Original Title: "Nvidia and the World Without Jensen"

Original Source: SemiAnalysis

The core company of the AI era——Nvidia's leader Jensen Huang is stepping down amid scandals, regulatory doubts, and geopolitical games, leaving the biggest suspense for the global tech community: Is Nvidia, without its soul figure, still the unique company it once was? As one of the world's leading semiconductor and AI research institutions, SemiAnalysis provides an extremely in-depth exclusive analysis through its profound industrial tracking capabilities. The research report not only reveals the dark lines behind Huang's resignation involving the RASA Act, chip bans, and intelligence community games but also dissects what system-level crises "Nvidia without Jensen Huang" will face from dimensions such as organizational structure, supply chain control, financial safety nets, and open-source ecosystems. The following is the original text of the report:

Jensen Huang is no longer the CEO of Nvidia. The Nvidia board has been investigating Huang's behavior and weighing his future in the company he founded; clearly, behavioral issues do exist—there are ethical problems associated with relationships with a Chinese client and several subordinates.

Moreover, we believe there are deeper reasons behind his resignation merely four days later. We think this is directly related to large-scale GPU shipments avoiding customs regulation. Our accelerator model tracks the production and whereabouts of every AI chip; the data show that over 1.5GW of regulated Nvidia chips ultimately flowed into China. This figure does not even account for the chips specially designed by Nvidia to circumvent US regulations—this report was initially disclosed by us.

We believe that the US intelligence community is a significant force pushing Huang to leave Nvidia. Specifically, we believe that RASA (the Remote Access Security Act) or similar legislation is likely to continue making progress; RASA has already passed in the House of Representatives. With Huang, who has always supported China, departing, AI and chip restrictions in relation to China will accelerate. We firmly support certain restrictive measures, but poorly designed restrictions will harm US interests.

Such legislation will have consequences, and we do not believe politicians have fully considered these consequences. Under the current regime where Chinese laboratories can remotely use GPUs, there are the following clear benefits:

· Profits flow to US companies;

· The US can know what Chinese laboratories are doing;

· The US can, if willing, almost immediately cut off GPU usage rights;

· US companies, such as Nvidia, AMD, Broadcom, Google, Amazon, Microsoft, Meta, Supermicro, and Dell, all rely on Chinese companies in their supply chain;

· Nvidia and AMD employ Chinese engineers residing in China to develop hardware and underlying software for their GPU systems. Without these engineers, the user experience of these GPUs would undoubtedly be much worse than it is now;

· US companies, especially the entire startup ecosystem, benefit from Chinese companies, students, and independent researchers utilizing these GPUs to publish open-source research成果;

· The motivation for China to use domestically manufactured accelerators will be slightly weaker;

· Innovations from China flow globally through open innovation methods such as extensive expert paralleled efforts and open-source kernel libraries, e.g., MSA, DeepEP, DeepGEMM, etc.

That said, we do support implementing certain AI restrictions on China, but these restrictions should be closer to the MATCH Act rather than RASA.

Perhaps the most important point is: open source is under attack. It is an open secret that Nvidia collaborates directly with Chinese laboratories on open-source projects and helps them optimize their GPUs and models. These laboratories use Blackwell GPUs, and Nvidia is aware of this. The US open-source ecosystem and many startups rely on Chinese AI models. With Huang's departure and the rise of hawkish positions, this ecosystem will be severely damaged.

But for shareholders and the AI supply chain, the truly important question is: What will Nvidia be like without Huang?

Our view is that this company was never built to operate without him. He has led for 33 years; even if the board forces him to leave, or he "resigns," Nvidia is still better off with him than without him. The company is uniquely constructed around him. Without this person, the company could become a completely different entity. Without Musk, SpaceX and Tesla cannot operate; without Huang, Nvidia cannot operate either.

A King and His 60 Direct Officials

For a company of such large scale, Nvidia's organizational structure is quite peculiar. According to Huang's public statements, he has about 60 direct reports. He has stated that a CEO should have the most direct reports, as these individuals require the least management, and he does not meet with them one-on-one.

However, few can advance matters in the absence of Huang. The company operates like a kingdom with only one king; regardless of seniority, others merely execute his will. Huang has said that who reports to whom is not important because titles do not correlate with actual importance of work. This is also one of the reasons why Nvidia is considered "the smallest big company in the world": despite its massive market value, it has only about 42,000 employees. There is little internal politics at Nvidia. Huang has stated that this flat network empowers employees and lets the company adapt more quickly to changes, as its technology iterates extremely fast.

This is also one reason Nvidia acts swiftly. Once Huang determines the direction, everyone works together towards it. Former employees have told us that product development and engineering work often extend until the last moment to allow the company to change its mind if necessary. One former executive stated that, for example, the final design of the GB200 was much closer to the product cycle than anyone hoped. Huang has said he wishes the company could forever maintain a startup mentality.

However, the cost of this structure is that Nvidia does not have a deep reserve of successors. Look at Nvidia's website: among Huang's direct reports, only five are considered "well-known," and two of them are legal and financial executives. Most of his direct reports have not been trained to handle the public spotlight, and the company explicitly does not encourage them to seek exposure. Few leave Nvidia to take CEO positions at other large companies; Arm's Rene Haas is one of the few exceptions. Very few at Nvidia can stand out externally because all major decisions are made by Huang. But many in Nvidia's executive team do not mind because the crown bearer bears heavy responsibility; many are happy to work until retirement age because they are well-compensated, treated well, and believe in the company's mission and leadership.

No One Really Wants This Job

Colette Kress has been appointed as interim CEO, but Nvidia is searching for a formal CEO. This person cannot have a financial background. Potential internal candidates being discussed for replacing Huang on the technical side are Ian Buck and Jonah Alben.

Buck created CUDA while pursuing his PhD at Stanford University, joined Nvidia in 2004, and is currently the VP of Hyperscale and High-Performance Computing, overseeing the data center business. When Huang is on a business trip, he is often tasked to handle related affairs and frequently speaks at meetings.

Alben has been with Nvidia since 1997, serving as the Senior VP of GPU Engineering, responsible for overseeing the design of the company’s GPUs.

But the reality is that no one individual can manage all affairs alone. We do not believe Buck has the capability to lead the company's hardware engineering division. Alben and his team have technical autonomy, but even he does not control the entire chip business; other executives responsible for multiple key tech and business divisions report directly to Huang as well. Both are excellent, but may not carry the same supply chain control as Huang.

Additionally, there is another issue: is there anyone willing to take the CEO role? One executive told us earlier this year, "Ian would rather die than take over from Huang; he doesn’t want the job. Jonah is more respected, and people say he can handle it, but really, no one wants that job." However, the situation has obviously changed. Kress is the interim CEO, but integrating over 60 vice presidents into a unified vision for the company requires a very particular kind of person; otherwise, the company's direction risks falling into noise.

Founder-like Bets, Unable to Be Sustained by Professional Managers

Huang is one of the last old-school tech founders, a contemporary of Elon Musk, slightly older than Mark Zuckerberg. Other companies that grew alongside Nvidia, including AMD, Google, Microsoft, Apple, and Amazon, have all completed the transition to professional management.

Nvidia's success relies on massive bets, which may take years or even decades to yield returns. Founders who have the full support of the board can take these risks without worrying about internal rebellion. CUDA was unprofitable for many years, and Wall Street analysts continuously questioned it, but Huang stuck with it, and that bet ultimately paid off massively. One former employee said, "Despite many opposing the investment in the limited-use programmable shader, CUDA, and Tensor Core, Huang still bet the whole company on these directions."

Omniverse and autonomous driving are Nvidia's new bets internally. Former employees say Huang referred to them as "zero billion dollar bets." They do not make money today, but he believes their potential is as immense as CUDA. He also created the Shield TV consumer entertainment box and the DGX Cloud service based on the belief that "Nvidia needs to understand customers to design better products"; even though both businesses generated nearly no profit and were eventually abandoned.

A successor to Nvidia who is not a founder, regardless of how long they have worked in the company, will have a different manner of being accountable to the board and shareholders. Such long-term bets and steadfast beliefs will become less likely. One of the biggest problems is that Huang's legendary ability to "turn stone into gold" and transactional acumen will be hard to replicate.

Customers Buy Huang, Not Just GPUs

Huang is the key decision-maker in allocating chips to hyperscale cloud vendors and other customers. Customers are always vying for quotas and trying to grab Nvidia's attention, and an increasing number of Nvidia deals require substantial support from them.

As we wrote in July, Nvidia can choose to allocate more capacity to hyperscale cloud vendors, neocloud, or its safety net plan's neocloud. In this plan, Nvidia provides credit support for clusters and shares cloud service revenue. This "Nvidia Central Bank" is supported by Huang's complete credibility and credit. The next CEO is unlikely to make bets as bold as Huang did.

Nvidia prefers to make decisions based on feasibility, but many things ultimately depend on what holds strategic significance. Skipping over hyperscale cloud vendors may not be strategically sensible; partnering with them and giving them preferential treatment may not always have strategic significance either. Nowadays, many customers want assurance: that Nvidia will continue to support them financially or with quotas.

Nvidia disclosed that as of the second quarter of its fiscal year, its total off-balance-sheet guarantee amount was $530 billion, up from $184 billion the previous quarter; this includes a $36 billion take-or-pay minimum commitment for neocloud capacity. Nvidia also holds a $99 billion equity investment portfolio.

Many customers want to strike deals with Huang for reasons that go beyond the numbers themselves. Neoclouds around the world have been able to secure financing because Huang endorsed them, not because they chose Nvidia after comparing chips. Nvidia's safety net measures allow neocloud and new laboratories to raise funds without the support of hyperscale cloud vendors: Nvidia sold GPUs at full margin from day one, then set a revenue floor for buyers, provided guarantees for buyers' landlords, or signed leases themselves, enabling operators without investment-grade ratings to borrow at investment-grade prices. Without the "Huang’s Put Option," this ecosystem could shrink significantly.

Note from BlockBeats: Nvidia's "safety net plan" is essentially a financial support system leveraging its own credit to facilitate large-scale financing of AI computing power infrastructure. Its core mechanism is known as the "Residual Value Support Agreement": in the typical transaction structure where customers leverage special purpose vehicles to finance the purchase of GPUs and rent out computational power, if rent payments are interrupted or a gap exists after asset disposition, Nvidia promises to cover the shortfall as per the agreed proportion, thereby alleviating lenders' concerns about defaults by AI companies and chip depreciation risks.

Control Over the Supply Chain Will Loosen

Nvidia has strong control over its suppliers. If competitors could obtain supplies, their shipments would be much higher. They are unable to do so because Huang personally locked down storage supplies in Korea and chip supplies in Taiwan.

As we wrote in "The Great AI Silicon Supply Shortage," Nvidia has locked in much of the supply of logic wafers, memory, and other necessary components. Huang's trip to Korea in 2025 was to secure supply before other executives fully realized the extent of the memory shortage. Our industry clients previously thought we were too aggressive in our 2026 forecasts, but later they found we were right. Nvidia's supply and capacity commitments skyrocketed from $119 billion in one quarter to $279 billion, primarily allocated for memory.

This control has a strong personal color. Huang negotiated these deals with major suppliers globally in a CEO-to-CEO manner. We estimate that in Korea, the price SK Hynix will charge Nvidia for HBM4 in 2027 could be about $3.0 to $3.3 per Gb; whereas other tier-one customers including AMD, Broadcom, Amazon, and Google might pay around $3.7 to $4.1 per Gb. We believe there is an understanding between both parties: in exchange, Nvidia grants SK Group priority quotas for GPUs. Without Huang, these companies would no longer prioritize Nvidia to the same degree, making it easier for others to access capacity. We believe that by 2028, over 70% of total capacity may not belong to Nvidia.

More importantly, in addition to locking in supply, Huang's vision has also driven Taiwanese supply chains to increase investment, leading them to believe in this round of construction more than anyone else. Huang’s legendary supply chain dinners and commitments have propelled Taiwanese suppliers to execute aggressive plans at technical and logistical levels.

The AI Industry Loses Its Open Source Advocate

Without Huang, the AI industry has lost its most important advocate. He was everywhere, always speaking as the chief advocate of AI. He was the face of AI, and no one can replace that.

Huang also formulated Nvidia's China policy and pushed for its support of open-source AI models. These models provided nations, enterprises, and neocloud with reasons to purchase GPUs, not just rely on a few pioneering laboratories developing their own chips.

As we wrote in September, no one needs open-source AI more than Nvidia, which is heading towards a world with "only Anthropic and OpenAI as real customers, and if you count hyperscale cloud vendors, a total of seven customers." Among these seven customers, except for SpaceX, all are trying to build their own XPUs to erode Nvidia's profit margins. AI models and applications are complementary to GPUs.

Chinese leading laboratories use Blackwell GPUs, and Nvidia's DevTech closely collaborates with these Chinese leading laboratories, learning from its end-users. This has contributed to innovations in the CUDA ecosystem, such as extensive expert parallelization, DeepGEMM, and DeepEP. Subsequently, these innovations spread to the Western open-source community and startups. Huang's departure will have a large-scale negative impact on these startups.

Without Huang as an advocate, and without financial and engineering assistance for open models, Chinese laboratories may not be able to open-source kernel libraries aimed at Nvidia's Blackwell GPUs, such as MiniMax's MSA, DeepGEMM, DeepEP, and FlashKDA from the Dark Side of the Moon. Globally widely used inference engines like SGLang and vLLM have adopted these innovations from Chinese laboratories. Chinese laboratories may have to completely deny that they own Nvidia GPUs and limit Nvidia software to internal use, rather than open-source Blackwell software.

Innovation from China will no longer be shared within the Nvidia ecosystem and the entire machine learning ecosystem, which, fundamentally speaking, is a bad thing. This may be beneficial for closed AI laboratories, but detrimental for startups.

A manager-led company might deprioritize the underperforming Nemotron. As many models from US leading laboratories shift to closed-source, Nvidia has been solely supporting other US open-source models through investments in companies like Reflection, Thinking Machines, Poolside, and the acquisition of Hugging Face.

Nvidia has consistently pushed to maintain continued contact with China, arguing that China should use and rely on US technology. As we wrote last year, the US government banned H20 from entering China, and afterwards stated it would grant export licenses to Nvidia, necessitating a balance between allowing China to use the US AI tech stack, slowing its autonomous development, and limiting product quality and quantity exported to China. Today, what if he is operating under constraints? That remains to be determined by investigations. Clearly, hawks are winning, which leads us to ask a somewhat strange question: why now?

Why Are Huang's Personal Issues Being Exposed Now?

We are not keen on conspiracy theories, but a few things occurring at the same time seem to form a narrative that's hard to ignore. The intelligence community benefits immensely from this. The New York Times published a second report on the topic, clearly stating that its reporting was based on US intelligence. Trump, Rubio, and Luteinick were aware of all this before visiting China in May. Additionally, there are reports that Luteinick had delayed actions against Chinese chips. This relationship has changed, and the winds are blowing in the other direction now.

We do not actually think Alice Huang has any special status—she is merely a well-connected business figure. It is normal for someone with widespread connections in China to have links with the authorities.

Our accelerator model has tracked over 1.5GW of Nvidia GPUs that should not exist in China. The US Department of Commerce now also believes that there are over 1.5GW of Nvidia chips that should not be in China. With the midterm elections shifting towards the Democrats, someone decided that strong action was needed: not only to stop the active circumvention of existing export controls but also to widen the restrictions to completely cut off China's access to cutting-edge chips.

It should be clarified that there were divisions within the Trump administration on this issue, but Trump had previously provided cover for Luteinick. Trump is no longer providing that cover now. More importantly, a critical piece of legislation known as the RASA (Remote Access Security Act) has passed in the House and awaits a vote in the Senate, and that vote is happening

RASA is About to Vote

On January 12, RASA passed in the House with a vote of 369 in favor and 22 against. It was passed through a "suspension of rules motion," which requires a two-thirds majority. The House Foreign Affairs Committee approved the bill with a vote of 51 in favor and 0 against in April 2025. Although the bill has been submitted to the Senate Banking Committee, it has not yet garnered a vote. We know that Nvidia and Oracle have been heavily lobbying against it.

Of course, RASA is not the only way the US government can prevent Chinese companies from remotely accessing GPUs. But RASA is the most hotly debated legislation in Washington right now. It is clearly popular because it has already passed the House; moreover, the substance of the legislation seems evident to many: why allow Chinese companies remote access to GPUs that they are not permitted to import?

It is worth mentioning that we know of neoclouds in Asia, Australia, the Middle East, Northern Europe, Europe, Canada, and even the US actively selling computing power to Chinese companies. They will all be impacted by this legislation. This is another global export control bill with far-reaching implications. Huang, and possibly Luteinick, do not want to see this happen.

Therefore, let us oppose the popular narrative and explore a possibility: Despite the various headlines in recent times, passing RASA or similar legislation could lead to unexpected consequences. According to an old saying in American politics, we believe that if all politicians agree on something, it’s probably a bad idea.

As previously mentioned, under the current regime, Chinese laboratories are able to remotely use GPUs, which brings the following clear benefits:

· Under the current regime where Chinese laboratories can remotely access GPUs, profits flow to US companies;

· The US can monitor what Chinese laboratories are doing;

· The US can, if willing, almost immediately cut off GPU usage rights;

· US tech companies rely on Chinese enterprises and engineers;

· US companies benefit from Chinese open-source research;

· China will also use domestically manufactured accelerators less;

· Innovation from China flows globally through open-source kernels and extensive expert parallel mechanisms like MSA, DeepEP, DeepGEMM, etc.

It must be made clear that in a world where a Chinese laboratory is completely cut off from using Nvidia GPUs, their motivation to turn to domestically produced accelerators and break away from Nvidia GPUs will only become stronger.

And this is already happening.

In February, Zhiyu AI claimed that GLM-5 now supports inference on seven domestic accelerators.

In April, DeepSeek released a V4 paper and subsequently open-sourced an inference superkernel for the V4 Pro model on Huawei's Ascend NPU.

In June, Meituan released LongCat 2.0, claiming that the model was fully pre-trained on Huawei's Ascend 910B NPU.

Politicians continue to ignore how tightly interwoven China, Taiwan, and the US actually are. Many Chinese nationals work in leading laboratories; many Chinese engineers are employed by Nvidia and AMD, writing the underlying code for launching new chips; many Americans learn from Chinese open-source contributions; many of the raw materials needed to produce these chips come from China; many ODM board and component suppliers from L1 to L6 are also Chinese companies.

Rising tensions will inevitably reduce chip output and slow AI advancement for everyone.

We believe the US should limit or control the scale of computing power delivered to Chinese laboratories; prohibiting exports to China while allowing trusted cloud service providers to offer remote access is the best way to accomplish this goal.

The Situation May Be Worse Than We Imagine

Therefore, if we must answer "why now," one possibility is: this move aims to clear the opposition to an extremely hardline export control bill on China. As significant changes occur in Congress, ahead of entering a lame-duck session, it is most beneficial to clear out the adversaries to push the bill through. This is our most conspiratorial judgment of the current situation, and it aligns with various aftershocks circulating in Washington.

So, back to the initial question.

What Will Nvidia Be Like Without Huang?

At such a scale, the only true case of a company continuing to operate well after losing its legendary founder is Apple and Steve Jobs. We believe that Huang's departure clearly signifies the end of an era of visionary founders; without the person who built this company at the helm, it cannot operate with the same intensity.

Using a car as a simple analogy: No one wants to ruin a car built by someone else; meanwhile, the car's manufacturer——Huang can drive comfortably at 150 miles per hour on the highway.

Whoever is promoted to maintain the status quo and facilitate the shift from founder-led to manager-led will likely have to deal with a coveted problem: how to handle such a massive cash flow. Huang clearly envisioned making the company a balance sheet backer for the ecosystem, providing and expanding credit for neocloud and laboratories. This is a natural extension of supply chain dominance.

But professional managers are very likely to do another thing that is indeed somewhat effective: stock buybacks. Apple post-Jobs is the best example of a company performing well after a founder's handoff. We believe the only script that has previously worked at such a scale could quite possibly play out again: a founder hands off to professional managers, the company becomes more conservative, and cash is returned. Shareholders will be pleased, but the company's pace of technological advancement will no longer be as it was before.

This is a bad outcome for technology. It will also be a bad outcome for shareholders. But clearly, people will miss Huang; the unique Nvidia will become more ordinary. As the supply chain shifts from following Huang's bold commitments to following the mundane vision of managers, internal struggles and the digestion process within this giant will manifest both within the company and globally. Without him, the tech world will be different.

Our current judgment is that Ian Buck is most likely in the lead, followed closely by Kress, who could become the formal CEO. External CEO candidates, such as Johny Srouji, are also rumored. But given Nvidia's uniqueness, we believe it is best to promote from within the company. The pool of candidates is small, and whether anyone is willing to take on this job remains to be seen. Only industry giants would stand a chance of being as exceptional as Huang; we doubt that giants like Lisa (AMD's CEO Lisa Su) or Hock (recently departed former Apple CEO) would make the switch.

In any case, this is highly beneficial for all Nvidia's competitors and very detrimental for Nvidia’s supply chain.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到[email protected]，本平台相关工作人员将会进行核查。