Hot Chips has always been one of the industry's best venues for getting beneath product-level specifications and into the architectures driving the next generation of computing. This year's conference so far has highlighted a broader shift in AI-accelerated architectures, where memory capacity and bandwidth, data movement, power, interconnects and rack-level integration are becoming critical design constraints alongside compute performance.

NVIDIA, AMD and Intel detailed new CPU, GPU and system architectures at the Stanford University campus venue, while Micron discussed how memory architectures need to evolve to keep pace with AI. Meanwhile, recent Qualcomm disclosures around High Bandwidth Compute (HBC) and its new Oryon CPU with FlexCache offer another perspective on many of the same challenges.