AI News

After Jensen Huang Leaves Nvidia, Who Is Shaking the Compute Throne?

36Kr · AIUpdated Nov 16 21:25

AI summary

After Jensen Huang's departure from Nvidia, the company is facing a contest over the CEO succession and governance adjustments. The board has approached Arm CEO Rene Haas, while U.S. Treasury Secretary Bessent advocates finding a new CEO from outside. Customers' in-house chips are accelerating in parallel: Google's seventh-generation Ironwood TPU has entered commercial use, Amazon's Trainium3 has entered the commercial stage, and Anthropic has already used more than 1 million Trainium2 chips to train and run the Claude model. In the China data center AI accelerator market in the first half of 2026, Nvidia's share fell to 34% and its shipments fell 44% year on year, while Huawei rose to second place with about a 28% share.

Body · AI translation

After Jensen Huang left Nvidia (NASDAQ: NVDA), that was not the end of the matter.

On November 15, the Financial Times reported that Jensen Huang was discussing with close associates the possibility of returning to Nvidia as CEO, and that his Silicon Valley allies were also forming a lobbying team in the hope of winning support from the board and the White House. On the other side, U.S. Treasury Secretary Bessent publicly stated that Nvidia should look for a new CEO from outside and carry out major adjustments to the company. Previously, the market had also circulated news that Nvidia's board had approached Arm CEO Rene Haas.

In just a few days, the contest over who would lead Nvidia had extended from internal corporate governance to U.S. political-business relations and the struggle over interests in the AI industry. On November 17, Nvidia will also announce its latest quarterly results, an important moment for the new management to stabilize market expectations.

For a company whose market value and industry influence both rank at the forefront of the global technology industry, Jensen Huang's departure is clearly not just a personal matter. Behind his departure are conflicts involving founder power, corporate governance and capital interests, as well as a contest between the two U.S. parties over technology industry policy. At the same time, Nvidia's customers had long since begun looking for alternatives, and China's AI industry is also accelerating the building of self-reliant computing power under external pressure.

Behind the departure: corporate governance and the two-party contest

On the surface, the direct cause of Jensen Huang's departure was the recent controversy surrounding his personal conduct and the resulting corporate governance review.

From the perspective of corporate governance, although Jensen Huang is the soul of Nvidia, that does not mean he holds irreplaceable control over the company. Unlike some Silicon Valley technology companies that use dual-class share structures, Jensen Huang holds about 3.5% of Nvidia's shares and does not have super-voting rights.

As Nvidia grew into a major global listed company, the board needs to bear responsibility to all shareholders. When a founder's personal controversies may affect corporate reputation, regulatory relations and long-term operations, the board has an incentive to prioritize the company's overall interests, and also has an impulse to use this opportunity to optimize Nvidia's governance structure so that the board can more effectively control company management.

But if the incident is observed in the American context, the matter takes on another layer of meaning.

On November 3, the U.S. midterm elections concluded, with the Democratic Party winning a majority in the House and Senate and set to formally take over both chambers of Congress in January 2027. This means the Trump administration will face more obvious congressional checks over the next two years. Jensen Huang's departure happened to occur in the window of power transition between the old and new Congress.

Against this backdrop, Nvidia and Jensen Huang became one of the focal points of the two-party contest.

On the Democratic side, Senate Democratic leader Schumer, House Democratic leader Jeffries and others demanded that Congress further examine Nvidia-related matters and pay attention to the Trump administration's previous chip policy decision-making process. For the Democrats, this is not only a matter of technology company governance, but also an important entry point for overseeing the Trump administration's industrial policy and reviewing its administrative decisions.

There are also hardline voices within the Republican Party. Representative Moolenaar, chairman of the House Select Committee on the Strategic Competition between the United States and the Chinese Communist Party, and others advocate strengthening chip export supervision and believe the United States needs to re-examine existing policies. Some Republican lawmakers also demanded that relevant corporate executives and Commerce Department officials face congressional questioning.

It is worth noting that although the two parties differ on how to hold the government accountable and how to regulate companies, they share a considerable common position on maintaining America's AI technological advantage and strengthening advanced chip management. Both want to use this opportunity to show a tough stance on technology company governance in order to win votes.

This also explains the awkward position Jensen Huang currently faces. As an entrepreneur, he needs to consider global markets, customers and commercial profits; but in the eyes of some politicians in Washington, AI chips are no longer just ordinary commodities, but also important bargaining chips in industrial competition, and his "yellow face" adds to the government's distrust.

Previously, Jensen Huang had long advocated retaining the Chinese market, hoping to maintain Nvidia's global competitiveness through commercial cooperation, which conflicted with the increasingly hardline industrial policies of some U.S. politicians.

Therefore, Jensen Huang's exit cannot simply be understood as political liquidation of one party by another, but rather the result of intertwined corporate governance risks, two-party political competition and conflicts of commercial interests. The claim that traditional Wall Street capital is joining forces to suppress Silicon Valley technology upstarts still lacks sufficient evidence.

What is truly worth watching is how a company highly dependent on its founder's personal influence can find a new balance under political and capital pressure.

Customers develop their own chips, and Nvidia's ecosystem moat faces challenges

Compared with the reasons for Jensen Huang's departure, the market has begun to focus on whether Nvidia's long-established industry dominance can continue in the post-Jensen Huang era.

In the past few years, technology giants such as Microsoft, Google, Amazon and Meta have invested hundreds of billions of dollars in building AI infrastructure, but a considerable portion of that money has flowed to Nvidia. For these companies, purchasing Nvidia GPUs can quickly provide advanced computing power, but it also means bearing higher chip procurement costs and being influenced by the supplier in terms of technology roadmap and supply pace.

Therefore, leading customers had long since begun trying to break away from sole dependence on Nvidia.

Google was one of the earlier movers. After years of iteration, its TPU has formed an independent AI computing system. The seventh-generation Ironwood TPU launched in 2026 has officially entered commercial use and can support large-scale model training, inference and complex computing tasks. Google will also open TPU access to cloud service customers to further expand its own software and hardware ecosystem.

Amazon, meanwhile, continues to advance its Trainium series of chips. Trainium3, launched at the end of 2025, has already entered the commercial stage; Amazon is combining its in-house chips with cloud services through AWS in a bid to lower AI training and inference costs.

Its important customer Anthropic is also expanding its cooperation with Amazon. As of April 2026, the cooperation plans announced by the two sides include up to 5GW of new computing capacity, and Anthropic has already used more than 1 million Trainium2 chips to train and run the Claude model.

Microsoft and Meta are also accelerating similar arrangements. Microsoft's Maia 200 entered the production environment in 2026, and the company plans to begin large-scale expansion of Maia-series computing deployments starting in 2027. Meta, for its part, has proposed advancing four generations of MTIA chips over the next two years, covering businesses such as recommendation systems and generative AI inference. Information it has disclosed shows that MTIA has already been deployed internally in the hundreds of thousands of chips.

These moves show that the goal of leading tech companies is not simply to find another GPU supplier, but to gradually build their own computing power systems.

Customers' de-Nvidia-ization: on the one hand, in-house chips can be optimized for their own businesses, reducing unnecessary computing consumption; on the other hand, multi-supplier procurement can increase bargaining power and avoid concentrating huge capital expenditure on a single chipmaker over the long term.

Of course, Nvidia's advantages will not disappear easily. The CUDA software ecosystem, developer habits, cluster interconnection and product iteration capabilities still constitute high competitive barriers. The in-house chips of companies such as Google and Amazon are currently also mainly aimed at specific business scenarios, which does not mean they can fully replace Nvidia GPUs.

But from the perspective of industry trends, customers' in-house chips and products from competitors such as AMD are forming new choices. Huang Renxun's departure may further weaken customers' psychological dependence on a single supplier, pushing the global AI computing market to gradually move from Nvidia's dominance toward a landscape in which GPUs, ASICs and other dedicated computing architectures coexist.

China's computing self-reliance: US restrictions may be counterproductive

Today, this worry is gradually becoming reality.

Reuters, citing IDC data, reported that in the first half of 2026, Nvidia's share of China's data center AI accelerator market had fallen to 34%, with shipments down 44% year on year; Huawei's shipments reached 572,000 chips, for a market share of about 28%. At the same time, Alibaba Pingtouge, Cambricon (688256.SH), Baidu Kunlunxin, Hygon Information (688041.SH) and other companies are also expanding their market presence.

More noteworthy is the change in the software ecosystem.

In the past, one of the biggest obstacles for domestic AI chips was not simply computing performance, but the developer ecosystem that Nvidia's CUDA had built up over many years. A large number of models, algorithms and applications are built around CUDA, and for enterprises to switch underlying chips, they need to carry out code migration, operator adaptation and system optimization anew, at no small cost.

Image source: Ascend Community

But as Huawei's Ascend and CANN ecosystem continue to improve, this situation has begun to change. Huawei disclosed in 2026 that external developers accounted for 61% of CANN developers, with more than 5,200 monthly active developers, and that more than 40 models had completed native pre-training on Ascend and CANN. Domestic AI companies are also promoting model adaptation, enabling domestic computing power to develop from simple hardware substitution toward software-hardware coordination.

In its fiscal 2026 second-quarter filing submitted to the US SEC, Nvidia has explicitly warned that the loss of the Chinese data center market is helping competitors expand their developer and customer ecosystems and may have a material impact on the company's future operations. This in fact confirms Huang Renxun's earlier concerns.

Huang Renxun's departure may become a new node in the changing landscape of the global AI computing industry. Nvidia still has deep technical accumulation, but customers' in-house development, supplier diversification and China's computing self-reliance are all changing the market structure that was once dominated by a single company.

Whether US industrial restrictions can achieve the expected effect remains to be seen, but if partisan struggle and conflicts of interest accelerate the growth of competitors, those ultimately affected may also include the long-term commercial interests of US tech companies themselves.

(This article is for reference only and does not constitute investment advice. Markets carry risks, and investment should be made with caution.)

This article comes from the WeChat public account "Investors Network - Siwei Finance", author: Wu Wei, published by 36Kr with authorization.

Source

36Kr · AI