Publication details

Advisor Voice

Sources
Date
Tags
  • Articles

Unveiling the hidden winners in Nvidia’s AI powerhouse

Author: Qiao Ma

Nvidia’s dominance in the artificial intelligence (AI) processing unit (GPU) market isn’t a solo act. While Nvidia has grabbed the headlines with its powerful H100 chip, a network of suppliers is silently reaping the rewards according to Qiao Ma, partner & lead portfolio manager of the Munro Global Growth Small & Mid Cap Fund, Munro Partners.

Nvidia’s H100 GPU is the most powerful GPU chip on the market and is designed specifically for AI applications. The H100 contains 80 billion transistors, which is six times faster than its predecessor, Nvidia’s A100 chip.[1]

According to Ms Ma, Nvidia’s H100 GPU is like a multi-lane highway, boasting thousands of ‘lanes’ called cores. This translates to lightning-fast AI tasks like image recognition, language translation, video editing, etc. This phenomenal performance has created a surge in demand, benefitting not just Nvidia, but the entire supply chain.

“The H100 features specialised lanes dedicated to specific types of data, like images or text. This specialised approach boosts efficiency, allowing the H100 to handle complex AI tasks such as self-driving cars, medical diagnosis systems and other AI-powered applications.”

While Nvidia has grabbed the spotlight, numerous other companies are key to their production and are benefitting from the GPU maker’s strong sales. From specialised power delivery systems being made by Monolithic Power, to intricate cooling solutions inside the data centres made by Vertiv, each element contributes to the functionality of Nvidia’s chips. Most of these suppliers have seen the AI-related portion of their revenues and profits soar over the past few quarters, and Ma thinks it is just the start.

“It can help to think of Nvidia’s GPU as a dedicated express lane for data, enabling faster information flow compared to traditional memory architecture. This speed boost is crucial for handling the massive data demands of AI tasks. These memory chips are made by SK Hynix,” said Ms Ma.

“Importantly, the H100 features innovative chip stacking. Imagine multiple microprocessors layered on top of each other, working together. This miniaturisation allows for more computer processing power within a compact space, which is crucial for maximising performance and efficiency of GPUs. The continued development of advanced chip manufacturing will rely heavily on specialised technologies like BESI’s hybrid bonding and ASMI’s atomic layer deposition,” she said.

Nvidia’s H100 also requires intricate connections between hundreds of components, demanding microscopic precision. “Companies like Onto Innovation and Camtek play a crucial role in this stage, using specialised metrology techniques to ensure everything is perfectly aligned and free of defects. These companies might not be household names, but their contributions are vital to the H100’s success,” she said.

“Munro anticipates the aggressive pace of interest rises will subside in 2024, and in which case, we believe it will be a good environment for growth companies such as these semiconductor suppliers, as well as Nvidia. In such an environment, companies will go back to following their earnings, which is what the Munro team spend most of our time researching and investigating.”

Munro Partners is capturing growth in the artificial intelligence hardware and semiconductor supply chain investment opportunity through its Munro Global Growth Small & Mid Cap Fund, with ASM International and Onto Innovation as two of the Fund’s top five holdings as at 31 March 2024.