$NVDA Thoughts Post 3Q Results
NVDA 0.00%↑ continues to defy gravity reporting record revenue of $18.1 billion, up 34% sequentially and more than 200% year-on-year, exceeding the $17.5 billion buy side estimates. Data Center revenue was $14.5 billion, up 41% sequentially and 279% year-on-year, driven by the NVIDIA HGX platform and InfiniBand networking. About half of the data center revenue came from CSPs and the other half from consumer internet companies and enterprise ramping to deploy Gen AI. Gaming revenue was $2.86 billion, up 15% sequentially and more than 80% year-on-year. ProViz revenue reached $416 million, up 10% sequentially and 108% year-on-year. Automotive revenue increased to $261 million, up 3% sequentially and 4% year-on-year. Even more impressive than the topline growth is that 3Q operating margin came in at 57% compared to 10% last year and is at 50% through the first 9-months of the year compared to 14% in ‘22.
Jensen's 20-year journey in building an AI platform is paying off, with AI use cases becoming embedded in everyday life. InfiniBand's 5x growth over the year, driving networking revenue past $10 billion run-rate, underscores NVDA's shift from mere chip provider to a comprehensive AI model-building platform. Jensen clearly understands that the current boom is driven by LLM training, where the AI model learns from data, but the more sustainable opportunity will be driven by Inference, the AI model using what it learned to make predictions or decisions on new data. NVDA's TensorRT LLM software is key here, optimizing AI models for NVDA's hardware. It's akin to building a super-efficient brain (H200) and providing it with a guide (TensorRT LLM) for better understanding and processing human language. This will revolutionize AI's speed and intelligence while halving Inferencing costs.
NVIDIA AI offers the best inference performance and versatility and thus, the lower power and cost of ownership. We are also driving a fast cost reduction curve. With the release of NVIDIA TensorRT LLM, we now achieved more than 2x the inference performance or half the cost of inferencing LLMs by NVIDIA GPUs.
The Grace Hopper based CPU architecture is ramping quickly and will play an integral part in Inferencing. The Grace CPU is based on ARM architecture, which is known for its energy efficiency and is widely used in mobile devices. NVDAs adoption of ARM architecture for its CPU signifies a strategic move towards creating more power-efficient and versatile computing solutions which are ideal for AI use cases that need to reference a lot of data or remember long contexts for better and more accurate interactions. The GH 200 chip combines the Grace CPU with the Hopper GPU architecture, creating a powerful integrated solution for computing tasks. This combination is expected to provide exceptional data processing capabilities, crucial for handling complex AI algorithms and large-scale computing tasks efficiently. The integration aims to optimize both energy efficiency and computational performance, addressing two critical aspects of modern data center operations. The GH 200 is likely to significantly enhance the capabilities in AI model training and inference, allowing for more complex and sophisticated AI models to be developed and deployed. The chip's advanced processing power would be beneficial for a wide range of scientific applications, including simulations in physics, climate modeling, and biomedical research. Blending the Grace CPU's efficiency with the Hopper GPU's power will significantly impact the realms of AI, deep learning, and HPC, marking a pivotal advancement in NVDA's pursuit of high-efficiency, high-performance computing platform.
And you could also have applications or generative models where the context length is very high. You basically stored an entire book into its system memory before you ask the questions. And so the context length could be quite large. This way, the generative models have the ability to still be able to naturally interact with you on one hand; on the other hand, be able to refer to factual data, proprietary data or domain specific data, your data and be contextually relevant and reduce hallucination. And so the -- that particular use case, for example, is really quite fantastic for Grace Hopper. It also serves the customers that really care to have a different CPU than x86. Maybe it's European supercomputing centers or European companies who would like to build up their own ARM ecosystem and like to build up a stack or CSPs that have decided that they would like to pivot to ARM because their own custom CPUs are based on ARM. There are a variety of different reasons that drives the success of Grace Hopper, but we're off to just an extraordinary start. This is a home run product.
DGX cloud is another cog in the platform approach wheel. NVDA is becoming the de facto AI training platform where companies can bring their proprietary data to create custom AI models utilizing NVDA LLM training expertise and infrastructure. Companies like SAP 0.00%↑ , NOW 0.00%↑ , SNOW 0.00%↑ , do not need large capex outlays to build proprietary AI models but can utilize NVDAs platform, which combines software and hardware, to train and deploy their models. NVDA has now created a recurring revenue by essentially licensing their infrastructure in order to allow companies to create and deploy models based on their data. As more enterprises enter the AI fray looking to leverage their internal data to create AI applications DXG cloud offerings will become a larger part of the NVDA story.
NVIDIA AI Enterprise is $4,500 per GP per year. That's our business model. Our business model is basically a license. Our customers then, with that basic license, can build their monetization model on top of. In a lot of ways, we're wholesale. They become retail. They could have a per -- they could have subscription license based. They could per instance or they could do per usage. There's a lot of different ways that they could take to create their own business model, but ours is basically like a software license, like an operating system. And so our business model is help you create your custom models. You run those custom models on NVIDIA AI Enterprise. And it's off to a great start. NVIDIA AI Enterprise is going to be a very large business for us.
NVDA HGX platform is designed specifically for AI and HPC workloads. This means it's tailored to handle complex calculations and data processing tasks that are essential in these fields. It combines NVIDIA's powerful GPUs with advanced networking technology (like NVLink and InfiniBand). This integration allows for rapid data transfer and efficient communication between GPUs, which is critical for high-speed data processing and AI model training. HGX has a modular design which allows data centers to scale their operations by adding more GPU and networking modules as needed, allowing for flexibility and scalability in deploying AI and HPC solutions. The platform is supported by comprehensive software stack, including CUDA, cuDNN, and other AI and HPC software libraries. This software ecosystem enables developers to optimize their applications for the HGX platform efficiently. The HGX platform is widely adopted by major cloud service providers and enterprises, which speaks to its reliability and performance capabilities. This wide adoption also means there's a large community and a wealth of knowledge and resources available. It’s clear that NVDA has become the AI platform and in order to get optimal performance out of NVDA GPUs companies will need to utilize the full stack offering.
Remember, people think that the GPU is a chip, but the HGX, H100, the Hopper HGX, has 35,000 parts. It weighs 70 pounds. Eight of the chips are Hopper. The other 35,000 are not. It is -- it has -- even its passive components are incredible, high-voltage parts, high-frequency parts, high-current parts. It is a supercomputer and therefore, the only way to test the supercomputer's with another supercomputer. Even the manufacturing of it is complicated. The testing of it is complicated. The shipping of it is complicated and installation is complicated. And so every aspect of our HGX supply chain is complicated.
NVDA's strategy has evolved from producing chips to providing a comprehensive platform (DGX Cloud) where businesses can come and build their own AI models using tools (like NEMO) and foundational AI models. Businesses bring their own data and expertise, and NVDA provides the technology, support, and infrastructure. Once the AI models are built, they can be deployed in a secure and optimized environment suitable for business use. This approach allows them to cater to a wide range of business needs while leveraging its expertise and AI optimized technology. Jensen has said “the more you buy, the more you save” by which he means that utilizing NVDA’s entire platform (TensorRT LLM and H200, InfiniBand, DGX Cloud) ,not just their chip, will make AI tasks like understanding and generating LLM faster and cheaper by a significant factor. NVDA made a decision long ago to ensure that all their technologies are compatible with each other and improvements in one area benefit their entire ecosystem.
And so the complexity includes, of course, all the technologies and segments and the pace. It includes the fact that we are architecturally compatible across every single one of those. It includes all of the domain specific libraries that we create. The reason why you -- every computer company without thinking can integrate NVIDIA into their road map and take it to market and the reason for that is because there's market demand for it. There's market demand in health care. There's market demand in manufacturing. There's market demand, and of course, in AI, in financial services and supercomputing and quantum computing. The list of markets and segments that we have domain-specific libraries is incredibly broad. And then finally, now we have an end-to-end solution for data centers. InfiniBand network -- InfiniBand networking, Ethernet networking, x86, ARM, just about every permutation combination of solutions, technology solutions and software stacks provided. And that translates to having the largest number of ecosystem software developers, the largest ecosystem of system makers, the largest and broadest distribution partnership network, and ultimately, the greatest reach. And that takes -- surely, that takes a lot of energy. But the thing that really holds it together, and this is a great decision that we made decades ago, which is everything is architecturally compatible. When you -- when we develop a domain-specific language that runs on one GPU, it runs on every GPU. When we optimize TensorRT for the cloud, we optimized it for enterprise. When we do something that brings in a new feature, a new library, a new feature or a new developer, they instantly get the benefit of all of our reach.
NVDA’s decision 20-years ago to create programable CUDA and GPUs and build out their install base has created a formidable and still somewhat underappreciated competitive advantage which allows them to use software to optimize hardware performance. Currently, they have no real competitors and while CSPs are building their own chip infrastructure and AMD is releasing its own GPUs, they are way behind NVDAs integrated approach, and given the pace of innovation it is hard to imagine anyone catching up within the next 5 years. NVDA should do $19-$20 in EPS next year and $16-$17 in FCFPS, putting the multiple at 25x EPS and 30x FCF.
So why is NVDA trading at such an undemanding multiple given its recent growth trajectory and place at the forefront of what could be one of the greatest transformations in computing? NVDA's modest valuation, despite its groundbreaking role in computing, stems from a few key concerns. The lack of backlog disclosures creates uncertainty about the sustainability of GPU demand. Is the current surge just pulling forward future demand, with a possible drop-off by 2025/26? This uncertainty clouds the long-term growth outlook. Then there's the China factor. NVDA's significant revenue from China ($4 billion this quarter) faces a potential 25% hit in Q4, a roughly $1 billion impact, due to export restrictions. While these restrictions are a short-term hurdle, I'm confident NVDA will innovate and adjust its chips to align with U.S. export requirements, minimizing long-term effects. The trickier issue is demand pull forward. Over-ordering by CSPs and LLM developers, notably in China, suggests a scramble for GPUs due to limited supply. Despite overordering occurring I thin this is the early stages of an AI arms race, indicating a sustained, albeit uneven, demand trajectory. In this context, NVDA's valuation seems conservative, hinting at market caution in the face of these uncertainties and the evolving landscape of AI and GPU demand.
And as I just said earlier, we have a substantial reserve of AI chips, which can help us keep improving ERNIE bot for the next year or two. "
-BIDU
The global realization of AI's importance is just dawning, with many countries and enterprises at the beginning of their AI infrastructure journey. The rise of AI co-pilots and bots will force companies to shift capex towards AI model development and deployment. NVDA's offerings aren't just chips; they encompass CPUs, advanced networking solutions (boasting a $10 billion run-rate), and software services (with a $1 billion run-rate). This integrated and expanding portfolio places NVDA in a prime spot to capture a large portion of the economics from the increasing adoption of AI technologies. Given the current market valuation, the downside appears limited. The market may be underestimating NVDA's growth durability, particularly as accelerated computing becomes the norm. NVDA's ability to consistently outperform market expectations suggests a stronger growth trajectory than currently priced in, potentially driving share prices higher as the company continues to exceed forecasts into 2024.