Most of the AI conversation happens in giant data centers. Racks of expensive chips, enormous power bills, and a steady stream of data traveling back and forth over the internet. It works, but it also raises a fair question: does every smart feature in your glasses, earbuds, or kitchen appliance really need to phone home to a server farm?
Sam Fok, co-founder and CEO of femtoAI, says no. In Episode 127 of Machine Minds, Sam sits down with host and Samson Rose CEO, Greg Toroosian, to talk about bringing AI out of the cloud and into everyday products. The conversation covers neuroscience, chip design, hard company pivots, and the unglamorous truth about what makes technology succeed in the real world.
Here are the ideas worth taking away.
From Studying Neurons to Building Brains in Silicon
Sam grew up in a household of scientists, so big questions about intelligence and consciousness were part of life early on. That curiosity pulled him into neuroscience, where he worked on recording brain signals. At some point, though, asking why the brain works the way it does was not enough. He wanted to build something that works the same way.
That desire led him to Stanford’s Brains in Silicon Lab and into neuromorphic engineering, the field of designing hardware inspired by how biological brains process information. It is a rare career arc, moving from observing neurons to engineering chips, and it gives Sam a perspective most hardware founders simply do not have.
What femtoAI Is Building
Today, femtoAI creates brain-inspired hardware and software that lets AI run directly on devices. Think wearables, appliances, robotics, and consumer electronics. The aim is simple to state and hard to execute: AI runs efficiently in the real world without straining power grids, supply chains, and balance sheets.
Running AI on the device itself, rather than in the cloud, brings a few clear advantages:
Lower power use, which matters enormously for small battery-powered products.
Less dependence on constant connectivity.
Reduced pressure on cost, since not every feature needs expensive cloud compute behind it.
Intelligence that can live inside glasses, earbuds, appliances, and robots.
Think about how many products around you could be smarter if intelligence did not depend on a connection. Glasses that respond instantly, earbuds that adapt to your surroundings, appliances that learn your habits, and robots that react in the moment. Each of those becomes more practical when the AI lives right on the device. That shift is the bigger story behind femtoAI, and it is why Sam’s work deserves attention from anyone designing the next generation of connected products.
Why Biology Points Toward Sparse Computation
One of the more interesting technical threads in the episode is how biology shapes femtoAI’s approach. The brain does not light up every neuron for every task. It works sparsely and processes information locally, which is a big reason it runs on so little energy.
Sam explains that sparse computation and localized processing are key to unlocking efficient edge AI. Instead of brute-forcing everything with massive amounts of math, the idea is to do less work, in the right places, at the right time. For anyone building AI-powered hardware, that is a useful design philosophy even outside neuromorphic chips.
The Pivot That Changed Everything
Turning research into a company is rarely a straight line, and Sam is candid about that. Technology that shines in a PhD setting does not automatically translate into something manufacturable and scalable. femtoAI made the hard decision to rebuild its core platform from the ground up so it could work as a commercially viable product.
That kind of choice is painful. It means admitting that what got you here will not get you there. But it also reflects a mindset that shows up throughout the episode: the real goal is not to preserve the clever thing you built first; it is to build the thing customers can actually use.
Sam also touches on why timing mattered for femtoAI’s founding. AI had just crossed from research curiosity into a major commercial force, and the demand for efficient, on-device intelligence started to look very real.
Academia and Industry Reward Different Things
Another honest part of the conversation is the tension between academic incentives and commercial realities. In research, benchmarks and novelty tend to win. In business, reliability and reproducibility matter more. A chip that posts an impressive number once is not the same chip that behaves predictably across thousands of units.
For founders leaving the lab, this shift can be jarring. Some key differences come up:
Academia rewards novelty, while customers reward consistency.
Benchmarks impress viewers, while reproducibility earns trust.
Technical elegance is nice, but ease of use and integration often decide who gets adopted.
That last point deserves repeating. Sam notes that ease of use and integration frequently matter more than technical elegance. Engineers love an elegant solution, but customers love a solution they can drop into their production without a six-month headache.
Two Kinds of Customers
femtoAI serves two distinct customer profiles, and Sam explains how each one thinks. Some companies want turnkey AI features they can ship quickly without building a team of specialists. Others want full-stack tools so they can build and tune their own models.
Supporting both takes discipline. A turnkey customer needs simplicity and fast results. A full-stack customer needs depth, flexibility, and good developer resources. Knowing which customer you are serving at any given moment keeps a product from becoming muddled.
Earning Credibility in Deep Tech
Deep tech is full of grand visions, and customers have heard plenty of them. Sam’s advice on credibility is refreshingly practical: deliver real, testable results rather than selling big promises. Let people try the thing, and decide for themselves.
That approach fits the nature of the field. When your product sits inside someone else’s device, trust is everything. A partner will not bet on a product launch on a slide deck. They will bet on hardware that works and a team that follows through.
Building a Team That Spans the Whole Stack
Hardware startups often need people across silicon, firmware, compilers, and AI tooling. Sam talks about building highly technical teams that share intellectual curiosity and cross-stack awareness. An engineer who understands how their layer affects the layers above and below it makes better decisions and catches problems sooner.
Leadership matters here too. Sam emphasizes trust and empowerment when leading engineers. Skills specialists do their best work when they are given real ownership and not micromanaged. It is a lesson that applies far beyond chip companies.
Beyond Today’s Dominant AI Approaches
Sam shares a thoughtful view on where AI is headed. He believes today’s dominant approaches alone will not be enough to reach truly general robotics. Physical machines operating in messy environments have different demands than a chatbot, including tight power budgets, fast local reactions, and reliability.
Looking ahead, the conversation turns to the next three to five years, when intelligence could spread across devices and environments instead of sitting in a handful of data centers. If that vision plays out, AI will feel less like a service you connect to and more like a built-in capability of the things around you.
Who Should Listen
This episode is a strong fit for several kinds of listeners:
Founders building AI-powered hardware. Learn why turning research into a commercially viable product requires more than technical innovation, including efficient hardware design, practical tradeoffs, and a willingness to rebuild when the original approach no longer works.
Teams deploying robots or smart devices. Understand how on-device AI can reduce reliance on cloud connectivity, improve responsiveness, and help products operate more efficiently in real-world environments.
Engineers curious about neuromorphic computing and edge AI. Explore how principles inspired by the brain, such as sparse computation and localized processing, can inform the development of more energy-efficient AI hardware.
Leaders wondering how to move a research idea toward a scalable product. Gain insight into the differences between academic research and commercial development, including why reproducibility, reliability, team collaboration, and customer needs matter when scaling deep-tech solutions.
The takeaway is clear and grounded. Making AI usable, efficient, and scalable in the real world takes more than clever algorithms. It takes the right architecture, honest tradeoffs, a willingness to rebuild, and a team that trusts each other.
Meet the Host: Greg Toroosian and Samson Rose
Conversations like this one are the reason Machine Minds has become a go-to listen for people in robotics, AI, and hard tech. Host Greg Toroosian of Samson Rose brings the right mix of curiosity and industry knowledge, steering guests like Sam Fok toward the practical details builders actually want to hear.
Samson Rose stays close to the robotics and hard tech community, and that closeness shows up in the quality of the discussion. If you are hiring for a technical team, scaling a hardware company, or simply want to keep up with the people shaping the field, Greg and the Samson Rose team are worth following. Subscribe to Machine Minds wherever you listen, and join the next conversation with the minds behind the machines. Contact us now!
Deepen Your Understanding of Edge AI
Explore More from Machine Minds and femtoAI:
Listen to the Full Discussion: Head over to Spotify or Apple Podcasts for Why the Future of AI Won’t Live in the Cloud with Sam Fok.
Analyze the Technical Architecture: Visit the official femtoAI website to explore their brain-inspired chips, on-device AI solutions, and developer portal.
Connect with the Speaker: To follow insights on neuromorphic engineering, edge AI, and deep tech leadership, follow Sam Fok on LinkedIn.