Reports 1Q on Wednesday, May 22 After Market Close (AMC). Nvidia (NVDA) made the case for strong demand on its Q4 earnings call driven by its Data Center business. Aggregate Capex in Q4 from the big four Hyperscalers grew 30% yr/yr and reinforced confidence about NVDA's outlook as the company reports Q1 earnings this week. The data center business contributed 83% of sales last quarter and META (META - NC) and Micosoft (MSFT-NC) have been major customers with Amazon (AMZN - NC) expected to come on strong as H200 ramps.

Hardware

The NVDA GB200 GPU is NVIDIA's latest flagship GPU, and is designed for significant performance and efficiency improvements. It is used in the NVDA DGX SuperPOD with DGX GB200 Systems, capable of delivering up to 40 exaflops of performance. The combination of the GB200 GPU and the DGX SuperPOD with DGX GB200 Systems provides performance gains in terms of AI computing power.
We understand NVDA's next generation R100 could see production starts at TSM late 2025 according to our checks. The R100 architecture is expected to showcase improvements in power consumption.

Software

NVDA NIM Microservices is a collection of cloud-native microservices that make it easier to deploy and manage AI applications on the cloud. NVDA's cuLitho is a new programming language for GPUs, designed for ease of use relative to traditional GPU programming languages, such as CUDA. NIM Microservices and cuLitho are both designed to make it easier for developers to build and deploy AI applications. NIM Microservices simplifies the deployment and management of AI applications on the cloud, while cuLitho makes it easier to write code for GPU's.

Capacity Update

According to our field checks Taiwan Semiconductor Manufacturing Company, Limited's (TSM- NC) 4nm utilization rate declined late last year due to reduced demand for iPhone processors. While Apple's iPhone 15 production may be further reduced this year, we believe the 4nm node has ample capacity to meet NVDA's needs, especially as major chip suppliers like Apple (AAPL-NC), Qualcomm (QCOM-Hold), and MediaTek transition their products to 3nm.
TSM's 3nm process, particularly the N3E variant, boasts high yield rates and substantial capacity, which we believe can sufficiently meet client demand. The upcoming N3P variant is further optimized for high-performance AI computing, catering to the needs of companies like NVDA. We anticipate a material increase in both 4nm and 3nm capacity from the second half of 2024 onwards. This should alleviate production and supply constraints impacting the H100, along with providing ample capacity for the recently launched H200. In turn this should enable NVDA to maintain its profitability in the high-end AI computing market.
NVDA is also expected to leverage TSM's N3P for its next-generation GPUs. While N3P's yield rates may not match those of N3E, we believe it should adequately meet NVDA's requirements. TSM's primary bottleneck currently lies in CoWoS advanced packaging, but the company has committed to significantly step up CoWoS capacity needs.

Industry Update

Industry Update

Semiconductors & Emerging Technologies

Nvidia (NVDA- Buy) Q1 Earnings Approach

关于

集团投资组合

​联系我们

联系我们