TeRAM exits stealth with $37M seed to break the AI memory wall
What's the deal? TeRAMDealroom has a profile for this one. Try Dealroom → emerged from stealth on September 15, 2026, with $37 million in seed financing to tackle the memory wall facing AI deployment. The Los Altos, California company is developing custom 3D SRAM for frontier AI applications, starting in data centers.
Who's backing it? Primary Venture Partners, B Capital, HyperionDealroom has a profile for this one. Try Dealroom →, and SemiAnalysis CapitalDealroom has a profile for this one. Try Dealroom → co-led the round, which closed on September 10. Alumni Ventures and Lightscape PartnersDealroom has a profile for this one. Try Dealroom → also joined.
What's the endgame? TeRAM wants to integrate 3D SRAM directly onto AI compute chips, delivering higher memory bandwidth at lower power. The funding will let it validate the technology and build out its team, with initial customer production targeted for 2029.
Why now? As AI models scale beyond two trillion parameters, existing memory technology struggles to keep up on capacity, throughput, and power. Since mid-2025, agentic inference has overtaken training as the main driver of AI compute, shifting the bottleneck from HBM capacity to memory bandwidth.
Generating each output token requires a pass through a model's trillions of parameters, making bandwidth the constraint — whether the model sits in on-chip SRAM or off-chip HBM.
Who's building it? TeRAM's founders bring more than 80 years of combined experience in semiconductor memory. The team has worked on AI compute system-on-chip projects with several hyperscalers and AI infrastructure companies.
"Memory is the constraint on AI progress, and incremental improvements will not be enough," said Brian Schechter, a partner at Primary Venture Partners and the lead investor. He described founder Charlie Cheng as "a rare founder who connects AI architecture, memory physics, and semiconductor manufacturing."
"Developing a new memory technology is an ambitious undertaking," said Cheng, chief executive officer of TeRAM. "We are fortunate to have investors who back our vision with both capital and hands-on support."
The signal: The round reflects a bet that memory, not raw compute, is becoming the gating factor for AI. As inference workloads grow, startups attacking the bandwidth problem at the chip level are drawing fresh investor interest.
Read more: uk.finance.yahoo.com
Image credit: TeRAM
