How does Nvidia make money?
????If you hope to meet more often, please mark the star ?????? and add it to your collection~
In 2023, NVIDIA's total revenue will be $60.922 billion, with a gross profit margin of 72.7% and a data center GPU market share of 98%. As of October 6, 2024, NVIDIA's market value will reach $3.064 trillion, ranking third in the world.
The author analyzes NVIDIA's current achievements and the main driving forces of its long-term development from three dimensions: product depth, product breadth, and sales channels. As shown in Figure 1, specifically, from the perspective of product depth, NVIDIA has keenly grasped the market rules, continued to "extend the chain" and develop in depth, greatly enriched the product connotation, and increased the added value; from the perspective of product breadth, NVIDIA uses open platforms and flexible architectures to expand product application areas, deeply integrates ecosystems in various fields to become mainstream tools; from the perspective of sales channels, NVIDIA builds a sales network that is collaborative and mutually beneficial with vertical market service platforms to achieve a win-win situation for all parties.
Figure 1: Three factors drive NVIDIA's long-term development
1. Product Depth
Continue to "extend the supply chain" and enhance product added value
The in-depth development of products is reflected in NVIDIA's continuous "chain extension" based on market demand, from GPU chip hardware, to "GPU chip hardware + software", to the "GPU + DPU + CPU" three-core strategy, and then to data center products. NVIDIA continuously integrates new technologies into existing products as infrastructure, continuously expands product definitions and connotations, and enhances their added value. At the same time, it integrates strategies such as system optimization, differentiated competition, and technology acquisitions.
1
Software-defined products to enhance added value
NVIDIA was originally a PC graphics chip company that focused on the gaming market and provided customers with industry-standard graphics chips, namely GPU hardware itself. NVIDIA has transformed from a hardware supplier to a computing platform in key vertical fields, from building industry-standard chips to providing customers with complete solutions, and its product connotation has changed from hardware to "hardware + software" (i.e. chips, systems and differentiated software built on them).
Figure 2: Schematic diagram of the product from GPU chip hardware to "GPU chip hardware + software"
NVIDIA has since entered a broader market space and opened up channels for profit growth. NVIDIA believes that in a computing era, the ratio of software to hardware is 8:2, and the market space for software and ecosystem is broader. If the business continues to be limited to the hardware market, it will share 20% of the market with a few market players. On the other hand, NVIDIA redefines the connotation of products, changing fixed-function, hardware-defined products to software-defined products, and continuously adding new functions and improving performance through software, thereby obtaining higher added value, which is conducive to product repricing, increasing gross profit margins, and opening up room for profit growth.
2
System optimization computing platform to achieve exponential performance improvement
NVIDIA's strategy is to consider the GPU and the systems and software on it as a whole, achieving performance improvements in the entire process from architecture, chips, chip design, system components, top-level design, system software, and algorithms, and is not limited to popular areas such as chip manufacturing and deep learning algorithms.
Based on the system optimization strategy, starting with the Kepler architecture, NVIDIA launches a new framework every two years. In just five years, Maxwell is 10 times better than Kepler, and Pascal is 10 times better than Maxwell, achieving a product iteration rate faster than Moore's Law. In recent years, NVIDIA has reduced the cycle of launching new architectures or new products to 6 months.
Improving product performance means bringing more economic benefits to customers or reducing costs, which means that the product itself is more valuable, which is conducive to improving product pricing and gross profit margins. Taking the data center business as an example, NVIDIA's Volta processor can replace a large number of commercial servers, network interface cards and cables. Customers can save hundreds of thousands of dollars for each Volta processor they purchase. The more GPUs they purchase, the higher the cost savings. Internet companies use deep learning to improve recommendation system services. The more complex the model and the greater the amount of calculation, the higher the accuracy. Every 1% increase in accuracy brings extraordinary economic impact.
3
Building infrastructure chip DPU,
Eroding the CPU market with differentiated value
When NVIDIA positions its products, it first considers the differentiated value created by the new products, and uses differentiated value as a competitive strategy to gradually erode the CPU market.
Taking DPU as an example, in today's environment of deglobalization and local conflicts, more attention is paid to the security and stability of data centers. The differentiated value of DPU compared to CPU is reflected in security and stability.
In the past, the data center architecture consisted of about 20% to 40% of the CPU running the infrastructure (including operating systems and security services), and the infrastructure and applications ran on the same processor. Since data centers are shared by everyone through multi-tenant clouds and public clouds, data centers have become a zero-trust environment. In order to prevent one of the applications from being malware and entering the data center operating system by attacking the processor, the infrastructure should be isolated from the application, and each application should be protected separately and zero-trust computing should be ensured.
Based on the aforementioned data center security requirements, NVIDIA created a programmable acceleration infrastructure processor DPU to replace the original CPU, specifically running the data center infrastructure, turning the infrastructure into a chip to achieve real-time monitoring of data packets and applications to prevent intrusion. At the same time, 20% to 40% of the original CPU running infrastructure was released to run applications instead.
Using similar competitive strategies, NVIDIA uses GPU parallel computing, mathematical capabilities, and high throughput as entry points to replace CPUs, and uses Arm CPU's energy saving as entry points to replace x86 CPUs. So far, NVIDIA has built a "GPU+DPU+CPU" three-core strategy. The result of cannibalizing the CPU market is specifically reflected in the widespread transformation of data centers around the world, with their budget expenditures shifting significantly from traditional CPU computing to GPU accelerated computing, smart network cards, etc.
Figure 3: Schematic diagram of the product's three-core strategy from "GPU chip hardware + software" to "GPU + DPU + CPU"
4
Extend the chain to develop data center products and keenly grasp market trends
With breakthroughs in deep recommendation systems, natural language understanding, and conversational AI, AI is experiencing explosive growth. For example, Microsoft and NVIDIA worked together to train the neural network model Megatron, increasing the number of model parameters from 7.5 billion to 17.5 billion. This reveals the changing trend of the computing market: the amount of data and computing demand are growing exponentially, and a single processor cannot meet computing needs, driving the computing market to gradually shift from using servers as computing units to using data centers as computing units.
NVIDIA has a keen insight into this market development trend and has included data centers and supercomputers in its product range. NVIDIA's data center product delivery is measured in weeks, and it has built five data centers. It took four weeks to build Selene, the fastest and most energy-efficient AI self-use data center, and it also helps companies around the world build data centers.
Figure 4: Schematic diagram of products from the "GPU+DPU+CPU" three-core strategy to data center products
"System performance optimization" has also been expanded from the original GPU-based system scope to full-process optimization at the data center level, from system software, algorithms, networks to data centers, to achieve the highest throughput at the lowest cost.
5
Acquire high-performance network technology to quickly fill technical gaps
With the rapid development of large models and big data, one server can no longer meet computing needs. Distributed computing based on data centers must be implemented. Therefore, data centers are turning to a new architecture of disaggregation and hyper-convergence. Specifically, a service or application is broken down into small pieces and run on multiple servers in a data center. The answers after the disaggregation and running are re-converged through the high-speed network inside the data center. Therefore, this new architecture of disaggregation and convergence makes high-performance networks between computing nodes essential. Small incremental investments in high-performance networks will increase data center throughput by billions of dollars, greatly improve economic benefits, and immediately recover investment costs.
NVIDIA was originally known for its chip-based technologies such as GPU and DPU. Given the important position of high-performance networks in data center products and the business needs of developing data center products, NVIDIA acquired Mellanox to acquire high-performance network technology, quickly filling in the technical gaps and entering the data center industry.
Mellanox is the network with the lowest latency, highest performance and highest bandwidth. It replaces the original communication function of the CPU in the data center, greatly improving the throughput of the data center. It is used in 60% of the world's supercomputers and 100% of artificial intelligence supercomputers. On April 27, 2020, NVIDIA acquired Mellanox to become its network department. 3,000 Mellanox employees joined, and NVIDIA acquired its deep network technology.
2. Product Breadth
Expand product application areas and become mainstream in various fields
Based on its existing products and professional advantages, NVIDIA has used open platforms and flexible business organizational structures to widely expand the application areas of its products. Through deep integration of vertical field ecosystems, it has been widely used in various application fields and has become a mainstream tool.
1
Extract universal needs from special applications,
Expand business areas based on existing advantages
NVIDIA started out in the gaming industry, solving computer vision problems in games. Founder Huang Renxun realized that in addition to gaming, medical and industrial design fields also have computer vision needs. In 2012, AlexNet convolutional neural network (CNN) won the ImageNet challenge with an overwhelming advantage, surpassing the closest competitor by 9.8 percentage points, using NVIDIA's GPU, demonstrating the GPU's excellent parallel computing capabilities.
NVIDIA extracts general market demand from the field of gaming. Whether it is games, visualization fields such as medical vision, or model calculations in challenge competitions, they essentially reflect the parallel computing needs in various industries. NVIDIA has accumulated profound parallel computing technology in the field of gaming. Therefore, by adjusting existing GPUs, efficiently launching GPUs suitable for new businesses, and solving parallel computing problems in other fields, NVIDIA has expanded its business areas from gaming to professional visualization and data centers (accelerated computing), broadening the application areas of its products based on its existing professional advantages.
2
The same architecture serves multiple fields.
The structure of the organization supports the expansion of vertical markets
Figure 5: NVIDIA uses the same architecture to serve multiple fields
NVIDIA invests in a core hardware architecture to serve four growth markets: gaming, professional visualization, data centers, and automobiles. That is, it unifies the same or similar parts of multiple business areas into one architecture, builds system configurations and software stacks suitable for different business areas on the same architecture, and serves vertical fields with software stacks.
This organizational structure supports NVIDIA in seizing market opportunities and expanding into new business areas. NVIDIA works with tens of thousands of startups and is exposed to potential business opportunities in all walks of life. When business opportunities with huge development potential emerge, NVIDIA can save the process of building hardware architectures for special fields, build software stacks for new fields based on existing architectures, quickly enter new industry tracks, seize rapidly changing market opportunities, and expand vertical markets.
In addition, this organizational form of architecture improves investment efficiency and operational efficiency. From the perspective of investment efficiency, different business areas share the same architecture to avoid repeated hardware development investment; when business opportunities with good development prospects and still in the early stages appear, NVIDIA does not need to invest in special field hardware equipment, avoiding over-investment in new markets, and at the same time, follow up on the development of new markets through small incremental software stack investments, without losing long-term development opportunities, and maximizing investment utility. From the perspective of operational efficiency, focusing on one architecture, by maintaining and upgrading one architecture, all business areas can be upgraded synchronously, reducing operating costs and improving operational efficiency.
3
Open platforms continue to generate innovative applications.
Business expansion has direction and grasp
Huang Renxun believes that every turning point in the way people develop computers is because computers are easier to program and easier to access, such as the PC revolution, the Internet revolution, and the mobile cloud. Because it is very easy to develop and deploy applications, 5 million applications have emerged in the mobile cloud.
NVIDIA has summarized the development rules of the computing industry and is constantly improving its accelerated computing platform while improving the applicability of its platform to eliminate any barriers to access for developers. Specifically, NVIDIA has built a broad partner network with original equipment manufacturers, cloud computing companies, and Internet companies, making the accelerated computing platform applicable to any development environment such as PCs, workstations, clouds, cars, robots, and embedded environments.
Because the accelerated computing platform is simple to program and easy to access, it gradually attracts developers from all walks of life to build tools on the NVIDIA platform. After 18 years of development, by the end of 2023, NVIDIA will be able to support more than 4.7 million developers worldwide. The expansion of developer coverage will accelerate the attraction of more developers to use the NVIDIA platform, forming a virtuous circle, because developers can reach the most end users after developing software, thereby building businesses and getting returns. NVIDIA is also constantly attracting new applications and expanding the application scenarios of its computing platform.
4
Deeply integrate vertical ecosystems,
Going mainstream in vertical markets
Huang Renxun believes that when building a service platform in a vertical field, it is necessary to deeply integrate with the large-scale ecosystem in the vertical field so that NVIDIA's platform and technology can be widely used in vertical industries, directly affect production and operations, and become the mainstream tools in the vertical market.
Specifically, NVIDIA has built vertical market teams, created APIs for each industry, assisted developers in each industry in using its technology, and worked with them to refactor applications so that applications using NVIDIA technology are more in line with market demand, thus bringing NVIDIA's platform and technology deep into all walks of life. NVIDIA's partners and customers are spread across dozens of vertical industries.
For example, VR technology has a wide range of applications and can be used in entertainment, industrial design, architectural design, medical imaging, scientific computing and other fields. NVIDIA has built platforms called DesignWorksVR and GameWorksVR, which provide APIs and SDKs to more than 250 companies involved in terminal applications such as video games, entertainment and professional graphics, and integrate VR technology into the existing systems of these companies through a cooperative development model, so that end users can use VR technology through these systems.
3. Sales Model
The sales network and vertical service market are mutually beneficial.
Achieve win-win situation for all parties
NVIDIA's sales strategy is to enter the terminal market with the help of a partner network. NVIDIA's end customers include cloud computing companies, Internet companies, thousands of enterprises in various industries, and tens of thousands of startups. Since sales coverage does not allow NVIDIA to contact every healthcare company, insurance company, or retail company, its partner network comes into play. They are able to provide NVIDIA products and architectures to enterprise customers around the world with a large sales force and distributor network.
NVIDIA's partner network includes global, regional and specialty cloud service providers, OEMs, original equipment designers, system integrators, independent software vendors, add-on board manufacturers, distributors, automakers and other ecosystem participants, connecting hundreds of thousands of IT sales professionals through the network.
NVIDIA's vertical market service platform and sales network work together to create a win-win situation for NVIDIA, sales partners and end customers. Taking the data center business as an example, NVIDIA hosts its AI infrastructure with the world's leading cloud service providers, which then open it to enterprises and startups from all walks of life. For NVIDIA, cloud service providers provide their products to enterprises from all walks of life, rapidly expand the scope and scale of product use, and occupy the AI service market. For cloud service providers, this model enriches their AI service content and expands the scope of services. More importantly, NVIDIA's vertical market service platform directly supports tens of thousands of AI startups, which can attract new customers to use cloud services and bring direct users and economic benefits to cloud service providers. For customers from all walks of life, they do not have the ability to independently build an AI knowledge system. Through the "NVIDIA+Cloud" approach, they have the most advanced AI infrastructure and an excellent AI infrastructure management team, and can easily obtain all the expertise from infrastructure to AI models, greatly reducing the threshold for AI use.
Information sources
NVIDIA quarterly earnings conference minutes from 2014 to 2023
Author: Wang Qing
Email: wangqing36@126.com
END
????Semiconductor boutique public account recommendation????
▲Click on the business card above to follow
Focus on more original content in the semiconductor field
▲Click on the business card above to follow
Focus on the trends and developments of the global semiconductor industry
*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.
Today is the 3926th content shared by "Semiconductor Industry Observer" for you, welcome to follow.
Recommended Reading
★ Important report on EUV lithography machine released by the United States
Silicon carbide "surge": catching up, involution, and substitution
★ Chip giants all want to “kill” engineers!
Apple , playing with advanced packaging
★ Continental Group, developing 7nm chips
★
Zhang Zhongmou's latest interview: China will find a way to fight back
"The first vertical media in semiconductor industry"
Real-time professional original depth
Public account ID: icbank
If you like our content, please click "Reading" to share it with your friends.