Gary Marcus: The Rise of Neurosymbolic AI Will Bring CPUs Back into the Hardware Mix
Gary Marcus · rss · 2026-08-08
Gary Marcus argues that the evolution of AI architectures is shifting the underlying hardware requirements. During the era dominated by pure neural networks—roughly from 2012 to mid-2023—commercial AI was almost exclusively driven by GPUs, which excel at parallel matrix arithmetic.
However, since mid-2023, frontier AI companies have quietly begun integrating technologies like symbolic code interpreters. Neurosymbolic AI, which combines neural networks with classic logical computation, inherently requires both CPUs and GPUs. Marcus suggests this marks a shift away from a purely GPU-driven era, with CPUs regaining importance in AI workloads.
More from AGI Musings
- AI-Generated Virus Event Sparks Debate: Author Calls It Most Alarming AI Event — GarrisonLovely · 2026-08-08
- Chollet: We Have AGI Capabilities, but Lag Humans by 3-5 Orders of Magnitude in Efficiency — mark_k · 2026-08-08
- Black Hat's Scariest Talk: AI Agents Are More Dangerous Than Tigers — JeffLadish · 2026-08-08
- KPMG: Nearly Half of Executives Delay AI Agent Deployments as Costs Exceed Benefits — Polymarket · 2026-08-08
- AI Sandbox Escapes: Genuine Security Crisis or Marketing Stunt? — alex_verem · 2026-08-08
- AI Product Forms Converging: Chat and Coding Boundaries Will Disappear — nbaschez · 2026-08-08