Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

tech.yahoo+1tech.yahootech.yahooQualcomm has secured Samsung Electronics and SK hynix as active collaborators on its High Bandwidth Compute technology, a near-memory architecture designed to break through AI inference bottlenecks that the company says existing memory solutions cannot solve.
The chipmaker laid out its broader AI infrastructure roadmap at the Deutsche Bank 2026 Technology Conference on Wednesday, August 26, with Durga Malladi, executive vice president of technology planning and edge data center, detailing a multi-year plan spanning custom silicon, server CPUs and software tools aimed at challenging Nvidia's dominance in data-center AI.
Qualcomm's HBC places compute directly adjacent to stacked DRAM, targeting the decode stage of AI inference where memory bandwidth — not raw compute — limits performance. Malladi said the architecture can deliver a sixfold improvement in tokens per second per watt relative to high-bandwidth memory and reach "hundreds of terabytes per second" of bandwidth, compared with roughly 21 to 22 terabytes per second for HBM today.tech.yahoo+1
Silicon for the first generation has returned to Qualcomm's lab and is "coming along very nicely," Malladi said, with commercial availability targeted for 2027 and a second generation planned for 2028. He said Korean memory vendors, rather than defending HBM against the new approach, asked "how can we work with you much closer on this" — a response he described as validating the technology's direction.investing
According to The Lec, Samsung Electronics and SK hynix are actively collaborating with Qualcomm on HBC's commercialization, providing the supply partnerships needed to bring the stacked-memory architecture to market.thelec
Qualcomm also confirmed its Dragonfly C1000 server CPU is on track for commercial availability in the second half of 2028, with Meta as its first announced customer. Malladi said the chip is designed to compete against projected 2028-era offerings rather than current products, and that hyperscalers have called its performance-per-watt claims "too good to be true".investing
The company is simultaneously pursuing RISC-V as a parallel path to Arm , characterizing the open instruction-set architecture as moving from academic interest to pre-commercial engagement. Qualcomm acquired Antenna Microsystems to bolster its RISC-V data-center efforts.tech.yahoo
To challenge Nvidia's CUDA lock-in, Qualcomm highlighted its acquisition of Modular, whose Mojo programming language and MAX compiler are designed to run AI workloads across multiple hardware vendors. Malladi said the stack can match or exceed native performance by up to 50 percent in certain cases, and that AMD Advanced Micro Devices, Inc. has publicly committed to evaluating Modular on its own hardware.investing
A Hugging Face partnership rounds out the developer strategy, aiming for single-click model deployment onto Qualcomm platforms. Malladi framed the combined effort as a "cloud-to-edge AI platform" spanning data centers down to smartphones and automotive systems — though he acknowledged that many hyperscalers want silicon validation before making firm commitments.investing