According to the South Korean media report by ZDNet, after the semiconductor business recorded a year-on-year dive of about 94% in its second-quarter earnings, Samsung is actively working to reduce the production cost of its HBM3e for AI applications in order to lower the selling price and attract Nvidia's adoption. Currently, Nvidia's AI GPUs are mainly supplied with HBM from SK Hynix.
Reporters say Samsung's strategy is simple: make its HBM3E memory more affordable and accessible than any other supplier, which will become an indispensable part of future AI computing.
Sales at Samsung’s memory division rose 11% in the second quarter compared to the first, reaching 21.2 trillion won ($15.2 billion), thanks in part to the expansion of HBM3E and server high-density DDR5. With server SSD sales on the rise, the clearance of NAND inventory is also accelerating. In the second half of this year, Samsung plans to boost production of 128GB DDR5, 24Gb GDDR7, and 8th-generation V-NAND, especially for AI server deployments.
Tesla recently confirmed a $16.5 billion partnership with Samsung, which will allow the South Korean company to produce next-generation AI6 chips at its Texas foundry, with the contract lasting until 2033. This agreement further strengthens Samsung's long-term prospects for its foundry business. This can inject much-needed scale and stability into Samsung's foundry operations, especially as the company faces fierce competition from TSMC and geopolitical pressures.
Just this week, President Donald Trump of the United States imposed a 15 percent tariff on Korean goods (dropped from 25 percent), casting new uncertainty over Samsung's second half business recovery. Samsung is walking a tightrope—bets on artificial intelligence to pull it out of the doldrums while weathering turbulent trade dynamics and trying to regain market share in the competitive high-end memory business.
If Samsung can achieve cheaper, high-yield HBM3E production, it could change the dynamic. reports that Nvidia is testing Samsung's HBM3E, but remains skeptical of the cooling and power efficiency, winning back Nvidia would be a inflection point. As Meta, Microsoft and Amazon are all expanding their in-house AI silicon, memory suppliers like Samsung are in a new race not only to prove capability but also to prove value.
(Reprinted from China Grid https://news.eccn.com)