Tensordyne Launches A Bold Approach For AI: Log Math

InnovationEnterprise TechTensordyne Launches A Bold Approach For AI: Log MathByKarl Freund,Contributor.Forbes contributors publish independent expert analyses and insights. Founder and Principal Analyst, Cambrian-AI Research LLCFollow AuthorJun 15, 2026, 12:00pm EDTSummaryTensordyne has unveiled the Napier processor, an AI accelerator leveraging 16th-century logarithmic math to revolutionize inference. By converting complex multiplication into simpler addition, this air-cooled system significantly boosts efficiency, claiming a 17-fold performance-per-watt improvement over Nvidia’s GB300 for DeepSeek-R1 inference. Designed specifically for LLMs, the Napier processor features a two-tiered memory architecture and offers substantial cost savings. A single Tensordyne rack delivers 1300 tokens per second for a 2T MoE model at 120kW, costing just $11 per million tokens, starkly outperforming Nvidia/Groq's 800 TPS at 1.5MW and $150. This innovative, lower-cost solution is poised to attract major hyperscalers and data centers seeking efficient AI infrastructure.Show More The Tensordyne "Napier" AI platform, named after the 16th century mathematician who discovered logarithms.TensordyneRemember in high school when your math teacher tried to teach (logarithmic) log math? Yeah, me either. But now Tensordyne, a startup based in Germany and California, has launched the first AI accelerator that operates in the logarithmic domain, claiming a 17-fold improvement in performance per watt over Nvidia’s GB300 rack (DeepSeek-R1) in an air-cooled 72-chip chassis. The new system will begin shipping later this year with volume production expected in mid-2027. (Disclosure: Cambrian-AI Research provides advisory and research services to Nvidia and Cerebras.)Log Math? Really?So, what is log-math? Invented in the mid-16th century by John Napier, log math enables a simple mathematical trick that allows multiplication to be rewritten as addition; log(ab) = log(a) + log(b). And tha...