Deepak Srivastava, founder of NuMorph, joins Sunny Ray to unpack why he believes today's brute force AI is heading toward a data and power wall, and how he plans to fix it with neuromorphic computing. A former NASA technologist with a PhD in quantum physics from the University of Florida, Deepak was among the first researchers to build computational nanotechnology models at NASA in the mid 1990s before quietly incubating a startup out of the agency's research park in Silicon Valley. He explains how watching chess and Go breakthroughs convinced him, years before ChatGPT, that scaling raw computing power was unsustainable, and how a swarm of fruit flies drawn to old bananas during the pandemic pushed him toward biological intelligence as a model. Inspired by how insects and honeybees achieve complex behavior at extremely low power, Deepak spent months studying computational biology and patent literature before building NuMorph's neuromorphic AI, chips and sensors, designed to plug into existing camera hardware and dramatically cut power use compared to conventional digital AI systems.
An ex-NASA quantum physicist explains how fruit flies and honeybees inspired his low power neuromorphic answer to AI's coming power crisis.
What are you building right now at NuMorph, and what problem inspired it?
There's a power problem in big AI. I wanted to build something more efficient, and the lowest power intelligent systems we know are biological, like human or animal intelligence. That's my model. In tech terms it's called neuromorphic computing, which is more analog and biological. My company NuMorph is a play on that. We're building neuromorphic AI, plus neuromorphic chips and sensors, all centered on analog computing rather than brute force digital scaling.
What made you leave a NASA research career in quantum physics and nanotechnology to become a startup entrepreneur?
I did my PhD in quantum physics at the University of Florida, then worked in materials physics at Penn State, which got me hired at NASA in 1995 to 97 as one of the first people doing computational nanotechnology. Being in Silicon Valley, people kept telling me I should start a company. NASA let me keep my job while incubating a startup in their research park, and I slowly transitioned starting around 2007.
Back in 2020, before ChatGPT, you decided brute force AI would hit a data and power wall. What made you so sure?
I'm a technical person who likes simplifying complex things. Watching AI beat chess and Go champions, I realized they were just scaling up supercomputing power to analyze billions of moves instead of millions, pure brute force. When the transformer paper came out around 2017-18, it confirmed that large language models throw all the world's data at training. I had a gut feeling that approach would hit a wall on data and power, long before anyone talked about gigawatts or nuclear reactors.
Everyone talks about the AI power wall now. Why was almost nobody saying it back then?
When ChatGPT came out, it felt like a gold rush moment, everyone was excited about its human-like prediction ability. That excitement lasted a year or two before people noticed hallucination problems from feeding it scraped internet data. The power and resource issue only became visible more recently, once the actual bills started arriving for corporations. Businesses are now trying to bring prices down, and that's when everyone really started realizing the power problem was real.
Where's the real break point, what task is the clearest proof that a neuromorphic system beats today's GPU based AI on both power and output?
Even without my own software, published literature and patents over the last 10 to 20 years show neuromorphic chips using crossbar architecture achieve about 95% less power consumption than digital systems. Big companies have proven this works, it just hasn't scaled to large stable systems yet. Neuromorphic sensors with event driven, focus sensing similar to biological sensing already exist commercially. The hardware progress came first, but there's a chicken and egg problem finding application developers to commercialize it.
If companies like Intel and IBM have had neuromorphic chips for years and it still hasn't taken off, why now, and what's your unique insight?
IBM and Intel ran mainframe neuromorphic systems that academic researchers used successfully for years, proving the technology works. But they couldn't get from lab systems to mass commercial developers, that's the chicken and egg problem. We're coming from a different angle, software and application first. Our neuromorphic AI plugs into existing digital frame based camera hardware without requiring companies to rip out their stack, so we become complementary to Intel, IBM, Sony and Samsung rather than competing with them.
You got the idea for NuMorph watching fruit flies over old bananas. Can you walk us through what you saw?
During COVID I had old bananas attract fruit flies, and instead of chemical spray I found a natural trick, putting out a red wine glass with a narrow neck traps them by smell alone since they barely have vision. That got me curious about drosophila, a well studied biological brain model. I moved up in complexity to honeybees, which act like a drone system, flying out, finding flowers, collecting nectar, and returning, with excellent vision and flight control.
What made you go all in on this rather than take a comfortable research job somewhere?
I could have taken a chair professor position at a big Midwest university, but I realized I'm someone who needs new challenges, that's just my personality. My family joked I should retire, and I did try gardening and cooking for a while, which actually gave me time to think and sharpen my ideas. But friends who knew my background told me I was too smart and not old enough to retire, so they encouraged me to do something more challenging.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interview