After Jensen Huang's Resignation, Silicon Valley Pressures the Board: Why Is It Hard for Nvidia to Cut Ties with Him
After Jensen Huang resigned as Nvidia CEO, Silicon Valley executives and shareholders are lobbying the board to reinstate him; a petition has collected more than 158,000 signatures, of which more than 7,500 people claim to be Nvidia employees. The article argues that Jensen Huang is deeply bound to Nvidia: he maintains supply chain relationships with TSMC, SK Hynix and others through verbal agreements, and leads internal collaborative design meetings of about 60 people. Jensen Huang holds only 3.58% of Nvidia's common stock, and the board has approached Arm CEO Rene Haas as a potential successor.
Why read this
AI score 77
From shareholder pressure and verbal supply chain agreements to internal collaborative design, this article breaks down the specific links that deeply bind Jensen Huang to Nvidia.
Topics
Body · AI translation
The resignation of NVIDIA CEO Jensen Huang, which seemed a done deal, has taken a new turn.
On November 16, foreign media reported that Huang is discussing with his "close friends" how to return to NVIDIA and continue as CEO, and that NVIDIA's allies in Silicon Valley have also begun pressuring NVIDIA's board of directors to demand Huang's return.
A company can change CEOs, but NVIDIA is rather special—it seems it cannot really replace Huang, as the two have long been deeply bound together.
After Huang's resignation, NVIDIA plunged about 14% at the open the next day, with the decline narrowing to 4.34% at the close, and its market value fell below $4 trillion. During intraday trading on November 13, NVIDIA fell more than 3% again, while AMD, Broadcom, and Intel fell about 9%, 6%, and 5% respectively, and TSMC, which manufactures for NVIDIA, also fell more than 3%.
A Bernstein analyst said: "Jensen Huang is NVIDIA, and NVIDIA is Jensen Huang."
Employees think so too. According to foreign media reports, a post by a senior engineer sent to the internal "suggestion box" before Huang's resignation has accumulated more than 1,600 comments, with most people wanting the board to bring him back. One comment read: "If you believe in the company's mission, bring the boss back."
Shortly after Huang left, NVIDIA conducted an internal survey in which nearly 90% of 232 employees voted for Huang to return.
I think NVIDIA will sooner or later bring Jensen Huang back.
Silicon Valley wants to bring Jensen Huang back
According to foreign media, one of Huang's allies is assembling a "war room" whose members are all Silicon Valley executives, and whose task is to lobby the board to reinstate Huang.
This ally said: "All the important people in tech love Jensen Huang." He added that these lobbying efforts are also supported by Huang himself.
The first to speak out publicly were shareholders.
On November 13, Altimeter Capital founder Brad Gerstner and Atreides Management head Gavin Baker sent an open letter to the board, urging it to "keep all options open, including a possible return of Jensen Huang."
According to data disclosed as of June 30, the two held a combined approximately $2.13 billion in NVIDIA stock.
Their reasoning was that Huang's long-standing leadership had "achieved extraordinary success," yet his departure this time was too sudden, the investigation had not yet produced results, no successor had been decided, and the company's future development roadmap had not yet been planned.
On the same day, Musk said on social media that he had signed a petition calling for Huang's reinstatement, and also posted the signing link online.
By the weekend, the petition demanding "bring Jensen Huang back" had gathered more than 158,000 signatures.
Although signatures can be made without real names, according to information already confirmed on social media, they even include Epic Games CEO Tim Sweeney and Y Combinator President Garry Tan.
Uber co-founder Travis Kalanick also signed, writing: "He made a mistake, yes, everyone makes mistakes. But I don't think anything he did rises to the level of separating him from his company."
Among the signatories, more than 7,500 claimed to be NVIDIA employees, equivalent to 18% of the company's total workforce.
But wanting to return and being able to return are two different things.
The core reason is that Huang does not hold control of NVIDIA.
For example, Zuckerberg and Musk both have special shares that let them control their companies, but Huang only holds 3.58% of NVIDIA's common stock, and there is little he can do about decisions made by the board.
However, NVIDIA's board has already taken some action.
On November 13, NVIDIA's board had already approached Arm CEO Rene Haas as a potential next CEO of NVIDIA. Yet the board's statement was that no "formal" process had yet been initiated.
There is precedent in Silicon Valley for a founder being ousted and then returning.
Jobs left Apple in 1985; at the end of 1996 Apple acquired NeXT, and he returned first as an adviser, only taking over as interim CEO in 1997.
Altman was faster. On November 17, 2023, he was removed by the board, and more than 700 of 770 employees signed a joint letter demanding his reinstatement and the board's resignation, otherwise they would all leave. Five days later, he was restored to his position.
Moreover, rehiring NVIDIA executives is not the first time this has happened at all.
In August this year, Ajay Puri, who had managed sales for many years, stepped down and then became a senior adviser to the company; he had worked at NVIDIA for 21 years. When Jobs returned to Apple back then, his first role was also as an adviser.
If NVIDIA does not act, others will. Around the time the resignation news came out, someone on social media tagged Musk and asked why he did not recruit Jensen Huang to his company, and Musk replied: "That would be an honor."
NVIDIA cannot do without Jensen Huang
Generally speaking, a company of NVIDIA's scale can often survive the departure of its CEO, because the rules of how a company operates are already deeply entrenched—everyone knows who should be where and what they should do.
Yet without Jensen Huang, many things at NVIDIA would not be able to proceed.
On November 14, Musk said on social media: "The reason Jensen Huang managed NVIDIA's supply chain so excellently is that many of the deals were verbal agreements he negotiated himself, not formal written contracts. Without Jensen Huang, at least some key deals would be lost."
Musk said this is how he himself does business with Jensen Huang.
He said: "As long as Jensen Huang has promised me when the compute will be delivered, or I ask him to move certain components up, he always gets it done. Of course, we'll fill in a purchase order afterward, but I never doubted for a second whether it would happen."
Previously, SemiAnalysis analyst Dylan Patel said in an interview that when Jensen Huang communicates with the small suppliers in the "Da chain," he calculates demand in advance, then goes to the executives of these companies one by one.
"Even for a small company with only a few hundred million dollars in annual revenue, as long as it has one critical component, Jensen Huang will negotiate with them in person, persuade them to sell him their supply allocation, and then build up the capacity," Patel said.
NVIDIA's chips are manufactured by TSMC, memory mainly comes from SK Hynix, and the racks are assembled by manufacturers such as Foxconn and Quanta.
In May of this year, Jensen Huang said that NVIDIA's partners in Taiwan, China had grown from 10 in the early years to 150.
In February 2026, media reports said that NVIDIA's estimated revenue contribution to TSMC was $33 billion, about 22%, surpassing Apple's $27 billion to become TSMC's largest customer.
NVIDIA booked 800,000 to 850,000 CoWoS wafers in 2026, accounting for more than 60% of TSMC's capacity. Broadcom, in second place, could only get just over 240,000 wafers, which is dwarfed by comparison with NVIDIA.
With TSMC's capacity locked up by NVIDIA alone, other companies that have designed chips and want to tape out or package them can only go to the back of the line first.
As early as December last year, there were rumors that Apple and Qualcomm had begun evaluating Intel's packaging technology because they could not get a slot at TSMC.
By the time competitors finally make it to the front of the capacity queue, NVIDIA's next generation is already shipping.
Therefore, without Jensen Huang "showing his face" everywhere, NVIDIA would inevitably lose its core competitiveness.
In January of this year, he went to Taiwan, China for four days. Right after landing, he visited Morris Chang, the two had dinner together, and afterward he socialized with TSMC's senior executives.
On May 28, for the fifth time, he hosted a supply chain dinner at the Brick Kiln Ancient Flavor Nostalgic Restaurant in Taipei. C.C. Wei, Young Liu, Barry Lam, C.C. Lai, and Jonney Shih all attended. Taiwan media called this meal the "Trillion Yuan Banquet."
During the meal, Jensen Huang went outside to hand out boxed meals to onlookers, and C.C. Wei handed them out with him, saying as he did: "Hand them out faster, like my chips."
The fact is that for NVIDIA, a dinner gathering works better than a contract.
A Morgan Stanley research report last year mentioned that Jensen Huang's trip to Taiwan, China led TSMC to decide to add 20,000 wafers per month of 3-nanometer wafer capacity for Rubin.
Suppliers dare to expand production in advance because everyone believes in Jensen Huang's word. Whatever he says he needs next year, everyone stocks up accordingly.
The downstream works the same way.
Jensen Huang promised to purchase up to $36 billion in cloud services from "neocloud" providers that buy its chips to build data centers, and to provide up to $105 billion in conditional guarantees for OpenAI and SB Energy's data center project in Ohio.
Columbia Business School professor Van Nieuwenhuyse estimates that such guarantees and commitments add up to about $400 billion.
In essence, this is a circle.
NVIDIA invests money in customers, or guarantees for customers; customers get the money and buy NVIDIA's chips; after the chips are sold, it becomes NVIDIA's revenue, and NVIDIA then uses that money to invest in the next one.
It is just like the mooncake and crab voucher coupons given out during holidays. Others who get the vouchers can convert them into cash. So on balance, the manufacturer does not need to actually produce mooncakes and crabs at all, yet demand and value are generated.
The largest of these is the $500 billion credit plan. Jensen Huang also negotiated it himself, one by one.
According to foreign media reports, Jensen Huang personally contacted the heads of several Wall Street institutions, including Goldman Sachs CEO David Solomon, and proposed that everyone form a consortium to lend jointly.
In August of this year, Jensen Huang announced cooperation with six institutions—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to provide more than $500 billion in funding for AI compute infrastructure.
Jensen Huang "closed the loop" on NVIDIA
In addition to external cooperation, Jensen Huang is also the "core component" of NVIDIA's internal operating loop.
In March of this year, Jensen Huang appeared on a Lex Fridman podcast. The host asked him what is the hardest part of designing so many things at the same time.
Jensen Huang replied: "The problem now is that GPUs alone can no longer accelerate things. You only have 10,000 machines, but you want a million times the speed, so you have to break the algorithm apart and slice the data and model out. Once you do that, then at that point everything becomes important. CPU, GPU, network, switches—if any one link is slow, the whole thing slows down with it."
This is the extreme co-design proposed by Jensen Huang.
From architecture, chips, systems, and system software to algorithms and applications, the entire stack is optimized together, and power supply and cooling must also be factored in.
For Jensen Huang himself, this approach means meetings.
He said on the podcast that more than 60 people report directly to him, most of them engineers—some who understand CPUs, some GPUs, some algorithms, some design. He does not meet with any one of them individually, "because that's impossible."
His approach is to call everyone together: "We throw out a problem, and then everyone attacks it together."
More than 60 people sitting in one meeting—it's impossible for every topic to be relevant to everyone.
Jensen Huang says: "Anyone who wants to zone out can zone out. The people on this team all know when they need to perk up their ears and listen." But if someone stays silent at a point where they should be contributing, he will call it out on the spot.
In meetings, he does not give conclusions directly; he lays out the entire reasoning process step by step. He says the benefit of doing this is that everyone "doesn't have to disagree with your conclusion—they can disagree with your reasoning steps."
In 2023, every time a new generation of product was developed, more than 50 people had to meet with Jensen Huang to co-design the new product. Later, Jensen Huang carried out a round of streamlining, and by October 2025, the number of people involved in co-design had dropped to 36.
However, as the business grew larger, Jensen Huang personally stated in a March interview that the scale of co-design had reached 60 people.
This group includes not only CTO Michael Kagan and NVIDIA Chief Scientist Bill Dally, but also technical executives such as Hardware SVP Andrew Bell, as well as functional executives such as CFO Colette Kress (currently NVIDIA's interim CEO), General Counsel Tim Teter, and Corporate Marketing Vice President Laura Fay.
It is fair to say that anyone connected to Jensen Huang has to come to co-design.
When it comes to products, the most typical example this year is Vera Rubin.
At CES in January, Jensen Huang released this generation of platform. It swapped in six chips at once: the Vera CPU, Rubin GPU, NVLink 6 switch chip, ConnectX-9 network card, BlueField-4 DPU, and the Spectrum-6 Ethernet switch chip.
At GTC Berlin on October 21, he put Vera Rubin on stage again, saying it had entered full mass production, was 100% liquid-cooled, and could be assembled in less than five minutes.
Later, Jensen Huang carried out another round of streamlining. In November this year, NVIDIA's latest filings showed that there were still 33 people reporting directly to Jensen Huang.
When setting goals, Jensen Huang uses another method called "speed of light thinking."
Jensen Huang says "speed of light" is his shorthand for the limit of what physics can achieve. "Everything we do is compared against the speed of light"—memory speed, compute speed, power consumption, cost, time, headcount, manufacturing cycle, all the same.
He gave an example. Someone told him that something currently takes 74 days, and that they could get it down to 72 days for him. He did not want to hear that. He would rather tear it all down and reset to zero, first asking how fast physics allows it to be at the fastest, and then looking back at what it would actually take to reach the upper limit set by physics.
The concept of speed-of-light thinking runs through almost all of NVIDIA.
With Hopper-generation H100, the area of a single chip was already close to the upper limit of what a lithography machine can do in one exposure. If you want to improve performance by making a single chip larger, there is no room left.
So with Blackwell, NVIDIA put two chips that were both at this limit into the same package, connected them with a link of 10TB per second, and externally it is still one GPU. This is what is commonly called the chiplet approach.
But the cost is a surge in power consumption. A single GPU rose from around 700 watts to about 1 kilowatt. H100 could still use air cooling, but a single Blackwell rack is 120 to 130 kilowatts, so only liquid cooling will do.
This is obviously not the approach of changing 74 days into 72 days.
Further ahead is Rubin Ultra, planned for mass production in the second half of 2027, with about 600 kilowatts per rack, and power supply simply changed to 800 volts DC.
When chip area hits the limit, they combine chips; when air cooling can no longer handle it, they switch to liquid cooling; when power can no longer be delivered, they change the power supply.
Sooner or later, NVIDIA will have to bring Jensen Huang back.
This article comes from the WeChat public account "Letter AI," author: Miao Zheng, published by 36Kr with authorization.