Key highlights from Nvidia Corporation (NVDA) Q2 2024 Earnings Concall
Management Update:
- [00:02:40] NVDA’s Data Center Compute revenue nearly tripled YonY, driven by demand for the HGX platform from cloud service providers and large consumer internet companies. Also, Networking revenue almost doubled YonY, driven by demand for the InfiniBand networking platform.
- [00:05:00] NVDA does not anticipate that additional export restrictions on its Data Center GPUs would have an immediate material impact to its financial results.
Q&A Highlights:
- [00:24:43] Matt Ramsay of TD Cowen asked about the emerging application of large model inference and how the market is segmented between small and large model inference, and how NVDA’s product portfolio is positioned for this. Jensen Huang CEO replied that LLMs are powerful language models that can learn and understand human language. They can be distilled into smaller models that are specialized for specific tasks.
- [00:28:18] Vivek Arya at BofA Securities enquired about the sustainability of demand for generative AI in the hyperscaler market and the confidence in its growth over the next 1-2 years. Colette Kress CFO replied that the company expects to increase supply of HGX and other products over next quarters and fiscal year.
- [00:31:20] Stacy Rasgon from Bernstein Research enquired how much of the Data Center growth in the quarter was driven by pricing, content, or units. Colette Kress CFO said that the HGX systems were the main driver of NVDA’s Data Center revenue growth in the most recent quarter. These systems include both the Hopper and Ampere architectures. DGX systems were also a significant contributor to growth, but to a lesser extent.
- [00:34:27] Mark Lipacis of Jefferies enquired how has NVDA’s software ecosystem evolved over time, and how does it contribute to the value of the NVDA platform. Jensen Huang CEO answered that NVDA’s AI Enterprise software stack is used for end-to-end machine learning and is flexible, versatile, and high-performance. NVDA has a large installed base and is present in the cloud and enterprise, allowing it to sustain a complex stack of software and hardware and introduce new products regularly.
- [00:40:44] Atif Malik with Citi queried about CoWoS less L40S and its potential impact on supply tightness, as well as its incremental profitability or GM contribution. Jensen Huang CEO answered that the L40S is a GPU designed for fine-tuning pre-trained models for hyperscalers and enterprises. It is easy to install and has a wide range of software support. It is different from the H100, which is designed for large-scale language models and processing large amounts of data.
- [00:43:36] Joe Moore from Morgan Stanley asked how much unfulfilled demand is there for NVDA products, and when NVDA expects supply to meet demand? Jensen Huang CEO answered that the demand for accelerated computing is growing rapidly as companies look to improve their throughput, energy efficiency, and cost efficiency. NVDA is well-positioned to meet this demand with its broad portfolio of products and solutions.
- [00:46:00] Toshiya Hari of Goldman Sachs asked for a breakdown for Data Center business between CSPs, consumer Internet, and enterprise customers. Jensen Huang CEO replied that CSPs are the largest customer segment for NVDA’s Data Center business, followed by consumer Internet companies and enterprise customers. NVDA is confident that there will be enough applications and use cases for its customers to generate a reasonable return on their investments in accelerated computing and generative AI.
- [00:52:00] Timothy Arcuri with UBS asked about the attach rate of NVDA’s Networking Solutions to its compute products, and whether this can be used to prioritize allocation of GPUs. Jensen Huang CEO answered that NVDA does not use the attach rate of its Networking Solutions to its compute products to prioritize allocation of GPUs. NVDA offers a variety of networking solutions for different data center environments, including InfiniBand and Ethernet.
- [00:55:01] Ben Reitzes from Melius enquired about the reception and momentum of DGX Cloud. DGX Cloud is a strategic initiative for NVDA to engage and work closely with the AI ecosystem around the world. Jensen Huang CEO said that DGX Cloud’s strategy is to enable a close partnership between NVDA and the world’s CSPs, work closely with CSPs to improve the performance of hyperscale clouds, and support NVDA’s own large infrastructure needs. DGX Cloud has multiple use cases and drivers and has been a great success, with CSPs, developers, and NVDA’s own engineers all clamoring for more.
- [00:55:11] Ben Reitzes from Melius also asked about NVDA’s software business, including its run rate, materiality, and impact on margins. Colette Kress CFO said that NVDA’s software business is growing rapidly, with hundreds of millions of dollars in annual revenue. NVDA is looking to include NVDA AI Enterprise with its products and in CSP marketplaces, which will further grow the software business.
