It pretty much seemed like a done deal when NVIDIA CEO Jensen Huang signed “Jensen approved” on Samsung's HBM3E modules at GTC 2024. A year later, Samsung has yet to be formally inducted into NVIDIA's HBM3E supply chain, as yield, heat management, and other issues have so far kept orders from at bay.

Huang has praised Samsung on many occasions, never doubting the company's ability to deliver what's asked of it. He did that again during GTC 2025, but fell short of providing any concrete information about Samsung finally getting HBM orders from NVIDIA.

Meanwhile, SK Hynix stole the show at GTC 2025 by showcasing a prototype of its 12-layer HBM4 modules that will be utilized in NVIDIA's next generation of AI accelerators. It has already taken the bulk of NVIDIA's HBM3E orders and looks set to meet the company's requirement for its upcoming chips.

Huang was asked about Samsung during at press conference at the event. He believes that Samsung will play an “important role” in the supply chain, but didn't say anything about orders. When directly asked whether Samsung's HBM3E modules will be used for the Blackwell Ultra, Huang didn't provide a clear answer.

He pivoted instead to an explanation of Samsung's ability to integrate the base die of the HBM with application-specific IC and memory, which is the expertise that's necessary for the fabrication of cutting-edge AI chips.

Huang's answers have led to further speculation about Samsung being able to win orders from NVIDIA, orders that it desperately needs, as its semiconductor division is in a dire need of a turnaround. Even the conglomerate's heir Jay Y. Lee has said that the company finds itself in a “do or die” situation.

At a shareholders meeting this week, top executives of the company promised that there won't be a repeat of what happened with HBM3E last year, and that Samsung will push to enter mass production of the next-generation HBM4 chips ahead of schedule this year.