Mon Oct 20 14:00:00 UTC 2025: Here’s a news article summarizing the text, aimed at a tech and finance audience:

GSI Technology’s New AI Chip Claims Game-Changing Energy Efficiency, Stock Soars

SAN JOSE, CA – October 26, 2025 – GSI Technology (NASDAQ:GSIT) is making waves in the artificial intelligence hardware market with its Gemini-I APU (Associative Processing Unit), claiming GPU-level performance at a fraction of the energy cost. A recent study by Cornell University benchmarked the Gemini-I against industry standards, revealing potentially disruptive results.

The Cornell study, running real-world AI workloads involving large-scale Retrieval Augmented Generation (RAG), found that GSI’s in-memory chip achieved comparable throughput to a high-end NVIDIA A6000 GPU. However, the real headline is power consumption. The Gemini-I consumed “over 98% lower energy” than the GPU on large datasets, using only 1-2% of the GPU’s energy draw, according to the study.

“Compute-in-memory has the potential to disrupt the $100 billion AI inference market,” said Lee-Lean Shu, CEO of GSI Technology, in a press release. The Gemini-I also drastically outperformed traditional CPUs, shortening retrieval processing time by up to 80%.

This efficiency stems from GSI’s compute-in-memory (CIM) architecture, which co-locates processing within memory arrays, eliminating the data transfer bottlenecks common in traditional CPU/GPU designs. The company is banking on this advantage to carve out a niche in edge AI applications, particularly within the defense and aerospace sectors.

The announcement sent GSI’s stock price soaring, with GSIT up nearly 200% mid-session on October 20th. However, financial analysts are urging caution. While the technology is promising, GSI’s revenue is only approximately $22.1 million annually, and the company faces significant net losses. Technical forecasting sites predict a potential price correction. The stock is also considered to have “very high volatility.”

The company’s focus on high-margin, power-constrained markets like defense and aerospace may offer a path to profitability. GSI’s partners already include defense contractors working on satellite and drone imaging, aligning with a growing edge AI chip market projected to reach $56-57 billion by 2030. The company is already developing its Gemini-II APU, explicitly aimed at embedded, low-latency AI applications, as well as a future “Plato” chip.

GSI’s Q2 fiscal 2026 earnings report, due on October 30th, will provide a clearer picture of real-world demand for its technology. Whether GSI can translate this technological breakthrough into sustained sales remains to be seen.

Read More