2020年2月15日 星期六

Nvidia Morningstar

Analyst Note | by Abhinav Davuluri Updated Feb 14, 2020

Nvidia reported fourth-quarter results ahead of management's guidance, as the firm's GPUs for artificial intelligence workloads enjoyed strong demand from major hyperscale and consumer Internet customers. We were pleased to see inventories return to a more normalized level following the cryptocurrency-related headwinds that plagued the firm in early 2019. Data center revenue grew considerably, as customers leverage both Nvidia's training and inference GPUs key AI applications such as natural language understanding, conversational AI, and deep recommendation engines. After rolling our valuation model forward and incorporating slightly stronger growth in data center sales for the current year, we are raising our fair value estimate to $160 per share from $145. Nevertheless, we view shares as overvalued as we think current levels imply Nvidia is the sole beneficiary of the burgeoning AI and self-driving trends.

Fourth-quarter sales grew 3% sequentially and 41% year over year to $3.1 billion. The sharp year-over-year spike can be attributed to an artificially deflated fourth quarter in fiscal 2019 (calendar 2018) due to the massive decline in gaming GPU sales during that period stemming from a decline in demand in GPUs for cryptocurrency mining. Gaming sales fell 10% sequentially to $1.5 billion due to seasonally lower notebook GPUs and Nintendo Switch chip sales, partially offset by growth in desktop GPUs. While we do not anticipate major share loss to AMD, we do expect a more competitive environment that should pressure Nvidia's ASPs going forward. Data center sales were $968 million, up 43% year over year and 33% sequentially. Both T4 inference and V100 training GPUs were shipped in record volumes, while T4 shipments were up 4 times year over year due to public cloud deployments. While major cloud players have all adopted the T4 in their data centers, we continue to expect a fragmented inference chip market, with FPGAs, ASICs, and even CPUs being prominent.

Business Strategy and Outlook | by Abhinav Davuluri Updated Nov 09, 2019

Nvidia is a leading designer of graphics processing units that enhance the visual experience on computing platforms. The firm's chips are used in a variety of end markets, including high-end PCs for gaming, data centers, and automotive infotainment systems. Enterprise customers use Nvidia's GPUs for professional visualization applications that require realistic rendering, including computer-aided design, video editing, and special effects. Nvidia has experienced initial success in focusing its GPUs in nascent markets such as artificial intelligence (deep learning) and self-driving vehicles. Hyperscale cloud vendors have leveraged GPUs in training neural networks for uses such as image and speech recognition.

The linchpin of Nvidia's current business is gaming. PC gaming enthusiasts generally purchase high-end discrete GPUs offered by the likes of Nvidia and AMD. Going forward, we expect the data center segment to drive most of the firm's growth, led by the explosive artificial intelligence phenomenon. This involves collecting large swaths of data followed by techniques that develop algorithms to produce conclusions in the same way as humans. As Moore's law-led CPU performance improvements have slowed, GPUs have become widespread in accelerating the training of AI models to perform a task. However, we think other solutions are more suitable for inferencing, which is the deployment of a trained model on new data. Today's basic variants of AI are consumer-oriented and include digital assistants, image recognition, and natural language processing.

The firm views the car as a "supercomputer on wheels." Although this segment currently contributes relatively little to the top line, we acknowledge the opportunity Nvidia has to grow its presence in cars beyond infotainment as drivers seek autonomous features in newer vehicles. Looking further, Nvidia's Drive PX platform is a deep learning tool for autonomous driving that is being used in research and development at more than 370 partners. Nonetheless, both the data center and automotive spaces are fraught with competition that could limit Nvidia's future growth.

Economic Moat | by Abhinav Davuluri Updated Nov 09, 2019

We believe Nvidia has a narrow economic moat stemming from its cost advantages and intangible assets related to the design of graphics processing units, or GPUs. The firm is the originator of and leader in discrete graphics, having captured the lion's share of the market from longtime rival AMD. We think the market has significant barriers to entry in the form of advanced intellectual property, as even chip leader Intel was unable to develop its own GPUs despite its vast resources, and ultimately needed to license IP from Nvidia to integrate GPUs into its PC chipsets. To stay at the cutting edge of GPU technology, Nvidia has a large R&D budget relative to AMD and smaller GPU suppliers that allows it to continuously innovate and fuel a virtuous cycle for its high-margin chips.

Nvidia's intangible assets originate with its popularization of GPUs in 1999, which could off-load graphics processing tasks from the CPU, thereby increasing the overall performance of the system. The firm has patents related to the hardware design of its GPUs in addition to the software and frameworks used to take advantage of GPUs in gaming, design, visualization, and other graphics-intensive applications. Additionally, the latest PC games typically require system software updates (drivers) that optimize the performance of GPUs. We note Nvidia tends to provide more reliable drivers for most games that allows gamers to take full advantage of its GPUs, while AMD is unable to match Nvidia in breadth and consistency of driver updates. Consequently, consumers have favored Nvidia's GPUs for gaming, with the firm boasting over 70% share in the discrete GPU market, with little resistance from AMD at the leading-edge. In turn, this has enabled economies of scale that allow Nvidia to invest in designing chips at the latest process node while offering regular driver updates and remain at the forefront of GPU technology.

While the market for discrete GPUs in PCs has continued to decline, as most PCs utilize integrated graphics chips from Intel, Nvidia has benefited from a resurgence for high-end GPUs driven by the growing enthusiast PC gaming space. In our view, AMD has been unable to design products capable of competing with Nvidia's GPUs at the high-end of the gaming spectrum. Consequently, Nvidia has gained share at the expense of AMD as gamers have moved from mainstream graphics cards to performance and enthusiast segments. We note these GPUs range from $150 at the low-end to over $1,000 for premium cards, with Nvidia's gaming gross margins in the high 50s. Although virtual reality is another trend that should benefit Nvidia's gaming GPUs, we think mobile VR applications will be more prominent relative to those on PC VR systems, at least in the near term.

Unlike gaming GPUs, which are dependent on the secularly declining PC market, Nvidia has taken steps to leverage its GPU prowess into other markets such as automotive and data center that represent a meaningful and sustainable growth opportunity. GPUs are being used to accelerate computation workloads with the goal of training AI systems to drive cars and perform medical diagnoses. We note these are computationally intensive endeavors that are more achievable with CPUs and GPUs working in tandem versus CPUs in isolation.

Internet behemoths such as Google, Facebook, Amazon, and Microsoft have found GPUs to be adept at accelerating cloud workloads that use deep learning techniques to achieve speech recognition (Siri, Google Now, Alexa, Cortana), photo recognition (identifying faces in pictures on Facebook, videos of cats on YouTube), and recommendation engines (Netflix, Amazon). To train a computer to recognize spoken words or images, it must be exposed to massive amounts of data with the goal of educating itself. Inference involves taking what the model learned during the training process and putting it into real world applications to make decisions (that is after reviewing 10,000 cat pictures during training, is this next picture a cat?).

These examples are not very efficient to run on server CPUs (predominantly Intel's Xeons) alone, as general-purpose CPUs consist of a few cores that are good at performing a wide array of tasks in a sequential manner. The training process is ideal for GPUs that have massively parallel architecture consisting of thousands of smaller cores designed for handling multiple tasks simultaneously. Nvidia has a first-mover advantage in the accelerator market, as it looks to drive AI adoption in both the cloud and on the road.

Within automotive, Nvidia currently has a presence in the infotainment systems of approximately 8 million vehicles through its Tegra processors. While increasingly complex digital cockpit computers will become the norm, we note this is a highly competitive market, with Qualcomm, Intel, among others also offering competitive infotainment solutions, and we do not see any competitive advantage from Nvidia that warrants a moat just yet. However, with its Drive PX self-driving system, Nvidia hopes to carve a dominant position in the self-driving space. Should the firm's autonomous platform win the lion's share of self-driving business, we think Nvidia would strengthen its moat via superior intangible assets and switching costs. We view this opportunity as in the early innings, and while more than 370 OEMs have tested Drive PX in R&D settings, Intel (with the 2017 acquisition of Mobileye) represents a formidable opponent that will challenge Nvidia, in our view.

Fair Value and Profit Drivers | by Abhinav Davuluri Updated Feb 14, 2020

We are raising our fair value estimate to $160 per share from $145 per share. Our fair value estimate assumes a forward adjusted price/earnings ratio of 25 times. We do not believe the market is accounting for the competitive forces that we expect to challenge Nvidia. We project revenue will increase at a 15% compound annual growth rate through fiscal 2025 as the firm continues to diversify its revenue sources to areas of strong potential. Fiscal 2020 was a challenging year for the firm, as gaming revenue fell due to a cryptocurrency mining-related hangover and excess channel inventories, but we anticipate high-teens growth in fiscal 2021.

We think the data center segment is poised to rise at a CAGR of 25%, accounting for 40% of total revenue in fiscal 2025. We expect the firm to dominate the training portion of deep learning, but we don't believe the inference market will be as lopsided in favor of Nvidia's GPUs as the current stock price suggests. Gaming should continue to be a major source of revenue, though we think recent growth rates will be difficult to replicate due to saturation and lengthening replacement cycles of gaming GPUs, greater competition, and softer cryptocurrency mining-related GPU sales.

In automotive, most of Nvidia's current sales are infotainment-related. Its Drive PX autonomous driving platform is still in the early innings. By 2025, the firm expects the AI opportunity in automotive to be $30 billion with roughly 40 million vehicles on the road with capabilities ranging from basic level 2 to fully autonomous level 5. Ultimately, we think Nvidia will capture a healthy portion of this opportunity, culminating in a 25% CAGR in automotive revenue through fiscal 2025. Nonetheless, we also believe Intel-Mobileye will factor prominently into the self-driving equation, as it boasts core competencies vital to the endeavor.

We expect gross margins to hover around 60%. Gaming accounts for over half of total sales and has a gross margin at the corporate average. The Tegra chip, used in automotive infotainment systems, is relatively lower. In contrast, the enterprise and data center units have high gross margins, ranging from 65% to 70%. However, we also foresee margin deterioration due to competition leading to long-term gross margins of 60%. Nvidia must invest heavily in R&D to maintain its competitive edge in GPUs. Thus, we model long-term R&D as a percentage of sales at 24%, implying operating margins in the high 20s.

Risk and Uncertainty | by Abhinav Davuluri Updated Nov 09, 2019

Consumer spending in the PC space has undergone a significant structural shift over the past decade with the proliferation of mobile devices that serve as many users' de facto computers. This has pressured sales in desktops and laptops. Nvidia's discrete GPUs are also challenged by Intel's chips that feature both the CPU and GPU. These combo chips are more cost-efficient but still lack high-end graphics capability. We note there is still strong demand from gaming enthusiasts, who are willing to purchase high-end GPUs from the likes of Nvidia.

We remain concerned Nvidia generates the majority of its sales from gaming. The firm has benefited from strong PC gaming momentum in recent years, as gamers shift from consoles to PC gaming. However, many of the most popular games are competitive multiplayer online games (esports) that require low-end discrete GPUs for latency reasons versus high-end GPUs for cutting-edge graphics. The firm is also expected to benefit from Virtual Reality, however, a shift to mobile gaming VR over PC VR could curb these opportunities, as Nvidia's GPUs aren't formidable in smartphones (similar to Intel's CPUs in smartphones).

Adjacent markets (AI, automotive) are still in the early stages, and though Nvidia has a first-mover advantage in both, its lead may not last if superior alternatives arise (other forms of acceleration for AI or other self-driving platforms). Also, the rate of disruption tends to be quicker in these markets that are very performance-sensitive. We note GPUs were designed to do one thing very well: render graphics for realistic images, games, videos, and so on Leveraging GPUs in deep learning applications among other areas mostly occurred due to lack of better alternatives. As alternatives arise (via current competition or startups), Nvidia's recent explosive growth will be difficult to sustain, in our view. Ultimately, the risky nature of Nvidia's nongaming GPU segments leads to our very high uncertainty rating.

Stewardship | by Abhinav Davuluri Updated Nov 15, 2019

We believe management has demonstrated Exemplary stewardship of shareholder capital. CEO Jen-Hsun Huang cofounded Nvidia in 1993 after stints at LSI Logic and AMD. Colette Kress became CFO in September 2013, having previously worked with Cisco. Management compensation appears reasonable compared with industry peers.

Nvidia's management team has shown a willingness to invest in new opportunities in the past several years outside the firm's core PC graphics processor business. As a result, Nvidia has become a key player in the artificial intelligence accelerator market with its GPUs for AI training and inference workloads. The firm has also sought to drive the push toward autonomous driving with its Drive PX platform.

The firm has periodically made acquisitions in the past, though the deals tend to be smaller relative to ones made by competitors such as Intel. One notable acquisition was the $367 million purchase of Icera in 2011, a baseband processor firm, to complement Nvidia's foray into mobile devices. In May 2015, Nvidia wound down its Icera modem operations primarily because of its shift in strategy to focus on high-growth opportunities such as gaming, automotive, and AI acceleration instead of the cutthroat integrated application processor and modem markets for smartphones. We view this move as shrewd because it shows that management is willing to adapt when a particular venture isn't performing as intended.

In March 2019, Nvidia announced it will acquire Israeli-based Mellanox Technologies for $6.9 billion or $125 per share in cash. Mellanox sells networking products that focus on efficient data transfer in data centers via its InfiniBand and Ethernet technologies for interconnects. We note there will be no initial revenue or cost synergies as the GPU titan intends to maintain all of Mellanox's existing investments. We think this makes sense, as the deal rationale is initially to bolster Nvidia's share of data center spend to potentially increase its switching costs. Nvidia's DGX integrated system for Artificial Intelligence, or AI, utilizes InfiniBand technology while the two firms collectively power over half of the world's Top 500 supercomputers with Nvidia GPUs and Mellanox interconnects.

Management initiated a quarterly dividend in the fourth quarter of fiscal 2013 to return excess cash to shareholders, and it currently has a stock-buyback program. The firm returns cash to shareholders through ongoing quarterly cash dividends ($0.16 per share) and share repurchases.

1 則留言:

  1. If you're trying hard to lose kilograms then you need to jump on this brand new custom keto diet.

    To create this keto diet, licensed nutritionists, personal trainers, and professional chefs joined together to develop keto meal plans that are useful, decent, price-efficient, and enjoyable.

    From their grand opening in early 2019, thousands of people have already completely transformed their figure and well-being with the benefits a good keto diet can provide.

    Speaking of benefits: clicking this link, you'll discover eight scientifically-confirmed ones given by the keto diet.

    回覆刪除