NVIDIA's Dual Real Woes: Antitrust and Delayed Shipments
NVIDIA is currently facing two troubles: an antitrust investigation by the U.S. Department of Justice and a delay in the shipment of its flagship chip GB200.
However, its real challenge lies in the fact that large model companies that have purchased NVIDIA's chips need to make real money.
NVIDIA, the global AI chip giant (NASDAQ: NVDA), is experiencing a continuous correction in its stock price.
In the first half of this year, NVIDIA once became the world's most valuable company.
On June 18th, Eastern Time, NVIDIA's market value climbed to $3.34 trillion (see the June 19th issue of "Finance" for related reports, "NVIDIA's market value tops the list, will it continue to grow explosively?").
In the following two months, NVIDIA's stock price continued to correct.
The reason is that NVIDIA's stock price increased too quickly, and some investors chose to cash out at high prices.
Affected by factors such as the U.S. Department of Justice's antitrust investigation, the delay in the shipment of the flagship chip GB200, and the collective decline of the U.S. stock market, NVIDIA's stock price continued to slide.
At the closing on August 5th, Eastern Time, NVIDIA's closing price was $100.45, down 6.36% from the previous trading day.
Advertisement
NVIDIA's current total market value is $2.52 trillion.
On August 2nd, media reports stated that the U.S. Department of Justice has launched two investigations into NVIDIA.
First, in April of this year, NVIDIA spent $700 million to acquire Run:ai, an Israeli chip management software company, which may be suspected of monopoly.
Second, NVIDIA has been reported by competitors for potentially abusing its dominant position in the AI chip market.
For example, it is alleged that NVIDIA has been pressuring cloud service providers to purchase a variety of products.
In response, a NVIDIA spokesperson stated that NVIDIA wins through strength.
NVIDIA has been competing through decades of investment and innovation, strictly abiding by all laws.
They also stated that NVIDIA is willing to provide any information needed by regulatory authorities.
On August 3rd, news broke that NVIDIA's next-generation flagship AI chip, GB200, will be delayed by more than three months.
The expected delivery time will be postponed from the original plan of the third or fourth quarter of 2024 to the beginning of 2025.
One interpretation is that the GB200 chip design is relatively complex, and TSMC's CoWoS-L packaging process is also quite new.
As a result, the production yield is not as expected, leading to a delay in production progress.
On August 5th, NVIDIA responded again to this matter, stating that it does not comment on market rumors.
The demand for the Hopper series (H100/H200) is strong, and the trial of the Blackwell (B100/B200/GB200) samples has been widely carried out.
It is expected that production will increase in the second half of 2024.
JPMorgan's latest research report on this news pointed out that there are no major issues with GB200 that would lead to a redesign or multi-quarter delay.
The judgment of several industry insiders consulted by "Finance" is that although NVIDIA is facing issues such as the U.S. Department of Justice's antitrust investigation and the delay in the shipment of the flagship chip GB200, these issues have a limited impact on NVIDIA's fundamentals for the time being.
Despite the stock price fluctuations, NVIDIA's performance is very stable from an operational perspective.
NVIDIA has maintained a revenue growth rate of over 200% for three consecutive quarters (Q3 2023 - Q1 2024) and a net profit growth rate of over 600% for four consecutive quarters (Q2 2023 - Q1 2024).
NVIDIA firmly holds more than 90% of the global data center GPU (Graphics Processing Unit) market share and is difficult to replace in the AI chip industry chain.
A long-term antitrust researcher told us that antitrust is an effective tool for regulatory authorities to deter companies from abusing their monopoly positions.
The logic of antitrust is generally that it does not oppose companies from occupying a monopoly share in the market through innovation and competition, but it opposes companies from abusing their monopoly positions to suppress opponents or customers.
The boundary for regulatory authorities to determine whether to initiate antitrust litigation against a company is whether the company has hindered the innovation of other companies.
NVIDIA holds an absolute dominant position in the data center AI chip market.
Data center AI chips include GPUs and specialized AI chips (such as NPU neural network processors, DPU deep learning processors, etc.).
The former has strong general capabilities and is generally used for AI training and inference; the latter is used for specialized scenarios, usually focusing only on AI inference, with a few also available for AI training.
Wells Fargo data in February this year showed that NVIDIA's global data center GPU market share in 2023 was 98%.
Semiconductor market research firm TechInsights data in 2024 showed that the data center AI chip market size in 2023 was $17.7 billion, with NVIDIA's market share at 65%, Intel's at 22%, and AMD's at 11%.
It is particularly important to note that the two antitrust investigations recently launched by the U.S. Department of Justice against NVIDIA are not aimed at NVIDIA's monopoly market share, but rather to assess whether NVIDIA is abusing its dominant position in the AI chip market.
One of the investigations by the U.S. Department of Justice is NVIDIA's acquisition of the Israeli startup Run:ai.
According to media reports, the value of this transaction may be as high as $700 million.
Run:ai mainly operates GPU management software and is committed to helping users complete more computational tasks more efficiently under limited chip resources.
This technology is particularly important in the context of a global chip supply shortage.
U.S. regulatory authorities have doubts about NVIDIA's motivation for acquiring Run:ai.
The logic is that Run:ai's technology can help customers reduce their demand for AI chips from companies like NVIDIA, which may threaten NVIDIA's revenue and profits.
There is a suspicion that NVIDIA may be suppressing innovation through acquisition.
In recent years, U.S. regulatory authorities have become stricter in reviewing the acquisition cases of large technology companies, but most acquisition cases can still be approved after adding conditions.
Another investigation by the U.S. Department of Justice is, at the request of other competitors, assessing whether NVIDIA is using its dominant position in the market to force cloud service providers to purchase its products.
The reason is that some customers and competitors have accused NVIDIA of adopting improper sales strategies, such as bundling key software with chips for sale, or charging excessively high fees for customers who choose to purchase AMD or Intel AI chips.
In response to this accusation, NVIDIA stated that it is willing to provide the required information to regulatory authorities.
The company believes that the reason for its success is precisely "based on long-term investment and innovation, and always abiding by the law, ensuring that customers can freely choose solutions."
In 2024, NVIDIA's business and market value are rapidly expanding, and the risk of antitrust is gradually increasing.
Regulatory authorities in various countries have expressed concern about NVIDIA's dominant position in the AI chip market.
In February of this year, NVIDIA publicly acknowledged that officials from the United States, the European Union, and the United Kingdom are also reviewing its business.
Due to its position in the AI-related market, the attention of global regulatory authorities to its business has increased.
In July of this year, the French regulatory authority officially filed an antitrust lawsuit against NVIDIA, accusing it of manipulating prices, unfair contract terms, and discriminatory practices.
The French regulatory authority also expressed concern about the AI industry's excessive dependence on NVIDIA's CUDA software.
Some U.S. congressional politicians are also supporting antitrust investigations against NVIDIA.
In August of this year, U.S.
Senator Elizabeth Warren of Massachusetts warned that allowing a single company to become the gatekeeper of the global artificial intelligence future is dangerous and may pose serious economic risks.
In the short term, NVIDIA will not be overly affected by the global antitrust trend.
The dense antitrust actions of various governments may lead to slight fluctuations in the company's market value in the short term, but in reality, NVIDIA has enough time and space to deal with its impact.
Because antitrust investigations usually take several years, and the results are usually high-profile and low-impact, technology companies such as Microsoft, Google, and Qualcomm have all encountered antitrust investigations by regulatory authorities in different countries, but the outcomes are all that they pay hundreds of millions of dollars in fees to sign settlement agreements.
However, in the long term, NVIDIA needs to comply and guard against this.
Because once it is confirmed that NVIDIA has monopolistic behavior, NVIDIA may face legal consequences such as fines and business restrictions.
This will have an impact on NVIDIA's long-term performance and stock price.
Liu Xu, a special researcher at the National Strategy Research Institute of Tsinghua University, believes that although NVIDIA may have a market-dominating position, there is still a need for sufficient evidence to argue whether it has abused this position.
For example, the U.S. Federal Justice Department is still investigating whether NVIDIA has required cloud service companies to enter into exclusive procurement agreements with it, and whether it has treated customers who also purchase competitors' AI chips differently, offering higher prices when they purchase NVIDIA products.
As early as 2023, France's investigation into NVIDIA had already begun, but NVIDIA's stock price still reached a new high in the first half of 2024, indicating that the impact of antitrust enforcement on the market value of high-tech companies is relatively limited.
Liu Xu further explained that NVIDIA's acquisition of AI companies is not an isolated case.
Other American high-tech companies are also carrying out similar acquisition activities.
Although the U.S. and European Union and the United Kingdom's antitrust regulatory authorities have been stricter in reviewing the acquisition cases of large technology companies in the past two years, most acquisition cases have not been banned, and even if they may cause concerns about restricting competition, they have all been approved after adding restrictive conditions, such as Microsoft's acquisition of Activision Blizzard.
GB200 is NVIDIA's next-generation flagship AI chip.
The chip was released at NVIDIA's developer conference in March 2024 and is expected to begin delivery in the third or fourth quarter of 2024.
It is actually composed of two Blackwell B200 GPUs and one Grace CPU.
Compared to H100, GB200's computing performance has increased by six times.Nvidia's new product delivery has been delayed, and the industry has various interpretations, mainly focusing on the chip design process.
This is because the complexity of the GB200 has increased compared to previous generations.
A chip technology expert explained to us that Nvidia's Blackwell series chips use a large number of new IPs (circuit modules with independent functions), making some bugs difficult to be tested during the Post-Silicon Validation phase.
High-performance memory like HBM3 for high-performance series chips like Blackwell even needs to be connected to AI systems for testing.
Typically, early samples are also provided to customers for testing in actual business scenarios.
Under highly complex product conditions, it is normal to delay shipment to ensure product compatibility.
On August 2nd, JPMorgan Chase and Morgan Stanley each released a report analyzing this.
The common view of JPMorgan Chase and Morgan Stanley is that the problem was discovered before mass production of the chip, so the impact is limited.
They even predict that the aforementioned issue can be resolved within a month.
The market originally expected Blackwell to start production in the third quarter of 2024, with mass shipments in the fourth quarter of 2024, and large-scale server shipments expected in the first quarter of 2025.
Therefore, the current impact is mainly on the production plan for the third quarter of 2024.
With TSMC's capacity expansion in the fourth quarter, the delay in production in the third quarter is expected to be partially compensated.
The shipment of Blackwell this year will be slightly reduced, and the impact on shipments in the first quarter of 2025 will be limited.
JPMorgan Chase believes that chip design issues are the main reason for the product delay.
Morgan Stanley believes that Nvidia may improve the stability of the new Blackwell architecture through redesign.
After researching the foundry supply chain, Morgan Stanley believes that the production of Blackwell chips may be suspended for about two weeks, and production progress will resume in the fourth quarter of 2024.
TSMC can still catch up with Nvidia's Blackwell's expected production time.
The report from JPMorgan Chase shows that the total shipment of GB200 in 2024 was originally expected to be over 600,000 units.
Affected by the delayed shipment, it is expected that the shipment of GB200 in 2024 will be between 400,000 and 500,000 units.
The shortage of GB200 is expected to slightly increase the sales of the previous generation product H200 by 500,000 to 600,000 units in the second half of 2024.
It is expected that the shipment of GB200 in 2025 will no longer be affected and will reach over 4.5 million units.
Citigroup estimates that the delay of GB200 will not have a substantial impact on Nvidia's revenue in the third quarter of fiscal year 2025 (end of July 2024 - end of October 2024), but it will affect 15% of Nvidia's data center business revenue in the fourth quarter of fiscal year 2025 (end of October 2024 - end of January 2025).
Citigroup has lowered its revenue target for Nvidia's fiscal year 2025 (end of January 2024 - end of January 2025) by 5%, and the revenue forecast for fiscal year 2026 remains largely unchanged.
Citigroup still believes that Nvidia is the leader in data center GPUs, with a long-term market share of about 90%, and maintains a "buy" rating with a target price of $150.
The financial impact of the delayed shipment of GB200 is temporarily limited, but as Nvidia's product models increase and the iteration cycle shortens, the difficulty of managing Nvidia's product line and supply chain is increasing.
Nvidia's chip release cycle is getting shorter and shorter.
The interval between the release of V100 and A100 is 36 months, the interval between A100 and H100 is 22 months, the interval between H100 and H200 is 20 months, and the interval between H200 and B100/B200/GB200 is only four months.
The normal usage cycle of an AI chip is three years.
At the stage of V100 to A100, the pace of Nvidia and technology companies' AI chip replacement was consistent.
However, after the release of H100, Nvidia's product iteration pace accelerated, and it needs enough time to ensure the reliability of the product.
Another chip technology expert believes that the acceleration of Nvidia's product iteration is a double-edged sword.
On the one hand, it is beneficial for Nvidia to maintain a market advantage.
Its competitors AMD and Intel each need to spend a certain cycle to adapt for customers when a new generation of products comes out, and the product iteration cycle is longer than Nvidia's.
Nvidia has the CUDA software, which does not require frequent adaptation for customers.
However, the price of Nvidia's acceleration is that the stability of the product may decrease.
If there is no problem, it is still possible, but once there is a problem, it will lead to a situation similar to the delayed shipment of GB200.
Many practitioners in the chip industry believe that antitrust and delayed shipments are not enough to hurt Nvidia's foundation.
Nvidia's real challenge is whether technology companies (Microsoft, Google, Amazon, Meta, etc.)
can achieve tangible performance growth through large models in the next 1-2 years, and whether they can maintain huge capital expenditures to purchase Nvidia's chips.
Nvidia's main customers are large technology companies such as Microsoft, Amazon, Google, and Meta.
They are still purchasing Nvidia's AI chips on a large scale, and have paid 50%-100% of the capital expenditure in the past 2-4 quarters.
The market expects that this high-intensity capital expenditure is unsustainable.
Therefore, the core issue is whether large technology companies can cash in on the AI dividend.
Only when they have made enough money can they pay a high capital expenditure and continue to purchase Nvidia's AI chips.
After the industry chain forms a virtuous cycle, Nvidia's story can continue.
In August of this year, a technology company's strategic planning person who once engaged in chip investment told us that the semiconductor industry usually has a cycle of 36-44 months.
In the upcycle, chip prices and chip demand will rise simultaneously.
However, in the downcycle, chip demand and chip prices will fall simultaneously.
This round of AI chip cycle started at the end of 2022.
Theoretically, the demand for AI chips may peak in the third quarter of 2025 or at the beginning of 2026.
At that time, Nvidia's challenge may really begin.
Leave a Comment