Envision a future where, donning VR glasses, you immediately enter a virtual metropolis to gather with distant friends in a digitized restaurant. Upon returning to reality, your smart home has deftly adjusted the temperature, a robotic assistant has served up freshly brewed coffee, and the ambiance—lights, music, and curtains—has harmoniously shifted to your preferred settings. Stepping outside, a smart car patiently awaits, with its onboard system analyzing real-time traffic data to chart the most efficient route. At the core of this orchestrated symphony is NVIDIA's ingeniously crafted chip, ensuring a tapestry of smooth operations. These AI chips are not merely linking devices but intricately weaving the fabric of a new era in intelligent connectivity—one where NVIDIA not only connects every device but also envisions every conceivable future.

Summary

• NVIDIA's revenue growth has found equilibrium through diversified operations. Notably, gaming and data center departments stand out as primary revenue streams. The gaming sector involves the sale of standalone graphics cards or gaming laptops equipped with high-end GPUs to individual end-users. Meanwhile, data center operations provide servers and cloud computing services to the enterprises in a B2B manner.  
• Riding the wave of the AI revolution, NVIDIA's GPU chips, with their parallel computing prowess, have positioned the company as the global engine of artificial intelligence. The data center business has progressively eclipsed gaming, constituting a substantial 76% of the overall business, as indicated in the latest Q2 financial report.
• NVIDIA exhibits a discernible sensitivity to market supply and demand dynamics, marked by cyclical fluctuations. The potential market for the data center business is estimated at around $250 billion annually. Anticipated is a steady rise in demand for AI language models from cloud service providers over the next 2-6 quarters, averaging a triple increase year over year. However, looming are supply chain risks, including delayed deliveries and geopolitical conflicts.
• Further risks for NVIDIA include potential U.S. export sanctions on China, heightened competition from other cloud service providers, and increased regulatory scrutiny in the AI industry. According to our model, $69 to $86 of NVIDIA's current stock price of $465 is presently tied to China, offering a strategic planning baseline of $380 even in extreme scenarios.
• Casting an eye towards the future, optimism prevails for NVIDIA's long-term prospects. The company's sustained investment in the realms of deep learning and AI chip technology is seen as erecting formidable competitive barriers, effectively translating technological leadership into a substantial market share and fostering profitable growth.

Operations Overview

NVIDIA, primarily an artificial intelligence computing company, is renowned for designing display chips. Its flagship product, the Graphics Processing Unit (GPU), initially designed for graphic computations in personal computers, has evolved into a global engine for artificial intelligence, boasting a market share exceeding 87%. Over the years, NVIDIA has diversified its GPU architecture to create a comprehensive computing platform, catering to various domains such as artificial intelligence, autonomous driving, robotics, the metaverse, and 3D internet applications—a successful pivot in its business strategy.

The company's business operations are segmented into four main categories: Gaming, Data Center, Automotive, and Professional Visualization.

1) Gaming: This segment encompasses the sale of graphics cards, comprising GPU, memory, power supply, and heat dissipation components. Many contemporary games, especially AAA titles, rely on NVIDIA's high-end discrete graphics cards, such as the GeForce RTX40 series, along with their associated software, for scene rendering, elevating gaming experiences and graphical capabilities.

2) Data Center: In contrast to the B2C model in the gaming sector, the Data Center business focuses on a B2B manner, providing services to enterprise clients. NVIDIA's CUDA platform, a parallel computing architecture akin to iOS for iPhones, enables developers to write software for various compute-intensive tasks, including data analysis, high-performance computing, and AI training. Through the sale of GPUs (Volta, Tesla, A100, H100 series), servers, networking equipment, software, and services, NVIDIA generates revenue from cloud service providers, enterprises, and research institutions.

• A100 Chip: This GPU model holds paramount importance in the Data Center business. For instance, running ChatGPT requires approximately 30,000 A100 chips, with the market price for A100 ranging between $1,000 and $15,000.
• H100 Chip: An upgraded version of the A100 chip, primarily utilized for large language model generative AI.
• L4/L40 Chips: GPU chips tailored for AI video and AI image applications.

• Grace & Grace Hopper CPU Super Chip: Specifically designed for large-scale AI and high-performance computing, requiring integration with GPU chips.
• BlueField Data Processor: A network card based on NVIDIA's DPU data processing technology, dedicated to handling communication protocols to reduce the CPU's processing burden and accelerate computing.

• DGX H100 Server: Comprising 8 H100 GPUs, a CPU, DPU data processor, 2TB system memory, and SSD high-speed storage, the DGX H100 server can be locally deployed by enterprises or accessed through cloud services, with a price tag of approximately $400,000.

• DGX BasePOD Supercomputer: Offering higher computing speed, greater storage capacity, and enhanced functionality than conventional large computers, the DGX BasePOD is priced around $7-60 million.

• DGX GH200 Supercomputer: A high-memory AI supercomputer integrating 256 GH200 chips, slated for release later this year, with early access granted to companies like Google, Meta, and Microsoft.

• HGX Supercomputer: Similar to DGX but with improvements designed by NVIDIA's partners.

3) Automotive: NVIDIA provides end-to-end solutions for autonomous driving vehicles, including the DRIVE platform, Xavier chip, and AGX Orin in-car chip. While the current revenue share from this sector is modest, NVIDIA has already become a processor supplier for Tesla's in-car domain, and the future holds significant growth potential with the widespread adoption of autonomous driving.

4) Professional Visualization: This segment involves the sale of Quadro series GPUs for workstation-level tasks, assisting professionals in visualizing clusters and providing corresponding software and toolkits, including OpenGL and DirectX.

Gaming and Data Center stand out as NVIDIA's primary revenue sources. Gaming has been the company's core since its inception, and its strategic focus on GPUs has set it apart from competitors like AMD and Intel. The year 2021 saw historic opportunities in mining, severe supply chain shortages, a booming metaverse concept, and the pandemic significantly boosting gaming demand. The market gradually recognized the critical role of high-performance GPUs in computing, leading NVIDIA's stock price to soar from around $100 to a peak of $334 in 2021. The gaming business's revenue surged from $1.5 billion in the fiscal year 2014 to $9 billion in the fiscal year 2023, representing a CAGR of 19.62%.

On the Data Center front, the GPU's ability to handle larger volumes of parallel computing tasks compared to CPU makes it well-suited for AI and high-performance computing. This has propelled Data Center revenue from approximately $200 million in the fiscal year 2014 to $15 billion in the fiscal year 2023, with a CARG of 54.07%. The Data Center's share of total revenue increased from 4.8% in the fiscal year 2014 to 76.4% in the Q2 of the fiscal year 2024.  

NVIDIA is now entering a phase of multi-chip development, moving beyond monopolizing the GPU market. The complex and diverse computing scenarios of the future demand the interconnection of different types of chips to achieve higher performance. Through an integrated CPU, GPU, DPU strategy, NVIDIA has released groundbreaking products such as the world's eighth supercomputer DGX BasePOD, the breakthrough acceleration superchip CPU Grace Hopper, and the computing platform BlueField-3 DPU. These products solidify NVIDIA's dominant position in the field of artificial intelligence.

In summary, NVIDIA, the mother of GPU, has achieved specialization in GPU technology through substantial research and development investments. In the past fifteen years, the company has constructed a comprehensive computing platform using software as a carrier, exemplified by the CUDA programming system, CUDA-X application libraries, application interfaces, API interfaces, and SDK software tool development kits. The parallel computing capabilities of GPU, particularly in the context of trends like mining, the metaverse, and artificial intelligence, have positioned NVIDIA as the world engine for artificial intelligence (Nvidia now powers 70% of the Top 500 supercomputers). This strategic positioning, coupled with the diverse product lineup, is building a robust ecosystem, forming a deep moat for NVIDIA in the rapidly evolving landscape of artificial intelligence and computing.

Demand and Supply

NVIDIA is sensitive to supply and demand dynamics, experiencing periodic fluctuations. During periods of high market demand and tight inventory, NVIDIA can sell products at full price, yielding higher profit margins. Conversely, during weak demand and excessive inventory, NVIDIA may have to sell products at discounted prices, putting pressure on profit margins. Over the past five years, NVIDIA's stock price has experienced a maximum drawdown of 52.5% post-earnings season, while its maximum gain reached 90%.

On the demand side, sustained high prosperity is anticipated over the next two years. NVIDIA envisions the potential TAM for its data center business over the next four years to be around $1 trillion, with chips accounting for 30% and software systems for 15%. This implies an annual growth space of $112.5 billion in revenue (the company's latest disclosed revenue for 2023 Q2 is $13.5 billion). Simultaneously, NVIDIA's data center revenue surged to $10.3 billion in the second quarter, marking a sequential growth of 141%. Historically, the business has tripled on average each year, primarily propelled by the demand for HGX systems (NVIDIA's engine designed for generative AI and large language models) from dominant cloud service providers, including Amazon, Google, Meta, Microsoft, and Oracle.

In the upcoming 2-6 quarters, the demand growth seems nearly Infinite. Colette, NVIDIA's CFO, underscored in the latest earnings call that they are "substantially" expanding capacity to meet surging customer orders. Furthermore, major cloud service providers, constituting technology giants, contributed over 50% of NVIDIA's data center revenue in the second quarter. These giants boast robust cash flows and are actively fortifying their dominant positions in the field of artificial intelligence. Consequently, unlike Tesla, which may be susceptible to a dip in macroeconomic demand, NVIDIA's customer profile ensures the stability of its order volume. Even if a hard landing senerio transpires in the next two years, technology giants are unlikely to curtail expenditures on AI infrastructure.

On the supply side, NVIDIA, without chip manufacturing capabilities of its own, collaborates with suppliers such as TSMC, Samsung, Foxconn, and BYD, leveraging their foundry services to complete the entire manufacturing process. TSMC alone accounts for 43.21% of the entire supply. Analogous to Apple, NVIDIA wields substantial bargaining power. For instance, in the fierce competition between Samsung and SK Hynix for the opportunity to enter NVIDIA's HBM memory supply chain, Nvidia’s overall gross margins exceeded 70% in the last quarter. NVIDIA also develops the CuLitho software, incorporating deep learning technology to optimize the lithography process, increasing the speed of computational lithography by 40 times. This aids suppliers like TSMC and ASML in better controlling the focus and exposure of lithography machines, thereby enhancing the accuracy and yield of chip manufacturing.  

In the next two years, a continuation of an undersupplied market is anticipated. However, considerations related to supply alone introduce some risk factors. Since NVIDIA heavily relies on third-party manufacturers such as TSMC and Samsung, geopolitical conflicts in East Asia may result in extended lead times. Prolonged delivery cycles can lead to increased product costs, as evidenced by NVIDIA's experience with more than a 12-month lead time, necessitating payment of high premiums to ensure a secure supply. Additionally, NVIDIA needs to accurately gauge the intricate relationship between demand and supply. Any imbalance leading to excessive shortages or excess inventory can significantly impact financial metrics.

Downside Risks

U.S. Export Sanctions

As of the latest financial report released in August, NVIDIA derives 44.7% of its revenue from the United States, followed by Taiwan (21%) and mainland China (20.3%).

Previously, the U.S. announced stricter export restrictions on the Chinese semiconductor industry, imposing limits on products, including the A100 and H100 chips (primarily used for training AI language models), and any future cutting-edge NVIDIA products destined for mainland China, Hong Kong, Macau, and Russia. Currently, NVIDIA has circumvented sensitive technology by producing a special A800 chip for China, degrading some features of the A100 chip (e.g., reducing the interconnect data transfer rate from 600GB per second to 400GB per second).

However, with the upcoming 2024 U.S. presidential election, despite recent easing in U.S.-China relations, we believe that in an election year, both Democratic and Republican parties will exert mutual pressure on issues related to China's technology national security. In this scenario, further U.S. sanctions on the Chinese semiconductor industry cannot be ruled out, potentially affecting NVIDIA's profits. According to our estimates, Chinese business accounts for $69 to $86 of NVIDIA's current stock price of $465. In an extreme situation, if NVIDIA were to lose all its China-related business, the stock price could drop to around $380 per share.

Competition from Cloud Service Providers

Currently, technology giants are engaged in an arms race to secure a share of the AI market. As mentioned earlier, NVIDIA's stock price has seen significant gains this year due to the demand from major cloud service providers for HGX systems used to train AI language models. However, companies including Google, Amazon, Meta, Alibaba, and Baidu are also establishing internal teams to develop chips and software with AI capabilities. They aim to ensure the security and reliability of their supply chains and exploit the vast market.

Since GPUs use Mac algorithms, which are best suited for 3D rendering mapping, there is some energy loss when using this algorithm for AI language model training. However, as GPUs currently offer the best cost-performance ratio in the market, they have virtually monopolized the market for chips designed for AI models. Nevertheless, Google has already introduced its AI chip, the Tensor Processing Unit (TPU), widely deployed in its data centers. Therefore, it's challenging to assert that GPU chips will remain worry-free.

It is evident that the competition in the AI chip and software market remains intense. Over the past two years, NVIDIA has continuously adjusted chip architecture to adapt to AI characteristics, attempted to acquire ARM (ultimately unsuccessful), and is developing products more suitable for AI. We believe that since chip development is a long-term process requiring significant capex and manpower investment, NVIDIA still holds an advantage over other giants in the short term.

AI Regulation & Antitrust

With the rapid development of AI technology, regulatory bodies may impose more requirements on the compliance and ethical issues of its applications. For example, this year, Musk called on the U.S. government to intervene and ban OpenAI from developing the next stage of ChatGPT. This could pose some risks to NVIDIA.

The most extreme scenario is government intervention to restrict the sale of NVIDIA's AI chips and software, resulting in the company losing significant customers and revenue sources. However, we believe the likelihood of this extreme scenario is minimal. This is because AI is currently a battlefield for countries, especially in the competition between China and the U.S. Even if the U.S. tightens regulatory measures, a complete ban is unlikely.

More likely scenarios include regulators requiring NVIDIA to approve or certify its products, thereby invisibly increasing its costs. Alternatively, regulatory agencies may encourage or require other tech companies to develop products that can replace NVIDIA's AI chips and software, leading to a reduction in NVIDIA's market share. However, given NVIDIA's current leadership position in the industry, it can potentially pass on the additional costs arising from AI regulation to the downstream buyers.

Investment Ideas

With the current market sentiment relatively optimistic, a potential market pullback in the future could present favorable entry opportunities.

• Buy Cash Stocks: Plan entries around the $400 mark, possibly waiting for a pullback to this level before making arrangements.
• FCN: Constantly get the interest income during the time span before maturity of the notes. In a scenario where stock price experiences a decline, investors will have a chance to purchase the underlying at a discounted price of 70-80%.

• Accumulator: Consistently acquiring positions at a substantial discount over 9 months to 1 year, so that the probability of success over an extended timeframe could be enhanced.

‍