Artificial Intelligence

AI News Roundup: Huang Exits Nvidia, Baidu Bets on 'Intelligent Organizations', SpaceX Drops $74B on Compute

By 墨林陌 · Nov 13, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

A founder-CEO leaving the world's most valuable chip company, with DOJ antitrust probes running in parallel, forces every AI buyer and vendor to re-plan around Nvidia. The retail token sale and Microsoft's headcount adjustment show AI spending and job displacement turning into ordinary market events.

Summary

Jensen Huang resigned from Nvidia's president, CEO, and board roles effective immediately after 33 years at the helm. CFO Colette Kress took over as interim CEO while the board hunts a permanent successor. Nvidia's pre-market stock fell about 10%, and the US Justice Department is running multiple probes into the company, including one focused on antitrust, alongside investigations touching Megaspeed and Supermicro.

At Baidu World 2026, Robin Li declared 'the next act of artificial intelligence is the intelligent organization.' Baidu released a large-model trading platform called Token Exchange, an ERNIE 5.5 Flash preview, and four upgraded agents. Nvidia also signed a Seoul MOU to wire Korean startup Rebellions' R200 NPU into NVLink Fusion for GPU-NPU heterogeneous systems with SK Telecom, aiming for a 'full-stack sovereign AI' stack supplied from 2028.

The application layer moved too. Doubao Work added Kimi K3.1 Pro and other third-party models to its own two. AI tokens sold 2 trillion units in 30 minutes on Singles' Day, giving compute a retail price and storable quota for the first time. Microsoft is adjusting nearly 60,000 employees, and SpaceX plans $74 billion in supercomputing facilities across Texas and Arizona, including a $50 billion Arizona base.

Takeaways

— Jensen Huang stepped down from Nvidia's president, CEO, and board roles effective immediately after 33 years; CFO Colette Kress is interim CEO.

— Nvidia's pre-market stock shed about 10% on the resignation news.

— The US DOJ is advancing multiple investigations involving Nvidia, including a direct antitrust-focused probe.

— Baidu World 2026 declared 'intelligent organization' as AI's next act and introduced a large-model trading platform called Token Exchange.

— SpaceX plans $74 billion in supercomputing infrastructure across Texas and Arizona, including a $50 billion Arizona base.

— SK Telecom and Nvidia signed an MOU with Rebellions to connect the R200 NPU to NVLink Fusion for GPU-NPU heterogeneous systems, with supply expected from 2028.

— Doubao Work added Kimi K3.1 Pro and other third-party models alongside its own two in-house models.

— AI tokens sold 2 trillion units in 30 minutes on Singles' Day, giving compute its first retail price.

— Microsoft is adjusting nearly 60,000 employees in a round that surfaces AI-driven job replacement.

— China's MIIT issued a notice regulating agent terminal network access, covering both hardware and embedded AI.

Conclusions

The 10% pre-market drop gives a concrete number for the 'Huang premium' investors had priced into Nvidia, and the legal scrutiny landing at the same moment compounds the repricing.

Baidu is moving the competitive argument from model specs to how organizations deploy agents, which is where enterprise budgets actually sit.

Selling AI tokens as retail inventory on Singles' Day turns compute into consumer goods; the '2 trillion in 30 minutes' figure is marketing, but the shift to e-commerce channels is a real distribution change.

Microsoft's nearly 60,000-employee adjustment marks AI-driven restructuring graduating from forecast to actual headcount decisions at scale.

Concepts & terms

Token Exchange (词元所)

Baidu's large-model trading platform where AI tokens are bought and sold as a commodity, following the naming pattern of Chinese stock exchanges. It anchors Baidu's bet that AI's next phase is the 'intelligent organization.'

Intelligent Organization (智能组织)

Baidu's thesis that the next act of AI is not better models but organizations where autonomous agents coordinate planning, scheduling, and execution across business functions.

NVLink Fusion

Nvidia's interconnect that links its GPUs with third-party NPUs, here Korean chipmaker Rebellions' R200, into a heterogeneous GPU-NPU computing system scheduled to ship from 2028.

Full-Stack Sovereign AI (全栈主权AI)

The Nvidia-SK Telecom framing for AI infrastructure a country owns and controls end to end, from chips to data centers, in this case SK Telecom's self-built AIDC.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗