AI News

Why isn't the auto industry panicking over Nvidia's leadership change?

36Kr · AIUpdated Nov 14 12:58

AI summary

After Jensen Huang resigned as Nvidia CEO, the stock fell more than 13% at the open and shed nearly $500 billion in intraday trading, but Chinese automakers made no emergency adjustments, with design wins and OTA updates proceeding as usual. The article states that from January to September this year, Nvidia's three intelligent driving domain controller chips—Orin-N, Thor-U, and Orin-X—together exceeded 2.2 million units installed in China, with a market share of nearly 40%, while Horizon Robotics, Huawei Ascend, and automakers' in-house chips are already taking over the substitution. The author believes that existing orders will not be cut off in the short term; the real variables are the market access and iteration space for the next-generation DRIVE, and domestic computing power shifting from a backup option to a risk-control necessity.

Why read this

AI score 72

The article reviews Nvidia's market share and the pace of substitution in China's intelligent driving and computing power markets, and explains why automakers have remained calm amid the leadership change turmoil.

Topics

Tags

Body · AI translation

"Chief Qiao ran the Beggar Clan for decades, working tirelessly, and actually turned the Beggar Clan into the number one clan in the jianghu. But then Quan Guanqing fabricated a scandal about him having an affair with Lady Ma and slapped a 'Tongliao' label on him, and just like that Chief Qiao's contributions were wiped out, and he was even expelled from the clan! Where is justice? Where are Chief Qiao's supporters? Did he have no loyal inner circle?"

Because China's history and cultural heritage are so long, it has in a very real sense demonstrated that there is nothing new under the sun. And so, regarding the recent series of events surrounding the "Jensen Huang affair," Chinese netizens once again had a flash of inspiration. Even better, one commenter in the comments section chimed in with a quip—"No way around it, who told him not to be a Song dynasty man."

All these changes came too suddenly, because originally this man appeared in American political circles as a loyal and indispensable minister.

Just at the start of the year, Jensen Huang was still at CES talking up "strong demand from Chinese customers for the H200," and promising that if the US Commerce Department granted licenses for China sales, 25% of sales revenue would go to the national treasury. When Trump visited China in May, he specifically notified him to rush over and accompany the trip midway. Even just over a month ago, he was at Nvidia's GTC conference in Berlin talking up how "AI factories compress on-site deployment from six months to fourteen days." As for the $193.7 billion in data center revenue for fiscal 2026 and pushing Nvidia's market cap above four trillion dollars—those were achievements laid out on the table for all to see.

From the helmsman of a top tech giant driving the AI revolution, a patriot doing his utmost to reduce the national deficit—falling from the peak took just a little over a week. An apology letter, an independent investigation, outside lawyers, new revelations involving direct subordinates, reports that the board was preparing to dismiss him, and lawmakers' calls—all landed one after another. And with Jensen Huang's departure, Nvidia's stock fell more than 13% at the open, evaporating nearly $500 billion in intraday trading.

Yet what's interesting is that compared with the raging public opinion in the US and the Chinese public's gleeful gossiping, the auto industry was so quiet it felt a bit out of character. Although a few executives used it as dinner-table conversation, in reality no OEM held overnight meetings, let alone emergency discussions and adjustments by project teams. Designations proceeded as usual, OTA updates were pushed as usual.

Don't assume this calm is due to sluggishness, because Nvidia's penetration into China's auto industry runs to the bone.

Based on information synthesized from multiple public databases, between January and September this year, among smart-driving domain control chips installed in domestic passenger cars, Nvidia's three mainstream products totaled roughly over 2.2 million units, with market share heading straight toward 40%. At Nvidia GTC 2026, GM, Toyota, Mercedes-Benz, and Hyundai all sent representatives, and BYD and Geely were also on the list.

Deeper than the chips is the compute cards and various supporting development tools Nvidia provides—DriveOS, CUDA, Omniverse, and DRIVE Sim all live in the same environment, with cloud DGX training, OVX simulation, and vehicle-side AGX inference. Perception debugging, synthetic data, and virtual validation of in-car design all run on it. Tesla's Cortex 1 is equivalent to over 100,000 H100s, XPeng's cluster exceeds ten thousand cards, and the majority of the bill for each iteration of smart-driving large models is recorded here. And domestic giants like BYD and Geely have also joined Nvidia's autonomous driving development platform NVIDIA DRIVE AGX Hyperion.

For such a company to suddenly change leaders, and in such an bizarre way, at such rapid speed—how could the auto industry not be affected? Why are domestic automakers and even related supply chain companies so calm?

01 A key "foundation," but not the only one

When Nvidia launched the DRIVE platform in 2015, what OEMs bought was actually a compute board. But after ten years of development, what's being handed over is a whole package.

On the vehicle side, from Orin's 254 TOPS to Thor's 2000 TFLOPS, with DriveOS, DriveWorks, and the Halos safety system in between. In January 2026, Nvidia open-sourced the Alpamayo chain-of-thought model, the AlpaSim simulation tool, and the physical AI dataset together; NuRec can restore road-test videos into editable 3D scenes—change trajectories, swap weather, and generate dozens of training samples.

Even the underlying models are completely free and open, with Nvidia placing its expected returns on compute demand. And Chinese automakers have benefited from this dividend more than anyone.

Li Auto AD and XPeng XNGP—these urban navigation systems—were all built on Nvidia's platform. BYD's God's Eye B uses a single Orin-X on the Han L and Tang L, while God's Eye C uses Horizon Journey series. Xiaomi SU7 and YU7 use Orin and AGX Thor.

According to Gasgoo statistics, among domestic smart-driving domain control chips from January to September this year, Orin-N was about 929,000 units at 15.6%, Thor-U about 777,000 units at 13.0%, and Orin-X about 528,000 units at 8.8%—the three combined firmly hold the top spot in smart-driving chip installations. By comparison, Huawei Ascend 610 installations were roughly 860,000 units, at 14.4%.

Of course, based on experience since 2016, anything that cannot be controlled by ourselves, China will seek to replace. And our replacement efforts didn't start on November 12. The quiet-as-a-chicken performance not only reflects silence due to the sensitivity of the matter, but on another level shows that because preparations exist, there's no need for an overreaction of "the sky is falling."

Image | By the end of 2026, computing power self-reliance and in-house chip development had become an absolute consensus in China; the image shows the XHEART X7 Ultra launch event.

In the first half of 2026, in the full-stack smart-driving chip market for self-owned brand passenger cars, Horizon ranked first with 31.94%, and among urban NOA high-level suppliers, Horizon ranked second with 22.82%.

According to Horizon, the Journey 6 series has already secured design wins with more than 25 automakers and over 100 models in China. The author found data from NE Era New Energy stating that in September, Horizon shipped 183,583 chips in a single month, and market expansion is accelerating.

In addition, the in-house chips of XPeng, NIO, Li Auto, and BYD will also be gradually installed in vehicles in the future. Huawei Qiankun ADS, relying on its self-developed MDC platform and Ascend automotive-grade chips, has already achieved a full-stack closed loop, with its supply chain completely separated from Nvidia. Xiaomi's Xuanjie D100 is scheduled for 2027.

According to CAAM figures, the domestic localization rate of automotive-grade chips rose from 4.7% in 2022 to 11.9% in the first quarter of 2026. The current shortcoming is that for high-end computing SoCs, the share that China can independently supply is still less than 10%. We must acknowledge the gap in high-end products, but after all, the overall share is continuously narrowing.

02 No one can stop our pace

Nvidia's announcement explicitly stated that operations, customer commitments, and the product roadmap would not be interrupted. And as of today, no automaker has disclosed that supply has been affected.

So we can broadly conclude that this turmoil will not cause a cliff-edge supply cutoff for short-term existing orders. The key is that it will inevitably affect access permissions and technological iteration for the next-generation DRIVE—that is, the next round of design wins.

It takes two to three years for a chip to go from design win to vehicle installation, and already-awarded projects also have long-term framework contracts. Replacement costs are so high that no automaker would disrupt its plans. Likewise for Nvidia, for vehicle-side chips under signed contracts—contracts that neither violate existing U.S. Commerce Department rules and whose breach could also expose the company to huge compensation—continued execution is only natural.

In particular, given that a large number of Chinese companies are already trying to step in and replace it in this field, in order to optimize financial data as much as possible and prop up the stock price amid the current situation, it certainly will not resort to any sudden "supply cutoff."

But the key issue still has to land on the focal point of computing power cards. Without additional computing power, how can automakers and smart-driving suppliers complete various simulation, algorithm training, and other work? In fact, relatively positive changes have also been emerging in this area over the past year.

According to IDC data, in the China data center AI chip market in the first half of 2026 by shipment volume, Nvidia shipped about 678,000 units, with a market share of 34%, down about 44% from the first half of 2025. By comparison, Huawei's Ascend series reached 572,000 units, for a 28% market share. The gap between the two has narrowed from 31 percentage points a year ago to 6 points.

Ascend relies on the CANN ecosystem to support model migration, while Hygon and Cambricon are being deployed in different scenarios. Migrating cloud training to domestic clusters is no longer just a contingency plan.

Of course, we cannot only say the pleasant things here; after all, this "holding up" comes at a cost. At present, domestic computing power is still constrained by production capacity, interconnect bandwidth, and ecosystem maturity. CUDA's moat cannot be breached in the short term, and the toolchain must be honed cycle by cycle. At the same time, constrained by semiconductor processing equipment, our substitute products still have disadvantages in transistor density and energy consumption. In fact, this problem has always existed, so in the past two years China has generally shifted to running external procurement, in-house development, and domestic substitution in parallel.

Some media figures have made it very clear in the past two days that over the past year, the stable supply of domestic computing power cards has to a large extent relied on Jensen Huang's previous personal balancing within the boundaries of U.S. Commerce Department rules. We did indeed rely on his personal efforts to preserve many existing orders and signed contracts. However, his successor, Cress, when facing the current situation, will inevitably prioritize compliance and risk control, and will most likely not continue this flexible mediation. Even the remaining existing supply margin, fine-tuned quotas, and room for version iteration may all be tightened.

Under such circumstances, domestic computing power will change from a backup option for supplementary pilots and wait-and-see standby into a rigid risk-control necessity that must be pursued.

For automakers, the current optimal solution is to accelerate in-house development and domestic substitution plans while implementing overseas solutions that can still be executed.

03 The elegy of North American Chinese elites

After the Nvidia board sent the man away, a joint letter from hundreds of employees quickly circulated online, with fierce wording demanding that the board resign en masse and that Jensen Huang return, among other things. However, this letter does not withstand scrutiny. Some media compared its wording with the 2023 OpenAI employee pressure letter and found it almost copied verbatim. What exposed it even more was that the letter said the board fired him, while the announcement and the 8-K filed with the SEC both used "resignation."

The real thing is actually not that dramatic, but it carries more weight. Wired magazine found a post by a senior engineer in the internal suggestion box, under which more than 1,600 comments had accumulated, most of them asking for the boss to be brought back. Some sent company-wide emails demanding a clear explanation of the decision-making process, and someone placed flowers in front of the logo at headquarters. On Blind, an anonymous workplace community, 232 registered employees identified as Nvidia staff, nearly 90% of whom believed Jensen Huang should continue in the role. Quite a few even said that the company where they had worked for more than a decade had this time had its "central axis" pulled out by a single announcement.

Using the word "central axis" to describe Jensen Huang and Nvidia is, in fact, not an exaggeration. The most counterintuitive thing about this company is that there is no second-in-command. Thirty-six executives report directly to Jensen Huang; there is no COO, no one-on-ones, and no formal succession plan. Huang himself has explained the reason: GPUs, memory, networking, power, cooling, and software must be optimized together, and adding a filtering layer in between would make co-design impossible. Information is spoken to everyone at the same time, and the rhythm of one product generation per year is bound to this approach.

And that is exactly where the problem lies: these thirty-six people all reporting to one person does not mean that one of them can replace that person. Apart from Huang himself, no one has been given the authority and responsibility for cross-business-line integration, and no one has been institutionally groomed as a backup. Although this leather-jacket-loving CEO holds only a small stake and does not even have super-voting rights, he is truly Nvidia's irreplaceable core, the "central axis" of all operations. What has enabled him to hold this company for thirty-three years is not the charter but that set of daily operations with him as the bus. The flatter it is, the cleaner the cut.

His successor Cress has been CFO for thirteen years, which is enough to stabilize the transition period, but managing the books is one thing and stepping on the gas is another. Nvidia has a playbook that runs on personal credit: invest in customers first, and the customers then take the money and come back to buy cards. Swap in a manager who puts compliance first, and this kind of investment will certainly be curtailed. The new management will most likely also set up a COO and fold dozens of reporting lines into a conventional hierarchy. But is this problem really something that can be solved by the board issuing a resolution and holding a meeting?

A man who built a company to four trillion and ran around everywhere securing permits for his own chips has been shown to be able to hand over all his duties within a week, before the investigation conclusions are out. A scholar at Fudan put it bluntly: this incident reminds non-white elites that the American dream has a visible ceiling. Enough said.

Qiao Feng was expelled from the Beggars' Gang, and the Beggars' Gang is still there, only from then on no one could settle matters at a glance when trouble arose. Nvidia will indeed not stop operating without Jensen Huang; after all, the product line is already booked out several years ahead. But what about several years from now?

And here, we are actually very curious about one question too—now that Jensen Huang has fallen, which "little yellow man" should be next, or is it "Su Ma"?

On November 13, 2026, Beijing time, Reuters disclosed that Jensen Huang had an intimate relationship with another Nvidia employee, the third such relationship exposed in the past week. The turmoil is not over yet, and it has already turned into something in the style of "Aqina and Saisihei."

After Jensen Huang resigned as chief executive, many people are asking to what extent the company will still stick to its automotive business going forward. Actually, the answer is not in San Jose; it is in Washington.

After the midterm elections, the tearing between the Republican and Democratic parties in American political and business circles has intensified, and what all sides are fighting over is that money-related channel and who exactly should control it.

Jensen Huang is caught in the middle, and it is indeed awkward. He is a competent CEO who did his utmost to protect market value and profit, and lobbying to relax exports to China would not be considered out of line for the head of any multinational company. Yet in the current American political system, this makes him a special symbol who is both under scrutiny for seeking to expand business in China and carries a Chinese face, and one far easier to go after than a white face. In such a sensitive period, if not him, then who else would be sacrificed?

In May 2026, Huawei released the Tao Law, the core of which is to replace geometric scaling with time scaling, using logic folding to stack circuits in layers and connect them with vertical channels, pressing down signal propagation delay layer by layer and improving performance without upgrading the lithography node. This targets the current reality: Moore's Law is slowing, and more than 80% of the energy consumption of large AI clusters is spent on data movement.

Over the past six years, our companies have been following this path. Kirin 2026 was the first to use logic folding, and public measurements show transistor density up about 55% and energy efficiency up 41%, with the industry judging that by 2031 it can reach a level equivalent to a 1.4-nanometer process.

On the independent chip manufacturing side, SMIC's and Hua Hong's mature processes, JCET's and Tongfu's packaging, and NAURA's and AMEC's equipment are all making continuous efforts. And Huada Jiutian's EDA is also continuing to catch up.

The competitive core of China's automotive intelligence is shifting from involution over computing-power parameters to supply-chain resilience and self-reliance and controllability. This is a positive signal.

For automakers, what they should do is actually still the most plain things: deliver the already-designated projects well, get second suppliers running, and complete the automotive-grade validation of self-developed chips page by page.

Whose board changes, or even what happens to the company, actually has little to do with us. After all, for automakers, what intelligent driving ultimately delivers is not the computing-power number of the hardware installed on a single vehicle, and it is not even much related to how much total computing power is mastered domestically; its essence is whether the entire system can perform the functions it claims.

This article comes from the WeChat public account "Auto Commune" (ID: iAUTO2010), author: Zha Youyin, published by 36Kr with authorization.

Source

36Kr · AI