Source: NvidiaWelcome to the WeChat subscription number of “Sina Technology”: techsina
Author / Hao Junhui Editor / Wang Xin Kicked Girl
Source: IT Times
Nvidia’s “Spring Festival Gala” GTC on September 20 was unable to turn the tide. On September 26, Eastern Time, Nvidia closed at $122.28. A month ago, on August 26, Nvidia started this round of downtrend. It has dropped from $178.57 at the opening that day, and has fallen by more than 30% so far.
The capital market’s “abandonment” of Nvidia should be related to the recent merger of Ethereum. When mining is no longer a method of obtaining ether, the graphics card used for mining is naturally no longer a sought-after sweet and pastry. But “the leader” Huang Renxun obviously still has Plan B. The most surprising thing about this GTC is that Nvidia directly skipped the Atlan chip, which was just released last year and was not yet on the market, and announced that it will launch a new generation of autonomous driving chips in 2025-“Thor (Raytheon) ” Such a radical approach means that Nvidia pinned its hopes on filling the “deficient” of the mining card on autonomous driving.
At GTC, Huang Renxun revealed that the Thor chip will be the first to be mass-produced and delivered on the Krypton model. Extremely krypton, what kind of car? Geely’s high-end new energy vehicle brand, and in March this year, when Nvidia released Drive Hyperion 9, an autonomous driving technology platform based on Atlan chips, there was a long list of cooperation, including BYD, WeRide Zhixing, Yuanrong Qixing, Yunji China’s new energy vehicle start-ups such as Zhixing and Yopao Technology.
However, unlike the dominant player in the GPU field, the automotive business accounts for only 2.1% of Nvidia’s total revenue (FY 2022), while its competitors include American rivals such as Qualcomm and Mobileye, as well as China such as Horizon. To “replace”, let the car chip become the “hero” to save NVIDIA, especially in China, the powerhouse of new energy vehicles, the wishful thinking of Huang Jiao may not be successful.
It is undeniable that from the data point of view, Thor’s power is “suffocating”. At present, the computing power of mainstream in-vehicle AI chips is still hovering at the 100-level. The first generation of Journey 5 has a single computing power of 128TOPS. Mobileye expects the mass production of EyeQ Ultra by the end of 2023 to be only 176TOPS. How much is Thor? A single computing power of 2000TOPS.
According to the previous calculations of Horizon, the L5 computing power requirement for fully unmanned driving is exactly 2000TOPS. Lao Huang’s purpose is very clear. Tell car companies, don’t worry, I will give you one step.
But will Chinese car companies inevitably choose Nvidia? not necessarily.
The computing power of the chip has been rolled up. A few days after GTC opened, Qualcomm released a Flex SoC chip, which also covers computer vision systems, digital cockpits, ADAS/AD, etc. With the support of AI accelerators, the computing power can reach 2000TOPS. Horizon also said that it will launch Journey 6, a thousand-level TOPS, next year.
Judging from the announced partners, Qualcomm and Horizon have a good foundation with domestic car companies. As of the end of 2021, the cumulative shipment of Horizon Journey chips has exceeded 1 million, and it has cooperative relations with more than 20 car companies such as SAIC, GAC, and BYD. Not to mention Qualcomm, which is almost invincible in the field of smart cockpits.
Nvidia has a much smaller market space. Half of the customer list it lists are Chinese car companies, and there are a few scattered in Europe and the United States. No way, who makes it expensive. According to media reports, NIO ET7 is equipped with 4 Orin chips, and the cost of this chip alone may exceed US$1,500. If it is the “big unified” Thor… How much does it cost, guess what?
This means that car companies that can accept Nvidia are niche, at least at this stage. The price of Weilai ET7 is between 450,000 and 550,000 yuan, and the starting price of the ideal L9 (equipped with two Orins) is also above 450,000 yuan. On the new energy vehicle list of the China Passenger Car Association in August, Weilai ET7 sold 3,126 units, and the Ideal L9 did not enter the top ten.
Nvidia’s slogan is “the more you use, the more you save.” It sounds right. A centralized super computing power chip can beat all directions, and Thor can replace six single-function chips. The total price seems to be cost-effective. However, in the semiconductor industry, which attaches great importance to the scale effect, in the short term, Thor should still be a “luxury” that cannot be touched by ordinary domestic models.
Another thunder that may “explode” at any time is the control of high-end chips in the United States. In August this year, Nvidia’s high-end computing chips A100 and H100 were required by the US government to be “exported with a license”. These two chips are used for artificial intelligence algorithm training. Car companies and artificial intelligence companies, especially those new car-making forces that have invested heavily in autonomous driving, will directly reduce computing power due to chip restrictions.
China is a big car-consuming country, with sales accounting for one-third of the world’s total, but there is almost no possibility of catching up in traditional automotive fields such as engines. Electrified new energy vehicles, including the accompanying self-driving cars, are considered to be China’s The perfect time to overtake on a “next-generation wheel” corner.
This is obviously not what some people want to see. Nvidia’s “toughness” is that it can create top-level products for both the automotive and the cloud, and its “fragility” may be precisely because of this – it becomes the product that is most likely to be “targeted”.
Just imagine, if the U.S. government cares so much about the A100 and H100, which can provide super computing power for China’s autonomous driving AI training, will it miss Thor, which provides powerful brain power for every car? Especially when China’s new energy vehicles go overseas to “show muscles”.
Furthermore, car companies have suffered enough from the “lack of cores”. Even if they now pay for chip companies, the “heart” of self-developed research has never stopped beating. BYD has cooperated with Nvidia and Horizon, but its own BYD Semiconductor Company has been established for 17 years and has launched its own IGBT4.0 automotive-grade chip. Chery has strategically invested in Horizon to ensure stable chip input.
Eating a piece of money grows a wisdom. Since entering the new energy era, traditional car companies have been “stuck” by battery and chip manufacturers. Whether NVIDIA, Qualcomm or any chip company, which one will become the next “Ningde era”, it is estimated that Neither is a happy outcome.
Although the future is still in the fog, for Nvidia, automotive chips are crucial. Taking Qualcomm as an example, it revealed at the investor meeting that the QCT (semiconductor chip) automotive business revenue has grown strongly, from $975 million in fiscal 2021 to $1.3 billion in fiscal 2022 (expected revenue). The total business order valuation has now grown to $30 billion. The addressable market for the company’s automotive business is expected to grow to $100 billion by 2030.
Others are already “eating meat” with big mouths.
On September 29, Nvidia closed at $122.2, down 4.05%. When will the new cycle start? When did the “darkest hour” pass? Everything is still unknown.
(Disclaimer: This article only represents the author’s point of view and does not represent the position of Sina.com.)
This article is reproduced from: http://finance.sina.com.cn/tech/csj/2022-09-30/doc-imqqsmrp1079406.shtml
This site is for inclusion only, and the copyright belongs to the original author.