partnerScore 80/100Add to thesis
Weaver (OmniConnect) gearbox solution for inference workloads targets $2,000–$3,000 revenue per GPU; Positron's 2TB LPDDR memory system requires many Weaver chips, indicating rapid ramp potential starting FY2028
Bill Brennan· Credo Technology· AI· 2026-06-01· about Credo Technology (Weaver), Positron (customer reference) (CRDO)
“let's use our first customer, Pozitron, as an example. They are redefining memory attachment and bandwidth... they have announced their first product with a total memory LPDDR size of 2 TB... If 2 TB are deployed that requires many, many Weaver chips... the revenue contribution per GPU can be between $2,000 and $3 thousand.”
Why it matters
Memory bandwidth is emerging as a critical bottleneck in inference GPUs. Weaver's gearbox architecture and Positron's 2TB system suggest a new architectural paradigm that could drive substantial interconnect revenue per unit.
Investment implication
Investors should monitor Positron's ramp and adoption by other inference ASIC makers (Groq, SambaNova, etc.). If successful, memory-bandwidth-optimized interconnect solutions could capture significant revenue. Suppliers of high-speed memory interfaces and packaging materials could see incremental demand.