The reason why Nvidia surpassed 4 trillion
Written by: RockFlow
Key Points
① NVIDIA benefits from the overall ecological network effect: CUDA, installation base, system-level integration, optimization, etc. Each advantage complements and reinforces each other, forming a strong technical barrier.
② Nvidia has exceeded revenue expectations for the past six consecutive quarters. What surprises the market is that despite raising expectations every quarter, it still finds ways to surprise investors with further significant leads.
③ In the past, when Internet giants decided to use a certain technology to build business barriers, they considered growth more. AI For the giants, if they don’t keep up, their business foundation may be gone. Today, they no longer make decisions based on calculating the returns and profits, but on the fear of absolute losses. Buying Nvidia chips is essentially buying insurance, which has no flexibility.
④ In the long run, technology giants will continue to look for high-performance GPU sources outside of Nvidia or internal solutions to get rid of their dependence on it. Most likely, these efforts will gradually weaken, but not replace, Nvidia's AI Dominance in the field.
Every now and then, a star company emerges, attracting much attention because of its soaring stock price over the years, or causing huge controversy because of its violent fluctuations. This time it is Nvidia, AI The global chip giant that provides power has seen its stock price soar nearly 50 times in five years and is currently ranked among the top three in the world in terms of market value along with Microsoft and Apple.
In a recent podcast on Crossroads, Vakee, founder of RockFlow, expressed her opinion on whether Nvidia's stock price is in a bubble. She does not think Nvidia is in a bubble. A forward P/E (forward price-to-earnings ratio) of 40 times is a reasonable range for such a leading company.
The recent pullback of semiconductor stocks, including the overall correction of U.S. stocks due to excessive gains, was in line with normal market changes; in addition, the stronger expectations of interest rate cuts at the time led to increased risk appetite, and some funds flowed from leading companies to more aggressive small-cap stocks, which was also a normal sector rotation; further, the tightening of U.S. export restrictions and geopolitical tensions a few weeks ago were also factors in the decline of chip stocks represented by Nvidia, but again, these are all normal market reactions.
As for the sharp fluctuations in the US stock market in the past week or two, it is mainly due to a combination of factors such as weak US labor market data, the lack of surprises from the earnings season giants, the retracement of the yen carry trading, and potential conflicts in the Middle East. However, apart from these external factors, as far as Nvidia is concerned, we believe that with our mid- to long-term optimism about AI (it is likely to be the biggest opportunity for change in our generation), there is still no bubble in the absolute leading stocks in this industry.
Nvidia's performance in the past year is impressive enough: its share of the AI chip market is about 90%, its annual revenue exceeds 60 billion US dollars, and its net profit margin exceeds 50%. In the past five years, its annual revenue compound growth rate has reached 64%, far ahead of all S&P 500 companies.
Regarding the company's development history and investment value, RoXiaobai NavigationThe ckFlow investment research team has made a detailed analysis in an article at the beginning of last year——In the battle of big AI models, can Nvidia have the last laugh?
In this article, we hope to answer the question of Nvidia’s true moat, which the market has not paid enough attention to, and why we believe that Nvidia is not only a great company, but also an investment target with high potential returns.
1. Apart from CUDA, what is Nvidia’s moat?
Investors who are bullish on Nvidia believe they are betting on its future. After factoring in Nvidia's higher expected growth, it is no more expensive than other tech giants.
But there’s a problem: The further out the market projects earnings, the more uncertain its forecasts become. Microsoft and Apple are established businesses that rely on their existing, stable customer bases.make moneyNvidia serves a newer but more promising market, so investors currently have much greater differences in their views on Nvidia's prospects than on Microsoft and Apple.
Nvidia has long been considered a top producer of gaming graphics cards. With the rise of crypto mining, GPUs, the core of graphics cards, have become increasingly popular. Nvidia's GPUs are highly optimized for "parallel processing" - breaking down computationally difficult problems and distributing them simultaneously to thousands of processor cores on the GPU, thereby solving problems faster and more efficiently than traditional computing methods.
In addition to designing the most advanced GPUs on the market, NVIDIA also created a programming model and parallel computing platform - Compute Unified Device Architecture (CUDA), which has become an industry standard and makes it easier for developers to use the power of NVIDIA GPUs.
But NVIDIA relies on more than just the CUDA platform that everyone praises today. The RockFlow investment research team believes that NVIDIA benefits from the overall ecological network effect: CUDA, installation base, system-level integration, optimization, etc., and actively chooses advanced solutions in bandwidth and network to improve computing performance. Each advantage complements and strengthens each other, forming a strong technical barrier.
Let’s look at the installed base first. CUDA’s decades of leadership means that it benefits from a strong network effect—its large installed base incentivizes frameworks and developers to target it, which in turn attracts more users to adopt it. NVIDIA has a large number of users in areas such as gaming, professional visualization, and data centers. The large user base provides NVIDIA with a continuous source of revenue and product feedback, and this scale effect also enables NVIDIA to continue to invest in research and development and maintain its technological leadership.
The second is system-level integration capabilities. NVIDIA not only provides GPU hardware, but also provides a supporting software stack. It can be said that from drivers to development tools to optimization libraries, it has already formed a complete ecosystem. This vertical integration enables NVIDIA to optimize at the system level and provide better performance and user experience.
As for the optimization level, NVIDIA has carried out in-depth optimization at both the hardware and software levels. In terms of hardware, it continuously improves the GPU architecture to enhance performance and energy efficiency; in terms of software, it optimizes drivers and libraries to fully tap the potential of hardware. System-level optimization, such as multi-GPU collaborative work and GPU direct memory access, further improves overall performance.
In order to solve bandwidth and network problems, NVIDIA has also made many attempts. It has launched a series of technologies to improve the efficiency of data transmission between GPU and CPU and GPU, the most important of which is NVLink. NVLink directly connects multiple GPUs through high bandwidth, significantly improving the efficiency of AI computing. This allows NVIDIA to maintain a strong position in the fields of autonomous driving and AI computing. In addition, betting on InfiniBand, acquiring Mellanox, and moving closer to the Ethernet platform through NVIDIA Spectrum-X are also NVIDIA's active layout in the network.
The RockFlow research team believes that for NVIDIA today, the huge installation base provides the power and data for its continuous optimization of GPU design, while system-level integration and optimization enhance user stickiness and expand the installation base. Coupled with the continuous iteration of many advanced solutions to bandwidth and network issues, the virtuous cycle between them enables NVIDIA to maintain its leading position in the field of GPU and accelerated computing.
Therefore, even though competitors AMD and Intel have launched similar GPU products with price advantages, Nvidia still has an absolute dominant position in the AI chip market.
It is estimated that NVIDIA has firmly occupied more than 90% of the data center GPU market in the past 7 years. In 2023, its share will reach 98%. The operation of all large data centers and the training of large models need to rely on GPUs developed by NVIDIA.
In the long run, it may be difficult for Nvidia to completely maintain its current market share, but as the data center GPU and other AI chip markets continue to grow, its moat will ensure that it gets the vast majority of orders. Nvidia is expected to always occupy an important position in this new industrial revolution and maintain healthy and long-term growth.
2. Nvidia's performance miracle: expectations have been raised repeatedly, but it still brings surprises
More than two months ago, Nvidia announced its first quarter results and its stock price soared 10% the next day. Since then, its market value has skyrocketed, replacing Microsoft to become the world's top market value.
Last quarter, Nvidia's specific revenue is as follows:
Revenue increased by 18% month-on-month to $26 billion (exceeding expectations by $1.5 billion). Among them, the data center division's revenue increased by 23% month-on-month to $22.6 billion, becoming the largest and fastest growing division.
How exaggerated is its growth rate? You can see from the following figure:
In fact, Nvidia has exceeded revenue expectations for the past six consecutive quarters. What surprises the market is that despite raising expectations every quarter, it still finds ways to surprise investors with further significant leads. At the same time, the profit margins of Nvidia's chip business have also increased significantly.
In the subsequent earnings conference, Huang Renxun said:
“The next industrial revolution has begun. Major companies and countries are working with NVIDIA to shift trillions of dollars of traditional data centers to accelerated computing and build a new type of data center - the AI factory - to produce a new commodity - AI.”
There are two core stages in the operation of AI systems: training - AI learns from large amounts of data, developing intelligence and pattern recognition. Currently, Nvidia's powerful GPUs dominate this stage; reasoning - AI applies its knowledge to real-world tasks and decisions. Despite facing more intense competition, Nvidia is making significant progress.
Inference workloads contributed approximately 40% of Nvidia’s data center revenue in the past year. The current market consensus is that as more and more generative AI applications emerge, inference is expected to become a huge market and bring considerable returns on investment to Nvidia’s customers.
Currently, NVIDIA mainly divides customers into three categories:
-
Cloud service providers (CSPs) contribute nearly half of data center revenue. All the giants (Amazon, Microsoft, Google, etc.) are Nvidia’s customers;
-
B-side enterprises drove strong sequential growth. For example, Tesla expanded its training AI cluster to 35,000 H100 GPUs and used them for FSD V12;
-
C-end companies are also a key vertical category. Taking Meta as an example, its Llama 3 is trained on a cluster of 24,000 H100 GPUs. It is expected to use 240,000 GPUs, ten times the computing power of Llama 3, to train the next generation of multimodal Llama 4.
Nvidia management also pointed out some important directions on the conference call:
Regarding data centers, they believe:
“As generative AI enters more and more 2C Internet applications, we expect to see continued growth opportunities as inference scales with model complexity as well as the number of users and queries per user, driving more demand for AI computing.”
In addition, Nvidia is very optimistic about the concept of "sovereign AI":
“Sovereign AI refers to a country’s ability to produce AI using its own infrastructure, data, labor, and business networks. The importance of AI has attracted the attention of every country, and major powers will pay more attention to the control of AI technology. We believe that sovereign AI will bring billions of dollars in revenue this year.”
Huang Renxun also made relevant disclosures on the impact of US export restrictions, the new generation H200 and Blackwell architecture, etc. The continuous rise in Nvidia's stock price in the following weeks fully proved that the market recognized his views and was optimistic about the long-term development of the AI wave.
3. The biggest challenge is still the giants’ own research
As mentioned above, the network effect built by NVIDIA around the CUDA ecosystem is an excellent example of a commercial company shaping a complete ecosystem. NVIDIA's great success is not only due to the excellent performance of its chips, but also because it has firmly established itself at the center of the generative AI wave by virtue of the entire network layout of related hardware and software.
But at the same time, its strategy of bundling basic software with chips has also drawn strong criticism from customers and competitors, and even regulators are repeatedly launching antitrust investigations due to its extremely high market share.
Nvidia faces competition from AMD and other chipmakers, including Qualcomm, Intel, etc. These companies are essentially chip designers, and they almost all use the same outsourcing company to manufacture chips - TSMC.
Both AMD and Intel have launched their own data center GPUs, aiming to take back market share from Nvidia's H100/H200 - Intel launched the Gaudi3 AI acceleration chip, and AMD launched the MI300 series. 2024 may be the first time that Nvidia cedes a small portion of market share to AMD, and Intel is expected to take back a small portion of market share.
You only need to look at one piece of data to understand why competitors are so determined: According to the latest quarterly financial reports, Microsoft, Google and Meta's total capital expenditure in the second quarter of 2024 exceeded US$40 billion - most of which was spent on AI.
With so many giants choosing to invest heavily, is it possible that their decision makers will one day start to calculate the ROI of AI costs due to internal and external pressures, and find that it is not cost-effective, so they start to slow down their AI investment?
Our answer is almost impossible. In the past, when Internet giants decided to use a certain technology or enhance a certain capability to build business barriers, they considered more about growth - that is, whether it could accelerate business development. But this time, generative AI has a big change. For these giants, if they don't keep up, the company and business foundation may be gone. Therefore, this time, the arms race of generative AI, for the purchase of Nvidia chips, is essentially more like the logic of buying "medicine". When a technology is the difference between life and death for a company, it will not think too much about the details.
This is not an exaggeration. For example, the search business may be completely changed by AI. Whether it is Google or Baidu, their decision-making ideas today are no longer based on calculating the returns and profits, but on the fear of absolute losses. Buying Nvidia chips is essentially buying insurance, and there is no flexibility at this time.
Precisely because high-performance chips are so important, the giants are determined to develop their own or find alternative solutions while spending huge amounts of capital expenditures. Faced with Nvidia's comprehensive CUDA ecosystem layout, other companies (in fact, almost all competitors) are trying to jointly develop open solutions to break Nvidia's monopoly on the AI software and hardware ecosystem.
Intel, Google, ARM, Qualcomm, Samsung and other tech companies are working together to develop new software suites that will let developers’ code run on any machine with any chip. OpenAI is also working on it, releasing an open source language that allows researchers without CUDA experience to write GPU code, and the open source PyTorch Foundation, incubated by Meta, is also making similar attempts.
These companies are also working to replace Nvidia’s proprietary hardware, including developing new ways to connect multiple AI chips within and across servers. Intel, Microsoft, Meta, AMD, Broadcom and others hope to set new industry standards for this connectivity technology that is critical to modern data centers. The clash between proprietary and open solutions is a bit like Apple and Google’s Android in the smartphone market, and as we’ve seen, both closed and open visions are enough to create winners.
In the long run, tech giants will continue to look for sources of high-performance GPUs outside of Nvidia or internal solutions to get rid of their dependence on Nvidia. Most likely, these efforts will gradually weaken, but not replace, Nvidia's dominance in the field of AI.
4. Conclusion
In the past few years, Nvidia's chips have pushed the company's profitability to new heights. The RockFlow investment research team has long been optimistic about Nvidia. Previously, RockFlow founder Vakee was asked in the Crossroads podcast, "Does Nvidia have any signs that it may be attacked by competitors?" Her opinion is that there are no such signals.
NVIDIA's advantage today is not a certain product, or the hardware advantage of a single GPU. NVIDIA has built core competitiveness in a complex ecosystem, so it is particularly difficult to break through. In addition to the CUDA platform, its advantages also include the current number of users, the entire basic installation volume, the overall system integration capabilities, and the ability to continuously optimize. These advantages reinforce each other, so the overall competitive barriers will continue to stand out.
Faced with potential problems in the ecosystem, NVIDIA can spare no effort to solve them, whether it is acquisitions or investments. This continuous investment, whether in hardware or the integration capabilities of the entire system, and the solution of network bandwidth problems, is constantly increasing its own barriers. This virtuous cycle will make its dominant position more stable.
Therefore, we believe that it can still replicate its previous growth momentum and release greater value in the long run.
The article comes from the Internet:The reason why Nvidia surpassed 4 trillion
Polychain says Niraj Pant secretly served as a “consultant” to EclipseToken, violating the fund’s policies. Author: Sam Kessler Translator: DeThings Editor’s Note: Niraj Pant, a former general partner at a top crypto venture capital firm, and portfolio company Eclipse…
相关文章
