Key takeaways
- Naveen Rao previously sold two AI startups to technology giants
- His third venture reached a $4.5 billion valuation before commercialization
- NeuReality acquisition brings engineers to accelerate ambitious chip development
Naveen Rao has already built two artificial intelligence startups that ended in billion-dollar and multimillion-dollar acquisitions. Now the neuroscientist-turned-entrepreneur is attempting something considerably harder: reinventing how computers process artificial intelligence.
His latest company, Unconventional AI, raised $475 million at a $4.5 billion valuation before releasing a commercial product. On October 26, it announced the acquisition of Israeli chip startup NeuReality in an undisclosed cash-and-stock deal, according to Reuters.
The acquisition could help Rao tackle a fundamental problem behind the AI boom: computing hardware is becoming enormously expensive to operate, and traditional processors may not deliver the energy-efficiency improvements needed for future AI systems.
How Naveen Rao Built and Sold Two AI Startups
Rao's first major entrepreneurial success came in 2016, when Intel acquired Nervana Systems for a reported $408 million.
Founded in 2014, Nervana developed technologies designed to accelerate machine learning. Intel wanted the company's software expertise and specialized hardware capabilities as it sought to compete in the emerging AI chip market.
After leaving Intel in 2020, Rao cofounded MosaicML, which helped businesses train large AI models more efficiently without building their own computing infrastructure.
In 2023, Databricks acquired MosaicML for $1.3 billion, turning Rao's second startup into another lucrative exit.
The deals reflected the growing importance of AI infrastructure, where specialized hardware and software have become central to the industry's economics.
However, Rao's ambitions increasingly extend beyond making existing technologies cheaper or faster.
His Third Startup Wants AI Chips That Use 1,000 Times Less Energy
Unconventional AI aims to develop computing systems inspired by biological processes, with an ambitious goal of making AI hardware up to 1,000 times more energy-efficient than existing approaches.
Rather than relying entirely on conventional digital computing, Rao is investigating architectures where simple computing elements interact dynamically to produce results.
The approach could challenge the way Nvidia GPUs and other accelerators process AI workloads, although the claimed efficiency improvement remains a development target rather than a demonstrated commercial result.
The opportunity is substantial because data-center electricity demand is becoming a constraint on AI expansion.
Unconventional AI initially expected roughly three years of research and development before progressing toward a commercial product over approximately five years.
But Rao says AI-assisted engineering has accelerated its schedule. The company has already developed its first experimental chip and is approaching completion of a second.
Why Unconventional AI Acquired NeuReality
NeuReality, founded in Israel in 2019, has developed specialized AI inference hardware designed to improve the efficiency of running trained models.
Its engineers bring experience in chip design, production and commercial deployment. Approximately 40 employees are expected to join Unconventional AI, which plans to maintain operations in Israel.
That expertise matters because designing an experimental processor is fundamentally different from manufacturing reliable hardware at scale.
Even companies investing billions in custom AI chips must solve problems involving fabrication, software compatibility and integration with existing computing systems.
Unconventional AI is backed by investors including Andreessen Horowitz, Lightspeed, Sequoia and Jeff Bezos. Its financing reflects growing interest in alternatives to today's dominant computing architectures.
Meanwhile, Nvidia's infrastructure ecosystem continues attracting enormous capital, illustrating the scale of the market Rao hopes to challenge.
His previous startups succeeded by solving problems that became increasingly important to major technology companies. His third venture faces a much bigger test: proving that a radically different computing architecture can work reliably and economically outside the laboratory.
