Will AMD's New AI Chip Stack Up Against Nvidia?
On October 10, 2024, $Advanced Micro Devices (AMD.US)$ launched its latest AI GPU accelerator, the Instinct MI325X, in competition with Nvidia's Blackwell.
CEO Lisa Su stated that the new AI chip surpasses $NVIDIA (NVDA.US)$'s similar products in inference performance and features a new type of memory that significantly speeds up AI computing. Analysts describe the MI325X as a "mid-cycle upgrade." While it retains the same basic design and CDMA 3 GPU architecture as last year's MI300X, it is the "first AI GPU equipped with 256GB of HBM3e memory." AMD plans to introduce the MI355X chip in the second half of 2025, featuring 288GB of HBM3e memory and a CDNA 4 architecture, with an expected 35% boost in inference performance.
However, the competition for AI chip remains intense, particularly in the data center GPU sector. With its H100 series and strong CUDA programming ecosystem, Nvidia has attracted numerous developers and cloud service providers, solidifying its position as an industry leader. Although AMD's Instinct series AI chips are gradually improving, the company faces significant challenges in capturing Nvidia's market share. Data shows that Nvidia's data center GPU revenue is expected to reach $110.36 billion in 2024, compared to AMD's mere $12.83 billion, highlighting the stark difference. This gap is in revenue and the technology ecosystem surrounding AI chips, where Nvidia's CUDA programming language has become the industry standard, creating a strong lock-in effect. In contrast, AMD's ROCm programming ecosystem still appears relatively weak, even though it continues to be updated and enhanced.
While AMD's Instinct MI325X chip outperforms Nvidia's H200 in certain AI tasks, market analysts generally agree that this is not enough to challenge Nvidia's market position, especially when dealing with major clients like Meta, Microsoft, and Google, where Nvidia products remain the preferred choice. John Vinh, an analyst at KeyBanc, believes that despite AMD's significant lag behind Nvidia, its development trajectory shows positive trends. He expects the company to ship 500,000 units of the Instinct MI300X AI accelerator this year, further solidifying its position in the AI market.
AMD's stock price fell by more than 4% on announcement day, while Nvidia's stock rose by 1%. This indicates that investors expressed disappointment with AMD's new product announcement, particularly due to the need for more news regarding new clients, which failed to provide a clear direction for the company's future market expansion.
AMD's announcement did not include any new major clients; its existing customer base includes Meta, Microsoft, and OpenAI. Collaborating with these tech giants indicates that AMD is gradually expanding the application of its AI chips. However, analysts believe that Nvidia's dominance in cloud computing and AI development makes it difficult for AMD to increase its market share significantly in the short term. Some analysts are skeptical about whether AMD's MI series chips can effectively replace Nvidia's offerings. Despite companies like Meta and AWS utilizing AMD's AI chips, Nvidia remains their primary supplier. Unless AMD can demonstrate significant market growth, it will face formidable challenges in challenging Nvidia's position.
AMD plans to maintain its competitiveness by accelerating product updates, with plans to launch higher-performance AI accelerator chips in the coming years, including the MI350 in 2025 and the MI400 in 2026. KeyBanc analysts expect AMD to deliver 500,000 units of the Instinct MI300X chips this year, supporting its further growth in the AI market. Although AMD's current market share in AI is still relatively small, the increasing demand for AI and the expansion of market size provide opportunities for AMD to achieve greater breakthroughs in the coming years.
With the rapid growth of generative AI, the global demand for AI chips is surging. Morgan Stanley analyst Joe Moore points out that the construction of artificial intelligence data centers is still in its early stages, with every indication from management suggesting that we are in the early part of a long-term investment cycle in AI. Lisa Su anticipates that the market for AI accelerators will grow from $400 billion in 2027 to $500 billion by 2028. The application of AI technology is continuously expanding into various sectors, including autonomous driving, medical imaging, and smart city construction, all of which demand higher computational performance from AI accelerators.
For AMD, the rapid growth in AI demand presents a significant opportunity. CEO Lisa Su predicts that the explosive growth of the AI chip market will offer immense opportunities for AMD and many chip manufacturers. However, it also poses challenges, especially with the tight supply chain for AI chips, particularly high-end GPUs, where current demand far exceeds supply capacity. Despite the strong performance of AMD's new chips, the company still needs to enhance its production capabilities and market recognition. If AMD cannot quickly ramp up production to meet market demand, it may struggle to compete effectively with Nvidia.
"Investor expectations for clear details on the MI325 and subsequent products drove AMD's stock price up slightly over the past week, but we remain skeptical about AMD's ability to narrow the gap with Nvidia unless it can demonstrate a clear growth driver for market share; otherwise, the stock may retest its annual lows."
Besides, the Instinct MI325X features excellent inference performance, especially on Meta's Llama 3.1 AI model, which outperforms the Nvidia H200. However, competition in the market relies not just on hardware performance but also on a broader technology ecosystem, developer support, and chip scalability. Nvidia's long-standing presence in AI has allowed it to win in hardware by deeply integrating software and services with developers and customers.
Moreover, a significant challenge for AMD is strengthening its software support ecosystem. While the ROCm programming framework continuously improves, it still faces widespread application challenges. The weakness of the software ecosystem means that many developers are hesitant to switch to AMD, which is a major barrier to expanding its market share. How well AMD can continue to focus on market share, customer collaboration, and technology ecosystems will determine its ability to compete with Nvidia in the long run.
Analysts point out that more than simply improving hardware is required to achieve significant success in the AI market. The efficiency of running AI workloads depends on chip performance and the software stack, memory bandwidth, and network infrastructure. Even if AMD's chips theoretically outperform Nvidia's in peak performance, Nvidia's ecosystem provides a better overall solution in practical applications.
Some Wall Street analysts hold a cautiously optimistic view of AMD's prospects. According to KeyBanc's estimates, AMD's target stock price is $190.56, representing an 11% upside from its current price. Despite fierce market competition and Nvidia's dominant position, AMD's ongoing innovation and market strategies position it well for accelerated growth in the coming years.
Source: MarketWatch
Disclaimer: Moomoo Technologies Inc. is providing this content for information and educational use only.
Read more
Comment
Sign in to post a comment