Samsung to Cut HBM3E Prices to Amid AI Boom

According to Korean media outlet ZDNet, after Samsung's semiconductor business saw its Q2 profits plummet by nearly 94% year-over-year, the company is now aggressively cutting production costs for its HBM3E memory in hopes of securing adoption by major AI players—especially NVIDIA, whose GPUs currently rely primarily on SK Hynix's HBM supply.

Samsung's goal is clear: make its HBM3E cheaper and more accessible than any other option on the market. As AI computing becomes increasingly dependent on high-bandwidth memory, this pricing strategy could give Samsung a critical edge.

The company's memory division posted an 11% quarter-on-quarter increase in Q2 sales, reaching 21.2 trillion KRW (around $15.2 billion), largely fueled by expanded shipments of HBM3E and high-density DDR5 for servers. Samsung is also seeing NAND inventory clear faster as server SSD demand picks up.

Looking ahead, Samsung plans to scale production of 128GB DDR5, 24Gb GDDR7, and 8th-gen V-NAND in the second half of the year—key components for next-gen AI server deployments.

In a major boost to its foundry business, Samsung recently landed a $16.5 billion partnership with Tesla. Under this deal, the Korean tech giant will manufacture Tesla's next-gen AI6 chips at its Texas fab, with the contract extending through 2033. This long-term deal brings much-needed volume and stability to Samsung's foundry unit, especially as it navigates fierce competition with TSMC and rising geopolitical uncertainties.

However, recent developments in trade policy have added fresh headwinds. Just this week, U.S. President Donald Trump introduced a 15% tariff on South Korean goods, down from 25%, clouding Samsung's recovery outlook for the rest of 2025.

Samsung is walking a tightrope—banking on AI to lift its semiconductor business while contending with volatile global trade and trying to claw back share in the high-end memory market. If it can pull off low-cost, high-volume HBM3E production, the momentum may shift in its favor. Reports suggest that NVIDIA is already testing Samsung's HBM3E chips, though concerns remain over thermal performance and energy efficiency. Winning NVIDIA's trust would be a major milestone.

As Meta, Microsoft, and Amazon continue building out their in-house AI chips, memory makers like Samsung are now racing not just to prove performance—but to deliver real value.