Shenzhen Backs In-Memory Computing with New AI Action Plan

SHENZHEN, February 12, 2026 — The Shenzhen Municipal Bureau of Industry and Information Technology has issued the “Artificial Intelligence+ Advanced Manufacturing Action Plan (2026–2027)”, marking a strategic shift in the city’s AI and semiconductor policy. For the first time, the plan officially supports emerging chip architectures such as in-memory computing and compute-in-memory (CIM) processors, targeting AI applications in smartphones, intelligent robots, and autonomous vehicles. It emphasizes the development of high-performance SoC control chips and promotes domestic alternatives in core AI components.

Memory Technology Moves from “Supporting” to “Core”

Traditional von Neumann architectures separate the processor from memory, creating data movement bottlenecks that limit performance and raise power consumption. In-memory computing embeds computation directly within the memory array, dramatically reducing latency and energy use. By enshrining this approach in a municipal action plan, Shenzhen signals that memory-centric innovation is now central to the city’s AI chip strategy—no longer a peripheral enabler but a frontline focus.

Policy and Market Converge as Mass Production Window Opens

The same day the policy was released, SMIC disclosed that memory chips are in short supply and prices are rising. Industry sources say the clear policy signal is accelerating the shift from lab‑based in‑memory computing prototypes to engineering‑grade mass production. In the new energy vehicle sector, the plan explicitly backs domestic substitution of 14nm and smaller autonomous driving AI chips and cockpit SoC chips, opening a clear adoption path for high‑reliability embedded memory solutions.

From “Replacement Dividend” to “Architecture Dividend”

Until recently, the domestic chip industry’s growth relied largely on import substitution. The new Shenzhen policy shifts the focus toward architectural differentiation, elevating in-memory computing alongside mainstream AI controllers and prioritizing engineering readiness and real‑world deployment in high‑value scenarios. The goal is no longer just “making it work,” but “making it work at scale.”

Trillion‑Yuan Automotive Market Drives Transformation

The Action Plan highlights the trillion‑yuan new energy vehicle market, noting that L3+ autonomous driving demands data throughput that traditional memory hierarchies cannot sustain. Adoption of in‑memory and near‑memory computing in vehicles may now come faster than expected. As memory devices evolve from passive “warehouses” of data to active “factories” for computation, Shenzhen’s policy could mark the moment China’s semiconductor industry begins transitioning from an architecture follower to an architecture definer.