This is the AI Affairs briefing for Sunday 15 November.
First, Nvidia’s board is facing a public call to leave the door open for Jensen Huang. Huang, who co-founded the chipmaker and led it for 33 years, resigned as chief executive, president and director this week. Chief financial officer Colette Kress is serving as interim CEO while the board considers a permanent successor.
Two investors, Altimeter Capital and Atreides Management, want that choice made slowly. In a letter posted on Friday, they urged directors to consider Huang’s potential return. They did not ask for his immediate reinstatement. Instead, they called for what they described as a slow, measured and deliberate leadership process, saying the board should not rush before it has all the facts.
The firms had a substantial stake in the decision. At the end of June, Altimeter held Nvidia shares worth nearly $1.9 billion, and Atreides held shares worth nearly $250 million. Together, their holdings were worth about $2.1 billion. Nvidia was Altimeter’s largest disclosed equity holding in its most recent regulatory filing, CNBC reported. Altimeter chief executive Brad Gerstner posted the letter, which was also signed by Gavin Baker of Atreides.
The investors said Kress could lead Nvidia capably through the transition. She had been the company’s chief financial officer for 13 years. Nvidia announced Huang’s departure on Thursday. The board had accepted his resignation effective Wednesday, Yahoo Finance reported, citing a regulatory filing. Director Harvey Jones became the board’s first chair.
The search is still taking shape. AI Affairs has reported that Nvidia had not chosen a search firm, and that its board had approached Arm chief executive Rene Haas. The investors’ letter gives directors room to consider candidates, while explicitly keeping Huang among the possibilities.
A separate public campaign takes a firmer position. The Stand with Jensen petition asks directors to bring Huang back while an independent investigation establishes the facts, unless the board can show that his removal serves Nvidia’s interests. It also asks for details about the investigation’s scope, independence and timeline, and for Huang to have a chance to respond before a final decision.
That campaign has attracted prominent attention, but it is not the investors’ letter. Elon Musk said on Friday that he was signing the petition. On Saturday, he said some critical deals would be lost without Huang. Inside Nvidia, a separate post asking the board to consider employees’ views had drawn more than 1,600 comments by Thursday, most supporting his return, Wired reported. Those comments are not part of the public petition’s count.
That petition is our next story, because its numbers rose sharply on Saturday. By the afternoon, Stand with Jensen displayed more than 115,000 signatures, TechCrunch reported. Its site listed 5,965 signers as current Nvidia employees, more than a thousand as former employees, and about 63,000 as shareholders.
Those figures describe what signers said about themselves, not verified workforce or shareholder totals. The campaign says affiliations are self-reported and its screening does not verify identities. Its organisers said they were removing fake entries. The website says the campaign is independent of both Nvidia and Huang. Earlier on Saturday, the displayed total had passed 100,000 by mid-morning Eastern time.
Among the named supporters were Musk, Replit chief executive Amjad Masad, Epic Games chief executive Tim Sweeney and Y Combinator chief executive Garry Tan. The petition asks Nvidia’s board to reinstate Huang while an independent investigation establishes the facts. It also calls for a fair investigation and more information for the public.
The calls for an investigation come amid reports of extramarital relationships involving Huang. Huang has admitted to an affair, Yahoo Finance reported. Nvidia’s independent directors said on Tuesday that they had retained former acting US Attorney Joon H. Kim and the law firm Wachtell, Lipton, Rosen & Katz to conduct an independent investigation, The Defiant reported.
Public expressions of support should not be confused with a single position inside Nvidia. Bryan Catanzaro, the company’s vice president of applied deep learning research, said on Friday that an online letter claiming his involvement was fake. He said he trusted the board’s judgement and backed Kress without reservation. The board remains responsible for appointing Nvidia’s chief executive. Nvidia is scheduled to report quarterly results on Tuesday.
The leadership change also puts an earlier financing dispute in a new light. Kress, now interim CEO, reportedly pushed in August to limit Nvidia’s commitments to AI companies and cloud providers that buy its GPUs. One account centres on a proposed guarantee for OpenAI compute.
Wccftech reported that initial discussions contemplated roughly eight gigawatts of compute, valued at up to $250 billion. At a meeting of Nvidia’s senior finance team, Kress reportedly reduced the signed guarantee cap to roughly 4.25 gigawatts, or up to $125 billion. These figures concern potential support for compute purchases. They are not reported payments for GPUs already delivered.
The distinction matters because of how the reported arrangements worked. Nvidia would commit funding to AI companies and cloud operators, which would in turn buy Nvidia hardware. Wccftech described that as circular financing. Kress reportedly wanted to rein in commitments involving smaller cloud providers as well as larger AI firms, including OpenAI and Anthropic. TechPowerUp reported her concern that some smaller providers receiving Nvidia investment could not afford the GPUs without it.
Huang reportedly challenged her effort to scale the deals back. TechPowerUp reported that he called Kress a “deer in the headlights” during the August confrontation. Neither account turns the reported exchange into a complete picture of the companies’ eventual financing arrangements. The OpenAI number was described as a cap on a signed guarantee, not a measure of hardware in use.
The accounts also describe pressure elsewhere in Nvidia’s supply chain. Wccftech reported that the company faced silicon ordered from TSMC and memory for which it had no immediate buyers. It said Huang had privately acknowledged a shift by major cloud companies towards chips they designed themselves. On that account, prepaid component commitments increased the pressure to find other GPU customers. Nvidia reportedly contacted TSMC about reducing forthcoming chip orders.
Wccftech argued that financing buyers for surplus chips could keep additional supply from pushing down GPU prices. TechPowerUp characterised the alleged effect as preserving the appearance of strong demand during a GPU and memory shortage. Those are claims about the possible effect of the arrangements, not reported measurements of supply available to other buyers.
There is a separate question about chips that cloud companies have already bought. Analyst Ed Zitron estimated that at least $200 billion worth of Nvidia AI GPUs purchased by large cloud companies were idle in warehouses, Wccftech reported. The publication said those buyers lacked sufficient electrical capacity or immediate customer demand to run them. That estimate concerns hardware after purchase. Kress’s reported concern was Nvidia’s exposure while helping prospective buyers make a purchase.
Finally, away from Nvidia, DeepSeek is opening more of its consumer app’s functions to developers. The Hangzhou-based AI lab made search, file-reading, voice and safety tools available through its API on Friday, Pandaily reported. Developers can bring some of those functions into a model request, or call certain services separately.
The model is deepseek-flash, currently backed by the DeepSeek-V4.1-Flash-1017 checkpoint. Developers can enable tools through the existing Messages, Responses and Chat Completions APIs. DeepSeek’s servers make the tool calls and place retrieved material into the context the model uses to answer. That means an application already using one of those endpoints can add search or file reading within the request, rather than building every retrieval step separately.
With thinking disabled, the model can search the web or images and answer with citations. With thinking enabled, it can search and read links across several reasoning steps. A URL-reading tool fetches a supplied page. File input extracts text from PDFs, Office documents and images, or parses their layout before adding the contents to a conversation. Search seeks material online; link reading and file input work with material supplied to the application.
Developers can also call Search and Fetch separately. Fetch retrieves a particular page. Optical character recognition, or OCR, turns printed text in an image into machine-readable words. Speech recognition turns audio into text, and text-to-speech generates spoken audio from written output. These are separate services, so an application developer decides how to connect them to a conversation.
DeepSeek has released an open-source demonstration client that shows the functions together. It supports search, link reading, attachments, voice input and thinking mode, and can import chat history from the DeepSeek app. Importing that history replaces the conversations already in the demo. The demo stores its API key, conversations and uploaded images in the browser.
Using the tools adds charges on top of model-token prices. Pandaily listed search and web fetch at 75 cents per thousand calls each, and file parsing and OCR at 22 and a half cents per thousand pages. Text-to-speech was listed at 45 cents per ten thousand characters, and speech recognition at about 12 cents per hour of audio, with both voice services currently half price. At peak hours, deepseek-flash was listed at 30 cents per million cache-miss input tokens and $1.20 per million output tokens; off-peak rates were half that. Startup Fortune said it could not verify a rate on DeepSeek’s own pricing page.
DeepSeek’s model-safety setting checks inputs and outputs and indicates when a request should be declined. The company tells developers serving the public to enable it, while meeting their own duties for content review, model registration and AI labelling. The check can be part of a model request. Responsibility for the service built around that request remains with its developer.
Today’s stories leave two different kinds of decisions in view. Nvidia’s board must choose how to handle its leadership transition amid public pressure and reported financing concerns. DeepSeek’s new tools give developers more ways to build an application, along with separate costs and compliance duties to manage.