How AI Will Reshape Computer Systems by 2035: A Jeffersonian Dinner in San Francisco About Our 10,000x Future
The Computing Research Association’s Industry Salon gathered 20 AI and systems leaders in San Francisco on April 16, 2026 to envision computer systems in 2035. Participants, including Jeff Dean, Dave Patterson and academic pioneers, projected a 10,000‑fold increase in global AI inference driven by 50× gains in algorithms, 50× hardware specialization and a modest 4× data‑center expansion. They highlighted new efficiency metrics such as intelligence‑per‑Watt and debated whether future abstractions must remain human‑interpretable. The discussion also covered clean‑energy data‑center power, societal risks of AI, and the need for education and policy frameworks.
Fourth Data Prefetching Championship: Part I
The fourth Data Prefetching Championship (DPC‑4), held with HPCA 2026, showcased a range of innovative prefetching algorithms evaluated against a baseline of Berti at L1D and Pythia at L2 under tight storage budgets. Keynote speakers from Huawei and Google emphasized...
Beyond Qubits: A Systems View of Hybrid CV-DV Quantum Computing
At ASPLOS 2026 a tutorial introduced hybrid continuous‑discrete‑variable (CV‑DV) quantum computing, which treats qubits and oscillator modes as a unified computational resource. The session covered the physical foundations, new instruction set architectures, and the compilation stack that translates high‑level algorithms...
Computer Architecture’s AlphaZero Moment Is Here
The paper argues that computer architecture has shifted from idea scarcity to evaluation scarcity, driven by large‑language models and autonomous pipelines. The open‑source Gauntlet system reproduced authors' solutions in 48 % of 85 recent ISCA/HPCA papers and proposed alternatives in another...
Spilling the Neural Tea: A Journey Down the Side-Channel
Recent research highlights the growing use of side‑channel attacks to reverse‑engineer deep neural networks, revealing model architectures and, in limited cases, weight information. Physical side channels on edge devices and micro‑architectural channels in cloud environments have demonstrated success in extracting...
To Sparsify or To Quantize: A Hardware Architecture View
Hardware architects face a trade‑off between sparsity and quantization for compute‑bound generative AI models. Unstructured sparsity offers maximal pruning but forces complex routing and poor SIMD utilization, prompting a shift toward structured patterns like N:M and block‑sparse attention. Quantization reduces...