Jensen Huang Resigns as NVIDIA CEO, Kress Takes Over as Interim CEO
On November 12, NVIDIA announced that Jensen Huang, who led the company for 33 years, has resigned as President, CEO, and Director, effective immediately, with CFO Colette Kress taking over as interim CEO. Previously, Huang pushed NVIDIA to support customers in building data centers through investments, credit guarantees, and compute purchase commitments. As of July 26, the book value of equity investments reached $99 billion, and total compute purchase commitments amounted to $36 billion. In August, the company announced a plan to partner with six Wall Street institutions to mobilize more than $500 billion in third-party capital, but no actual transactions have been facilitated to date, and some institutions are demanding that NVIDIA raise its residual value support level by about 25%.
Why read this
SelectedAI score 81
The article reviews the scale of NVIDIA's financing guarantees and compute purchase commitments after Jensen Huang's departure, as well as the resistance caused by Wall Street's disagreement over GPU residual values.
Topics
Body · AI translation
Jensen Huang is gone—can the "compute central bank" still stay open? Letter List Download the app for exclusive early access November 15, 2026, 12:49:59 From Beijing
"A deer caught in headlights," "a CFO who only handles financial reporting."
At an internal meeting in August this year, Jensen Huang publicly interrupted Colette Kress's report and made those remarks about her.
At the time, the two were clashing over Nvidia's ever-growing AI financing business. Huang wanted to use real money and credit guarantees to build data centers and buy GPUs for customers, while Kress kept her eyes fixed on risk and return and pressed him with repeated questions.
Three months later, this internal clash took a highly dramatic turn.
On November 12, founder Jensen Huang, who had led Nvidia for 33 years, was announced to have resigned as president, CEO and director, effective immediately.
The person who took over as interim CEO was none other than Kress, whom he had once criticized for "not understanding strategy."
Under Huang's push, Nvidia had long since gone beyond selling GPUs. Through massive investments, guarantees and compute procurement commitments, it had become deeply involved in the financing and expansion of the entire AI industry, gradually taking on the role of a "compute central bank."
According to foreign media reports, at the time Huang's son-in-law Nico Caprez and another vice president, Raj Mirpuri, began frequently contacting AI cloud service providers and data center developers, pitching them an alluring "financing support plan": Nvidia would provide credit backing to help customers obtain bank loans at lower cost to build facilities and buy chips; in exchange, Nvidia would take a share of these cloud service providers' future revenue.
The two even told potential partners that Nvidia could put together a full package of financing support within weeks. Dozens of companies came forward upon hearing the news, trying to hitch a ride.
Just as the business was advancing at full speed, Kress began pouring cold water on the plan, raising a series of pointed questions that struck at its core:
By what standards would Nvidia select the recipients of support? How should the real risk of financing guarantees be priced? And how should the expected return on investment be calculated?
According to people familiar with the matter, Kress even tried to directly constrain Caprez's activities—Huang's son-in-law, who had just been promoted to Nvidia vice president in January this year.
In July this year, Nvidia was discussing with OpenAI support for a mega data center in Ohio. OpenAI planned to lease the entire campus and deploy about $300 billion worth of Nvidia chips. To help the project obtain bank loans, Nvidia considered providing financing guarantees of up to $250 billion.
On July 26, the media disclosed the plan. The next day, Nvidia's stock fell about 5%, wiping out about $250 billion in market value.
In the weeks that followed, Kress stepped into the deal and renegotiated, cutting the scale of the guarantee Nvidia had initially planned to provide by about half. It was not until August 17 that Nvidia signed this "slimmed-down" agreement with project developer SB Energy, with the guarantee cap for the initial project ultimately set at $105 billion.
Faced with the market alarm bell, Huang himself also began to wonder whether this financing expansion had gone too far. At the August internal meeting, he displayed anxiety about the future: major cloud providers were planning to adopt alternative chips in an attempt to break free of their dependence on Nvidia, while the massive volumes of wafers and memory chips the company had previously committed to purchase were like an oncoming "avalanche."
Since then, Huang has asked the team to establish a clearer guarantee evaluation mechanism that takes into account the GPU sales revenue a deal might generate. Nvidia also paused some financing support deals, and the related projects were subjected to stricter executive and board review.
Across Silicon Valley, head-on clashes between CEOs and CFOs are actually nothing new.
In April this year, the media disclosed a similar power struggle inside OpenAI. CEO Sam Altman was pushing the company to rush to an IPO in the fourth quarter of 2026, while CFO Sarah Friar worried that the company was not yet ready, especially since massive compute spending commitments could bring heavy financial pressure.
In June, the media further reported that Altman had argued with Friar over the IPO timetable and demanded that she find a way to clear the obstacles, or else find bankers and lawyers capable of completing the listing work.
However, the friction did not escalate into a rupture. Friar remains firmly in the OpenAI CFO seat and said in a CNBC interview on November 6 that the company was still considering a 2027 IPO.
Ultimately, such conflicts largely stem from the different functions of the two positions. The CEO needs to coordinate growth goals and strategic direction, while the CFO must ensure that expansion plans have sufficient financial support and raise questions about the risks involved.
Kress was not always opposed to Nvidia's financing expansion. On the earnings call in August this year, she also publicly responded to outside doubts about "circular financing," saying that demand from some frontier AI labs was growing faster than their balance sheets and credit capacity could support. Therefore, she believed it was necessary for Nvidia to help these companies obtain more compute.
Judging from what existing reports have disclosed, Kress's questioning of some deals was in fact more like a diligent CFO's routine gatekeeping on risk and return, far from enough to prove that she opposed Huang's overall strategy.
Now it is Kress's turn to make the decisions.
In the past, Kress could stand in the CFO's position, raise questions about aggressive proposals, and ultimately it was Jensen Huang who made the final call on the company's strategic direction. But now, having taken the top decision-making seat, Kress must simultaneously bear the dual responsibility of controlling risk and sustaining growth.
After news of Huang's departure emerged, the founder of an emerging AI cloud service provider (Neocloud) described it to the media as feeling like "a bomb had exploded."
According to TechCrunch citing PitchBook data, since 2026 Nvidia has participated in 105 rounds of startup financing, and was often the largest investor in them.
From CoreWeave and Crusoe, which specialize in GPU rental, to the French large-model unicorn Mistral AI, and on to a host of rising players such as Lambda and Black Forest Labs, this massive flow of capital has penetrated every corner of the AI industry chain.
Nvidia's latest financial report further reveals the scale of this investment network: as of July 26, 2026, the book value of the company's equity investments had reached as high as $99 billion, with another $25 billion in future equity investment commitments.
For AI startups hungry for compute, purchasing GPUs is often the heaviest capital expenditure on the entrepreneurial path. And Nvidia holds two of the scarcest resources in the entire industry at the same time: chips and cash.
According to media reports, when some infrastructure startups plan to purchase $100 million to $500 million worth of GPUs, Nvidia will even invest about half of that amount, lifting them to begin fundraising at valuations exceeding $1 billion.
Moreover, in Silicon Valley's venture capital circles, having the name "Nvidia" on the shareholder list is itself a highly valuable mark of credibility. When other investors see that the world's largest GPU supplier is willing to put in money, they too feel more confident about participating in the financing.
Huang's ambition went far beyond being a financial or strategic investor.
Since the beginning of this year, Nvidia has also begun using credit guarantees and long-term compute purchase commitments to directly solve customers' funding difficulties in building hyperscale data centers.
The OpenAI Ohio project mentioned earlier is a typical case of Nvidia's credit guarantees. Compute purchase commitments bind the interests of buyer and seller even more directly: Nvidia first sells GPUs to cloud service providers, then commits to purchasing the compute services provided by these GPUs over the coming years.
According to Nvidia's financial report for the second quarter of fiscal year 2027, the company has signed this new type of agreement with multiple AI cloud service providers. As of July 26, the total amount of related compute purchase commitments had climbed to $36 billion, with agreement terms typically lasting as long as six years.
Under the agreement, if a cloud service provider finds an external customer in the market willing to pay a higher price, it can prioritize selling compute to that customer, and Nvidia's backstop purchase amount will be reduced accordingly; when certain conditions are met, Nvidia can take a share of the cloud service provider's revenue from third-party customers.
As a result, Nvidia's chip sales and customer financing have formed a mutually reinforcing relationship: Nvidia supports customer expansion through investments, guarantees, and compute purchase commitments; after customers obtain funding, they increase GPU purchases, and then earn revenue by renting out compute.
On August 10, Nvidia also announced that it had signed memoranda of cooperation with six Wall Street financial institutions—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—planning to establish an independent AI compute financing platform, with the goal of mobilizing more than $500 billion in third-party capital over the long term.
Under the announced plan, the financial institutions are responsible for independent assessment and funding, while Nvidia provides compute infrastructure and may provide limited residual value support for individual projects.
In his view, GPUs can not only continuously generate rental cash flow and flexibly serve different customers, but can also extend their economic life through software iterations. Since power plants, highways, and communication networks can all be built relying on long-term project financing, AI data centers can do the same.
That this compute central bank could rise at such an astonishing speed stems largely from Huang's extremely distinctive personal will.
According to media reports, when it comes to investment, Huang is hands-on in everything—he personally studies startups' underlying technology, uses his own connections to help them connect with customers, and even personally sits in on financing conference calls to lend his support and help entrepreneurs raise investment.
Umesh Padval, managing partner at Seligman Ventures, who has known Huang for a long time and once served as a director of the network chip company Mellanox, which Nvidia acquired, revealed that many of Nvidia's major startup investment decisions were often made by Huang himself.
But the problem is that this kind of influence, built on the founder's personal charisma, technical intuition, and appetite for risk, cannot be directly handed over through a personnel appointment.
After Huang's departure, Nvidia still has ample funds, leading GPU products, and a vast investment network. It is just that how much capital this financing system, which he personally drove, can still mobilize and how large a scale of expansion it can support will no longer be decided by Huang.
On November 5, the "blue wave" triggered market concerns about stricter regulation of data centers, and the U.S. AI industry chain suffered concentrated selling. The Philadelphia Semiconductor Index plunged about 12.2% in a single day, Nvidia fell about 9%, and AMD fell about 17%. Neoclouds that rely on external financing for expansion also suffered heavy losses, with share prices of companies such as CoreWeave and Nebius continuing under pressure.
Nvidia's plan to leverage Wall Street capital has also encountered considerable resistance.
As early as October, foreign media reported that some Wall Street lenders were skeptical about the long-term residual value of GPUs. Nvidia believes high-end GPUs can continue generating revenue for as long as a decade, but some banks are accustomed to assessing depreciation on a three-to-four-year cycle.
Competitor Broadcom, meanwhile, provided residual-value guarantees of up to about $43 billion in a $63 billion financing arrangement for Anthropic. According to a filing Broadcom submitted to the SEC, its calculation base is as high as 72.5% of unpaid lease payments (after further deducting proceeds from equipment disposal). By contrast, Nvidia is still going back and forth with Wall Street over how to share the risk of chip depreciation.
On November 12, foreign media further reported that the $500 billion financing partnership Nvidia announced in August has yet to lead to any actual transaction. Some financial institutions are demanding that Nvidia raise its previous residual-value support level of about 25%, keeping more of the chip depreciation risk on its own books.
According to another report by The New York Times, at least one of the six Wall Street financial institutions participating in the plan discussed withdrawing from the partnership on the day Huang Renxun stepped down.
People familiar with the matter said Nvidia may not be able to announce the first transaction under the partnership until as early as January 2027. Whether that timetable can be met will also depend on whether Huang Renxun's departure further disrupts the pace of negotiations.
With Huang Renxun's departure, several investors interviewed by TechCrunch expect that Nvidia may in the future choose investment targets more cautiously and scale back its investments in existing startups.
These judgments have not yet been confirmed by Nvidia management, and Cress is only interim CEO while the board is still searching for a permanent successor. But no matter who ultimately takes over, the business environment Nvidia operates in has already changed direction.
In addition, Nvidia has already locked in a large amount of supply upstream. As of July 26, the company's supply and capacity purchase commitments reached $279 billion, more than double the $119 billion in the previous quarter. Once downstream customers slow data center construction due to financing difficulties, Nvidia may also face pressure from orders falling short of expectations and difficulty adjusting upstream procurement.
Investors interviewed by TechCrunch pointed out that with Huang Renxun's departure, Google, Amazon, Microsoft and AMD may all increase investment in AI companies to promote their own chips or cloud services. They may not be willing to write hundreds of millions of dollars in checks for infrastructure startups that have not yet generated revenue, as Huang Renxun did, but the investment gap Nvidia leaves behind is still enough to provide competitors with new entry points.
"Special statement: The above content (including videos, pictures and audio if any) is uploaded and published by users of "Dafeng Hao", a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."