TrendForce News operates independently from our research team, curating key semiconductor and tech updates to support timely, informed decisions.
In a major shake-up in the AI chip market, AMD has signed a multi-year deal with OpenAI to supply next-generation AI accelerators, potentially generating tens of billions in annual revenue and giving OpenAI the option to acquire up to 10% of Team Red. The impact goes beyond the two companies: according to SeDaily, Samsung Electronics is set to supply the crucial HBM4 memory, marking a significant shift in the global AI semiconductor landscape that had long centered around NVIDIA and SK hynix.
Samsung and AMD have a history of collaboration dating back to the MI350X and MI355X chips. According to Commercial Times and SeDaily, both chips—built on TSMC’s N3P process—feature 12-high HBM3e from Samsung and Micron.
Under the latest agreement between AMD and OpenAI, the 6-gigawatt partnership will power the ChatGPT maker’s next-generation AI infrastructure across multiple generations of Instinct GPUs. The first 1-gigawatt deployment, featuring Instinct MI450 GPUs, is set to roll out in the second half of 2026, AMD notes.
Citing well-known TF International Securities analyst Ming-Chi Kuo, SeDaily reports that Samsung is set to be the main HBM4 supplier for AMD’s MI450, giving the company a chance to challenge SK hynix, which has long dominated the HBM market through its strong NVIDIA partnerships and leadership in HBM3 and HBM3e.
Industry sources cited by the report project a fierce new rivalry between the NVIDIA–SK hynix alliance and the AMD–Samsung camp, adding that SK hynix’s competition with Samsung over technology and pricing will intensify.
SeDaily further points out that the partnership’s ripple effects extend well beyond the companies involved. AMD is spearheading an open-standard alternative, Ultra Accelerate Link (UALink), alongside Google, Microsoft, and others to challenge NVIDIA’s proprietary NVLink. With OpenAI now joining AMD’s camp, the UALink ecosystem could expand rapidly—eroding NVIDIA’s technical exclusivity and paving the way for broader industry participation, the report adds.
Samsung Further Strengthens Ties with OpenAI
Before teaming up with AMD, OpenAI had already struck preliminary deals with Samsung and SK hynix to secure memory for its massive Stargate data center project last week. According to Tom’s Hardware and Reuters, OpenAI’s demand could skyrocket to as many as 900,000 DRAM wafers per month — a staggering figure that could account for nearly 40% of the world’s total DRAM output.
As per Maeil Business Newspaper, Samsung is expected to supply not only HBM but also graphics DRAM (GDDR), high-capacity SSDs, and LPDDR5X-PIM — low-power DRAM with in-memory computing — for AI applications in OpenAI’s Stargate project. The partnership could give Samsung a significant lift, bolstered further by its growing HBM alliance with AMD.
Read more
(Photo credit: Samsung)