Huang leaves Nvidia as Google launches Gemini 4 Hydrogen

This morning’s briefing leads with Jensen Huang’s departure from Nvidia and Google’s Gemini 4 Hydrogen launch. There are also Microsoft connector outages, UK growth figures and new releases from Claude and Codex.

Jensen Huang leaves Nvidia and Colette Kress takes charge

What happened. On 12 November, Nvidia announced Huang’s immediate resignation as president, chief executive and director. Colette Kress becomes interim chief executive while remaining chief financial officer. Harvey C. Jones becomes the board’s first chair; the search for a permanent chief executive will consider internal and external candidates.

The departure follows reports of affairs involving employees and businesswoman Alice Huang. Reuters reports that US authorities suspect Alice Huang of spying for China, Beijing denies the allegation, and attorney general Todd Blanche says the Justice Department has reached no conclusion.

Nvidia shares closed 4.3% lower on Thursday. Huang has also been summoned to testify before Congress about the alleged spy affair, the BBC reports.

Why it matters. Nvidia supplies processors used to train and run AI models, as the Guardian explains.

ClearBridge Investments portfolio manager Michael Feldman identifies execution as the risk after Huang’s departure, rather than an immediate loss of Nvidia’s technology roadmap, Reuters reports. The Financial Times’ analysis argues that his departure could unsettle the financial dependencies that helped fund the AI boom.

Export policy is another pressure point: Reuters reports calls from both parties for tighter restrictions on China. For customers, chair Harvey Jones expects operations, commitments and the product roadmap to continue uninterrupted. That is the board’s stated expectation.

People are asking

**Who is running Nvidia?

** Colette Kress is interim chief executive and remains CFO. Harvey C. Jones chairs the board during the successor search, according to Nvidia’s announcement.

**Has the espionage allegation been proved?

** The US Justice Department has announced no conclusion, and Beijing denies the allegation, Reuters reports.

**Have customer commitments changed?

** Nvidia’s chair expects them to continue uninterrupted. Its next financial results are scheduled for 17 November.

Google brings Gemini 4 to free users with Hydrogen

What happened. Google launched Gemini 4 Hydrogen on 12 November. It is rolling out across the Gemini app, Search, enterprise services, Antigravity and eligible Google Sheets subscriptions. Free app users receive it through automatic model selection when appropriate for their prompt.

The API release notes confirm general availability, a free tier and four thinking levels: low, medium, high and xhigh. High is the default. Standard paid API prices are US$1.50 per million input tokens and US$7.50 per million output tokens, including thinking output.

Artificial Analysis measures Hydrogen high at 51 on its Intelligence Index, US$1.87 per index task and 7.9 minutes per task. Gemini 3.8 Flash high scores 31, costs US$7.77 and takes 14.7 minutes. Hydrogen xhigh scores 54, costs US$3.16 and takes 13.7 minutes.

Why it matters. Google is extending its fourth-generation models to Free Gemini users, widening access beyond the paid subscriptions associated with Carbon and Argon.

Our analysis of the independent results: token prices alone give an incomplete comparison. Hydrogen’s standard token rates are twice Flash’s current promotional rates, yet its measured Intelligence Index task cost is about 76% lower than Flash’s. This describes the evaluation workload; it does not establish savings on an organisation’s own tasks.

Effort also changes the result. Moving from high to xhigh adds three index points while increasing measured cost and completion time. That makes the effort setting part of any meaningful model comparison.

People are asking

**Can Free users select Hydrogen manually?

** Google specifies automatic selection for Free users; its announcement does not promise a manual Hydrogen picker on Free.

**Does 51 versus 31 mean every task improves by 65%?

** No. These are aggregate Artificial Analysis index scores, not a percentage improvement applicable to each workplace task.

**Should everyone choose xhigh?

** High is the API default. Artificial Analysis measures a higher score for xhigh, alongside higher cost and longer task completion time.

Microsoft’s outage disrupts ChatGPT and Claude connections

What happened. Anthropic’s incident report began at 17:52 UTC on 12 November. Its latest update, at 22:23 UTC, attributes the failures to a Microsoft 365 incident and says most mail, calendar and search requests are still failing; Claude for Outlook remains affected.

OpenAI separately reports that new Microsoft connections may be broken and tools are not working as expected. The incidents remained unresolved on OpenAI’s and Anthropic’s public pages when checked for the 06:30 GMT edition.

Why it matters. Our analysis: a task using Microsoft-held work data retains that dependency when moved between assistants. Anthropic identifies Microsoft 365 as the upstream cause, while OpenAI reports failures in its Microsoft connections.

A failed retrieval also leaves the assistant without the requested source material. It should not be mistaken for a current, work-grounded answer. The incident’s scope matters: Anthropic lists Claude API and Claude Code as operational, alongside degraded performance in Claude and Cowork.

People are asking

**Is the whole of Claude unavailable?

** No. Anthropic lists its API and Claude Code as operational; the incident affects Microsoft-connected features. Check the component status page.

**Will reconnecting Microsoft 365 fix it?

** That is not an established remedy: OpenAI says new connections may also fail.

**How can work continue?

** Follow the incident updates, identify unavailable sources explicitly and use approved documents already accessible to the organisation where suitable.

AI and cloud businesses help lift UK growth to 0.7%

What happened. The ONS’s first quarterly estimate, published on 12 November, puts UK GDP growth at 0.7% for July–September, following 0.5% in April–June. Reuters reports that this was the strongest calendar-quarter expansion since the first quarter of 2024.

Computer programming, consultancy and related activities grew 4.4%. The ONS says many businesses reporting the largest turnover increases work in AI and cloud computing, but it cannot quantify their exact impact. Services overall grew 0.9%, while construction fell 0.7%, according to the quarterly bulletin.

The monthly bulletin describes cross-industry themes as anecdotal and says dedicated estimates of AI’s economic impact are planned for January–March 2028.

Why it matters. Our analysis: this is evidence of expansion in businesses supplying AI and cloud activity. It does not measure productivity gains from using AI across the wider economy. The ONS has not yet published its dedicated AI impact estimates.

The figures also connect AI spending with the broader economic outlook. Reuters reports that the Bank of England attributes some economic strength to AI spending. Pantheon Macroeconomics economist Rob Wood says the growth figures make another interest-rate rise in December more likely. That is his assessment, rather than a Bank decision.

People are asking

**Did AI generate all the 0.7% growth?

** The data do not establish that. The ONS identifies AI-related strength, without assigning the whole increase to AI.

**Does programming growth prove that every AI user is more productive?

** No. The quarterly release measures economic output; it does not establish an organisation’s return from adopting an assistant.

**Are the figures final?

** No. These are first estimates subject to revision.

New releases

Codex CLI. Version 0.168.0 adds recording attachments for audio-capable models and optional GIF generation and editing, with a separately enabled side-conversation view. More

Claude Meetings. A Google Meet beta is rolling out to Team and enabled Enterprise accounts, with notes and transcripts included and meeting documents and whiteboards free through 10 February 2027. More

Claude for Manufacturing. Team and Enterprise users can install 17 manufacturing skills connected to Autodesk Fusion, Infor, Microsoft 365 and Oracle NetSuite to prepare quotes, quality reports and supplier requests. More

Claude Managed Agents. Approved private-beta customers can run the Managed Agents service and sandboxes in their own cloud or datacentre, while model requests go to their chosen inference provider. More

Claude Code. Version 2.1.323 introduces opt-in experimental Code Mode, in which Claude writes JavaScript to call its tools, with an organisation-level option to disable it. More

Gemini 4 Hydrogen. Hydrogen high scores 51 against Gemini 3.8 Flash high’s 31 on the Artificial Analysis Intelligence Index, at US$1.87 rather than US$7.77 per benchmark task; Free app access uses automatic selection. More

Google Antigravity. Google AI Ultra users are getting Carbon and Argon Orchestrator options that pair a planning model with Hydrogen; paid AI Pro subscriptions and eligible premium-device trials receive the Argon option. More

Elsewhere in AI

Four advocacy groups sued Anthropic, Google, Meta, OpenAI and SpaceXSI in California, seeking to halt new AI model development and deployment over alleged catastrophic risks, AP reports. Read

Australia’s Amanda Rishworth launched guidance urging employers to consult workers about AI adoption and explain the effects on jobs, The Guardian reports. Read

NHS England removed its claim that 137 trusts benefited from Palantir’s system after 29 reported little or no use, according to The Guardian. Read

Caine Prize organisers cleared Zimbabwean author Simbarashe Kundizeza of using AI for The Blue Fighter after a month-long investigation, The Guardian reports. Read