Flucta
Probabilistic computing stack for efficient AI hardware.
The founders
TS
The company
Flucta is designing a computing stack, from algorithms down to custom silicon, that uses probabilistic computing, sparsity, asynchrony and approximate computing to run AI models. The hardware reads only a fraction of the data dense accelerators read, and models do not need retraining or fine-tuning to run on it.
AI models now tolerate extreme sparsity and random sampling, but conventional accelerators still read memory densely, and reading a byte can cost over 250 times the energy of computing on it. Flucta's target is a 100-fold improvement in intelligence per watt.
Flucta is based in San Francisco and was founded in 2026.
Dealroom lists Tathagata Srimani as the founder and has no further background on file.
Who else is building in this space
Companies working on the same thing, at every stage.
Unicorn
AI inference chips and rack-scale clusters
Unicorn
Memory-first inference silicon and systems using commodity LPDDR for LLMs
Unicorn
Power-efficient AI chips for data center inference, optimizing performance per watt for advanced AI models
Unicorn
Developer of high-performance GPU chips and intelligent computing solutions for AI and data centers
Unicorn
Fabless provider of AI solutions for cloud and edge, offering cost-effective, energy-efficient SoC chips and user-friendly software
Unicorn
Specialized AI inference chips for enterprise data centers
Thousands more founded every month.
Dealroom tracks new companies as they appear — founders, funding and signals, over the API or in the platform.