Three weeks after Nvidia shocked the tech industry by announcing an unprecedented revenue leap, Wall Street has been scouting for other chip companies that could benefit from the latest AI boom.
But as the investigation progressed, the chasm between Nvidia and the rest of the chip industry only widened.
In one of the most anticipated attempts to catch up to Nvidia, rival AMD showed off a new AI chip this week dubbed MI300X. The chip contains the core of NVIDIA’s success: a GPU, a product originally designed for video games, and a more general-purpose CPU and internal memory that feeds both processors.
This design reflects the chipmaker’s attempt to bundle different technologies in the search for the most efficient way to handle the large amounts of data required to train and apply the large-scale models used in generative AI. increase.
AMD claims the new chip’s superior performance, saying it will outperform Nvidia’s flagship H100 in some respects. But it can’t show potential customers considering this chip, and it’s the product’s ability to handle AI inference (pre-trained AI models ) was only emphasized. It also said it wouldn’t begin ramping up production of the new chip until the final quarter of the year.
NVIDIA’s H100 will have been on the market for 18 months by the time AMD’s new chip hits general availability in the first half of next year, according to Bernstein analyst Stacey Rasgon. It is said that there will be AMD is “very late. They may get the dregs [of the AI market] — but that’s probably enough,” he said, to justify Wall Street’s recent frenzy for the stock.
Moor Insights & Strategy analyst Patrick Moorhead added that “NVIDIA is free and clear in this round” of the chip wars that have erupted around AI.
Wall Street has named a number of semiconductor companies that could benefit from generative AI. The combined stock market value of AMD, Broadcom and Marvell soared by $99 billion (20%) in two days after Nvidia released a staggering sales forecast last month. However, their AI-related sales are not expected to come from a market dominated by Nvidia.
Broadcom, for example, stands to benefit from the growing demand for its data communications products and its collaboration with Google on the design of its in-house data center chips known as TPUs. Earlier this month, Broadcom predicted that its AI-related business would account for about a quarter of its revenue by 2024, up from just 10% last year.
But processors, which are used to train and apply AI models at scale, are generating the most surge in demand and the most excitement in the stock market. As AMD’s new chips dominate Wall Street, NVIDIA’s stock has climbed again past the $1 trillion level it first reached two weeks ago.
AMD Chief Executive Officer Lisa Su said data centers are a key focus of investment and “there is no question that AI will be a major driver of silicon consumption in the foreseeable future.” rice field. She predicts that the market for AI accelerators (GPUs and other specialized chips designed to speed up the data-intensive operations required to train and run them) will grow from her $30 billion this year to 2027. predicted she would surge to over $150 billion.
As it struggles to compete with Nvidia on cutting-edge AI chips, companies such as AMD and Intel are looking to the evolution of the generative AI market to boost demand for other types of processors. Large-scale models like OpenAI’s GPT-4 dominated the early days of the technology, but the recent explosion in the use of smaller, more specialized models has increased sales of lower-performance chips. They argue that they could.
According to Intel vice president Kavitha Prasad, many customers looking to use corporate data to train models are willing to share their information rather than risk handing it over to companies that provide AI models at scale. He hopes to keep it close to his home. In addition to all the computing work to prepare the data to feed into the training, she said there will be a lot of work on her Intel-made CPUs and her AI accelerators.
But chip makers are having a hard time predicting how the market will develop, as demand for data centers is changing rapidly with the proliferation of services such as ChatGPT. Razgon said CPU sales could even decline in the next few years as data center customers pour money into AI accelerators.
Rivals looking to add a bite to Nvidia’s burgeoning AI business face an equally big challenge with their software. Nvidia’s chips are widely used in AI and other applications because GPUs originally designed for video games can be easily programmed for other tasks using Cuda software.
In an attempt to attract more developers to its AI chips, AMD this week highlighted its efforts to work with the widely used AI framework PyTorch. But it still has a long way to go before it can match the many software libraries and applications already being developed for Cuda, Razgon said. “It will take a decade” for rivals to match Nvidia’s software, he said, and during that time he said Nvidia would move quickly to extend its lead.
“No one wants an industry with one dominant player,” says Moorhead. But for now, the fastest-growing market for chips that can handle generative AI is Nvidia’s.
