Nvidia stock vs. custom AI chips: Should investors worry?

Nvidia stock has had a great year in 2024, nearly tripling to about $135 per share. Business has boomed, driven by the surge in demand for graphics processing units (GPUs), which have emerged as the backbone of the era of generative artificial intelligence. That said, investors have been turning their attention to another area of ​​the semiconductor market over the past month: application-specific integrated circuits (ASICs), which could play a larger role in AI computing. is.

This comes after two major ASIC players, Broadcom and Marvell Technology, reported surging demand for ASICs from large cloud customers in the most recent quarter. Do Custom AI Chips Make Marvell Stock a Buy?For example, Broadcom’s custom AI chip and networking processor sales will increase 220% in 2024 from the $3.8 billion in sales the company generated from AI silicon in 2023. This brought the total to $12.2 billion. Sure, Nvidia’s revenue is impressive, estimated at about $129 billion this fiscal year, but its growth rate is slowing. So could ASICs threaten Nvidia’s AI dominance as the market matures?Separately, if you want a smoother rise than individual stocks, you might want to consider the fact that ASICs have outperformed the S&P and are up 91% since inception. Consider high-quality portfolios that have returned more than %.

ASICS vs GPU

ASICs have been around for over 50 years. But they are gaining new interest in the AI ​​era. GPUs from companies like Nvidia are extremely versatile and can be programmed for AI and other tasks, whereas ASICs are custom-designed semiconductors built to perform specific tasks and are generally It offers certain advantages compared to conventional processors. These chips offer several advantages over GPUs for AI by focusing on targeted functionality. For example, these specialized chips can be more cost-effective than GPUs designed for a broader range of applications.

ASICs also consume less power, making them ideal for data centers looking to reduce power costs, a critical cost for operating large-scale AI systems. Because ASICs are purpose-built, they can deliver higher performance for specialized tasks than general-purpose GPUs from Nvidia and AMD. These chips could be well-suited to large cloud computing providers, given that they operate at scales that justify the design and development costs of ASICs. For example, Broadcom, seen as the biggest beneficiary of the potential pivot to ASICs, recently announced that three of its hyperscaler customers are deploying clusters of 1 million custom chips across a single network. announced that it plans to build a

How changes in the AI ​​space can benefit ASICs

Over the past couple of years, companies have poured enormous resources into building AI models. Currently, training these large models is a one-off task that requires significant computing power, and Nvidia’s GPUs are considered the fastest and most efficient for these tasks. Because of this, we are receiving the maximum benefit from this. However, the AI ​​landscape may be changing. We expect the incremental performance improvement to decrease as the model grows with respect to some parameters. Apart from this, much of the internet’s high-quality data is already running through large-scale language models, so the availability of high-quality data to train models can become a bottleneck. Considering this, the AI ​​training phase, which has been brought forward significantly, may end. The fundamental economics of the end market for GPU chips and the broader AI ecosystem are weak, and most of Nvidia’s customers may not yet generate meaningful returns on their investments. See how Nvidia stock drops to $65.

As shareholders ultimately seek higher returns, more customers may turn to ASICs to reduce initial and operational costs. The long-term focus of AI is on inference, where trained models are used in real-world applications. This stage could open the door to alternative AI processors that are less computationally intensive and less powerful. ASIC chips customized for inference may also be a good choice for these tasks. There is also historical precedent for this. For example, in the crypto industry, Bitcoin miners initially used GPUs for mining, but some large players are now using GPUs to increase (Number of guesses), we switched to ASICs as we scaled up. ) and overall cost-effectiveness.

What this means for Nvidia

While NVDA stock has experienced strong growth in recent years, the 30-stock Trefis High Quality Portfolio has delivered better returns with less risk compared to the benchmark S&P 500 index over the past four years. Masu. It’s not been a roller coaster ride, as evidenced by the performance metrics of our corporate portfolio. So what’s in store for NVIDIA stock?

Indeed, given Nvidia’s head start in the AI ​​market and its deeply ingrained CUDA software stack and development tools, which will result in higher switching costs for customers, it is likely that Nvidia’s business will benefit from these new waves of processors. We do not expect it to be replaced. Moreover, if we see significant changes in the market, Nvidia may also expand its presence in this space. However, NVIDIA’s premium valuation may not fully account for potential risks and slower growth. We value Nvidia stock at approximately $93 per share, approximately 32% below the current market price. Check out our analysis of Nvidia’s valuation: high or low.

Invest with a portfolio that outperforms the Trefis Market

View all price quotes for Trefis