AMD’s efforts to chase Nvidia have fallen flat. Can its new chips help?

AMD unveiled its Instinct MI350 Series GPUs, boasting up to four times more AI compute and 35 times more AI inferencing power than its previous generation, aiming to gain ground against Nvidia in the AI chip market. Meta Platforms and Oracle Cloud Infrastructure are among the first customers, with Meta planning to use the MI350 accelerators for its Llama AI models and Oracle offering up to 131,072 of the GPUs for AI model building. Despite a $5 billion revenue from AI chips in 2024, AMD's AI revenue has not fully met earlier expectations, and its overall data-center revenue remains significantly lower than Nvidia's.

Analysis

Advanced Micro Devices (AMD) has introduced its Instinct MI350 Series accelerators, including the MI350X and MI355X GPUs, in an effort to bolster its position in the competitive AI chip market currently dominated by Nvidia. These new chips reportedly offer significant performance improvements, with four times more AI compute and 35 times more AI inferencing power compared to AMD's previous generation, and the MI355X is touted to generate up to 40% more tokens per dollar than competitors, highlighting a price-performance advantage. The company also achieved a five-year goal for energy efficiency and set new targets, a critical consideration given the increasing scrutiny on AI data center energy consumption. Key tech giants Meta Platforms and Oracle Cloud Infrastructure have been announced as early adopters; Meta will utilize the MI350 for its Llama AI models and continue collaboration on future AMD AI roadmaps, including the next-generation MI400 series, while Oracle plans to offer up to 131,072 MI355X GPUs. Despite these advancements and securing over $5 billion in Instinct AI chip revenue for 2024, AMD's progress in the AI sector has been viewed with mixed success, as investor expectations had previously been set higher, according to Oppenheimer & Co. analyst Rick Schafer. AMD's overall data-center revenue for 2024 reached $12.6 billion, substantially less than Nvidia's $115.2 billion for its full fiscal year 2025. Bank of America analyst Vivek Arya suggests AMD could capture a "credible 3%-4% share" of a potential $400 billion AI accelerator market. AMD's stock has reflected this challenging landscape, down approximately 25% over the past year, though up about 20% in the last three months, and fell 2.2% on the day of the announcement. A recent $10 billion partnership with Saudi Arabian AI firm Humain is seen by Rosenblatt Securities as a positive development that could help mitigate US-China trade uncertainties.

Key Decisions for Investors

• Investors should closely monitor the adoption rate and revenue impact of the new Instinct MI350 Series, as its success is crucial for AMD to gain tangible market share against Nvidia, especially given that past AI revenue figures have sometimes fallen short of high investor expectations.
• Consider the valuation implications of AMD potentially capturing a 3-4% share of the projected $400 billion AI accelerator market, weighing this against the company's current data-center revenue being nine times less than Nvidia's and its recent stock underperformance.
• Evaluate the strategic partnerships with Meta, Oracle, and Humain as key indicators of market validation and potential growth drivers, but remain cautious about execution risk and the intense competitive pressure from Nvidia and custom chip developers before significantly altering positions.
• Monitor AMD's progress towards its enhanced energy efficiency goals for AI hardware, as this could become an increasingly important differentiator and selling point in an energy-conscious market.