may 3, 2026
2 links from the engineering internet.
theregister.comai
inference era opens a new lane for ai chip startups
as ai workloads shift from training to serving, inference is heterogeneous enough for specialized chips to win distinct niches. groq's sram-heavy lpu architecture (now under nvidia) and lumai's optical inference accelerator are among those making their move as the workload mix diversifies.
theregister.cominfra
uk pac holds up bank of england migration as rare public-sector tech win
parliament's public accounts committee published a report calling the bank of england's rtgs renewal a model for large-scale public sector tech, contrasting it with the emergency services network running 12 years late and the national savings & investments overhaul already £1.3b over budget.