NVIDIA Dominates AI Chip Market Amid Rising Competition
In the competitive field of artificial intelligence (AI) supercomputers, NVIDIA Co. (NASDAQ: NVDA) stands out with an impressive 94% market share of AI chips, commonly known as GPUs. While Marvell Technology Inc. (NASDAQ: MRVL) and others are vying for attention in the ASICs segment, fewer rivals can challenge NVIDIA’s stronghold in the broader computer and technology sector.
AMD Takes Aim at NVIDIA’s Lead
Applied Micro Devices Inc. (NASDAQ: AMD) has significantly improved its position in the market by developing GPUs aimed at PC gaming and bitcoin mining. AMD is optimistic that its upcoming MI325X AI accelerator chips can effectively compete with NVIDIA’s Blackwell GPUs, particularly in data center applications.
Data centers have been relying heavily on NVIDIA GPUs, resulting in substantial demand and putting new orders on a year-long waitlist. AMD hopes to attract more clients looking for immediate AI chip solutions without extended delays. Reports suggest that companies like Microsoft and Meta Platforms have already acquired some AI GPUs, alongside OpenAI, indicating a shift in purchasing habits.
NVIDIA’s Blackwell Chips: Fully Booked for a Year
NVIDIA’s Hopper 100 AI chips demonstrate robust potential; the company claims that just two of its HGX supercomputers—priced at $500,000—can replace 1,000 standard CPU servers that cost $10 million. Its next-generation Blackwell GPUs are said to be four times more powerful than the Hopper H100 AI chips. As of October 2024, all Blackwell GPUs (B100 and B200) are sold out for the upcoming year, with shipments slated to begin at the end of 2024. These systems feature eight Blackwell GPUs and an impressive memory bandwidth of up to 72 petaFLOPS for training and 144 petaFLOPS for inference.
Notably, large customers like Meta Platforms Inc. (NASDAQ: META), Alphabet Inc. (NASDAQ: GOOGL), Amazon.com Inc. (NASDAQ: AMZN), Microsoft Co. (NASDAQ: MSFT), and Oracle Co. (NYSE: ORCL) have already secured their allocations of Blackwell GPUs even before their public release.
Taiwan Semiconductor: The Backbone of AI Chips
Taiwan Semiconductor Manufacturing Co. Ltd. (NYSE: TSM) holds the title of the largest semiconductor manufacturer globally. Its advanced fabrication technologies are essential for developing AI GPUs, which depend on smaller manufacturing processes for improved performance and efficiency. Currently, top-tier AI GPUs are produced using 7nm, 5nm, and upcoming 3nm processes.
Taiwan Semi’s innovative chip-on-wafer-substrate (CoWoS) technology allows multiple components to be integrated onto a single substrate, enhancing overall chip performance. However, NVIDIA’s extensive demand has been stretching production capabilities, necessitating closer collaboration between Taiwan Semi and NVIDIA to resolve bottlenecks.
Capacity Constraints and AMD’s Challenges
AMD anticipates starting production on its MI325X AI GPU by the end of 2024. The company aims to introduce new AI chips annually, with the MI350 planned for 2025 and the MI400 for 2026. Nonetheless, consistent capacity constraints remain, particularly with mounting demand in the AI sector.
NVIDIA’s substantial orders may take precedence at Taiwan Semi, which generates a significant portion of its revenue—approximately 11%—from NVIDIA’s business. This relationship has contributed to Taiwan Semi’s impressive 36% year-over-year revenue growth, totaling $23.5 billion in the third quarter of 2024, while also underscoring NVIDIA’s position as the second-largest customer.
Why AMD Faces Uphill Battle Against NVIDIA
Considering the urgency of NVIDIA’s AI chip orders, it’s likely they will be given priority due to existing supply constraints. As NVIDIA’s Blackwell GPUs are booked solid for the next year, this situation could hinder AMD’s plans to ramp up the production of its MI325X AI GPUs.
Additionally, NVIDIA’s ecosystem is fortified with its CUDA programming language, which has become a standard for AI developers. This dependency traps many developers within NVIDIA’s framework, making it challenging for AMD to gain traction, regardless of the technical merits of its chips. Taiwan Semiconductor is expanding its capacity and forecasts capital expenditures of around $30 billion in 2025, yet NVIDIA’s ordering strength may limit AMD’s potential breakthroughs in the near term.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
