submission to the SemiAnalysis AgentX benchmark, running on Rubin NVL72, passes 1, 000 tokens per second per user on models above 700B parameters, at 20.9 times the performance per total cost of ownership of a GB300 NVL72 rack at 325 tokens per second per user.
Why it matters: Higher per-user throughput on giant models changes the math for serving large open-weight models yourself rather than routing to a hosted API.
US computer science enrollment falls 12.4% this fall
What happened: Early fall 2026 data published October 29 by the National Student Clearinghouse Research Center show computer and information science enrollment down 12.4% at four-year colleges, 12.3% at two-year colleges and 10.4% at graduate schools, while total enrollment rose 0.2%.
Why it matters: A smaller pipeline of graduates arrives just as teams lean harder on AI coding tools, which shifts hiring toward people who can direct and review agents.
AWS previews live app previews from agent session endpoints
What happened: An AWS post dated October 30 describes AgentCore session endpoints: authenticated URLs into an agent’s microVM, limited to between 1 and 5 ports you allow, for live web previews of a coding agent’s app.
Why it matters: Reviewers can open the running app instead of reading a diff, which removes a manual step from every agent pull request.