GSI Technology Stock Jumps 39% As Edge AI Bets Put GSIT Back In Focus

GSI Technology’s “Gemini-I” APU – a novel Associative Processing Unit that performs AI computations directly in memory – received independent validation from a Cornell University-led study unveiled on Oct 20 at the MICRO’25 conference ts2.tech. Researchers confirmed GSI’s chip can deliver performance on par with a high-end NVIDIA GPU for certain artificial intelligence tasks, while using roughly 98% less energy ts2.tech. In large-scale tests of retrieval-augmented generation, the Gemini-I APU achieved throughput comparable to an NVIDIA A6000 GPU but with a tiny fraction of the power draw ts2.tech. It also ran these AI tasks up to 5× faster than a standard CPU, dramatically cutting processing times ts2.tech.

On Oct 20, a Cornell University-led study confirmed that GSI’s Gemini‑I APU can deliver GPU-like AI performance using far less power investing.com economictimes.indiatimes.com. In benchmarks on large retrieval-augmented generation workloads, GSI’s chip achieved throughput comparable to an NVIDIA A6000 GPU while using ~98% less energy investing.com economictimes.indiatimes.com. The study also found the APU ran RAG tasks several times faster than standard CPUs investing.com ts2.tech. GSI CEO Lee-Lean Shu hailed the independent validation, saying “Cornell’s independent validation confirms what we’ve long believed – compute-in-memory has the potential to disrupt the $100 billion AI inference market,” and that the APU delivers “GPU-class performance at a fraction of the energy cost.” economictimes.indiatimes.com.