TeRAM Raises $37M for AI Memory
TeRAM has emerged from stealth with $37 million to develop 3D SRAM aimed at reducing the memory bottlenecks limiting AI inference performance.
The AI hardware race is moving beyond accelerators into the memory systems that feed them. TeRAM has emerged from stealth with $37 million to tackle what the company calls the AI memory wall.
What happened
The equity financing was co-led by Primary Venture Partners, B Capital, Hyperion and SemiAnalysis Capital.
TeRAM is developing custom 3D SRAM designed to place more high-speed memory close to AI compute. The architecture targets workloads where moving model data between memory and processors consumes significant time and power.
Why it matters
Modern AI models frequently become limited not by arithmetic performance but by how quickly weights and intermediate data can reach the processor.
That bottleneck becomes particularly important during inference, where cost and energy efficiency determine whether models can be deployed at large scale.
The bigger picture
AI infrastructure is becoming a full-system optimisation problem. GPUs remain central, but startups are now attracting serious capital for memory, networking, photonics, cooling and power.
TeRAM's approach sits squarely in that shift. Commercial success will depend on manufacturing, integration and performance in real systems, but the market increasingly recognises that faster accelerators alone cannot solve every scaling constraint.