GSI Technology ($GSIT) had their claims validated by Cornell U researchers that their Gemini-I APU were able to match NVIDIA A6000 GPU performance using 98% less energy.
If this technology can be commercialized at a fraction of the energy savings shown, and be a true competitor to NVIDIA chips, then the case for data centers driving prices in turn driving microgrid adoption will be weaker, but the case against EV adoption and transportation electrification (as electricity prices climb) would also be also weaker. Either way, this is a positive development for those of us concerned about climate change and the disruption to come from the data center and AI boom.
My algorithm is truly reflective of my interests. From an Instagram account @theartificialintelligence:
a tiny chip company just challenged nvidia’s power monopoly 🤯 GSI Technology built something wild — a chip called Gemini-I that matches NVIDIA’s A6000 GPU performance while using 98% less energy. Cornell University validated it: real AI tasks, identical benchmarks — same results at 1–2% of GPU power. Its secret? Compute-in-memory. Instead of wasting energy moving data between memory and processor, Gemini-I computes inside the memory itself — no transfer, no bottleneck.
Results: • 5× faster than CPUs on retrieval-augmented AI tasks • Edge AI for drones, satellites, IoT now practical • Data centers could slash energy bills by 98%
The $100B AI inference market? Officially disrupted. Cornell proved it. GSI built it. The era of energy-efficient AI just began.
