Artificial intelligence is not only big business, estimated to reach over $422 billion in global spending by 2028, it requires big data computing on a scale few can comprehend.
AI research efforts from Alphabet’s DeepMindAI, Meta, OpenAI and others are pushing the current known capabilities of today’s computing environments. These algorithms get better by getting bigger. The total computing power required to train the world’s biggest AI model has been doubling every three and a half months for the past several years, far outpacing Moore’s law.
Luminous Computing Founders (l) Marcus Gomez and (r) Michael Nahmias.
Luminous Computing is trying to change that dynamic. The Mountain View, California-based start-up, founded in 2018 by Marcus Gomez, CEO, Mitchell Nahmias, CTO and Michael Gao (who has since left the company), plans to push revolutionary technological advances to maximize artificial intelligence through the development of a new supercomputer built on photonics chips.
“Ten years ago, the largest AI models took maybe an hour to train on a single machine. The biggest models today take months on tens of thousands of machines requiring hundreds of engineers. The problem is that we’ve maxed out the capability of existing hardware. And what’s required to get species defining technology improvements, to do things like talk to your computer, deploy robotics at scale or completely automate medicine, is a supercomputer that can actually support the next generation of AI. That’s what we’re building here,” says Gomez.
The luminous founders set out to solve the challenge of providing orders of magnitude improvements in computing performance without giving up on ease of use, programmability and customer experience in order to make good on their mission to make AI fulfill its potential to automate many aspects of human activity to improve life. It turns out that the bottleneck is communication—between chips, between memories, between boards, between boxes and between racks. And according to Gomez and his co-founders, the answer lies in photonics-based computer architecture.
“I come from an optics background. And we know that optics solves this communication problem. And that’s why we use optics to send data across oceans and across racks in a data center. So, we’re putting optics into the computer architecture to solve this problem directly. But it’s not just using optics, you have to redesign the entire computer architecture from the ground up. And if you’re able to do those two things, one, use optics to solve this data movement problem and two, re-architect the system with optics, then you can actually break past this. [performance-usability] curve. And then you can offer both ease of use and performance and you totally eliminate the complexity of the software stack. And that is really the key thing. This solution completely breaks past this curve and offers something that is both more performant and easier to use,” says CTO Nahmias.
Four years into the creation of Luminous, the company is still about 20 months away from production, although Gomez says that the company already has customers ready to go once they do. However, the promise of the concept and the pedigree and progress the founders and their growing team of engineers have made to date has allowed the company to attract the level of venture capital required to build and produce a new photonics-based supercomputer.
The company has attracted $126 million in funding to date, according to Gomez. Their latest A round was raised on March 3, 2022 for $105 million. Investors include Modern Venture Partners, Horsley Bridge Partners, Third Kind Venture Capital, 8090 Partners, Bill Gates, Neo, Alumni Partners, Strawberry Creek Ventures, Gigafund and others.
Gomez is the son of Indian immigrants who came to the US with less than $1,000 to their name. They settled in a small farm town in Minnesota. “Growing up I didn’t really have exposure to technology. My exposure to cutting edge technology was through pop science magazines,” says Gomez. He was inspired by Steve Jobs’ admonition to think big and came to Silicon Valley to attend Stanford because he wanted to work on solving foundational human problems and artificial intelligence was supposed to be the bedrock on which next step innovations would come from. He graduated from Stanford with a degree in computer science and then worked in the field of artificial intelligence, both during his degree work and later for Google before meeting Nahmias in 2018 when they came together to form Luminous Computing.
“I didn’t get my driver’s license until I was 22 because I was so sure that self driving cars would be readily available. But when I came to Silicon Valley, I was disappointed and frustrated because the gap between what was in Star Trek and what was happening on the ground was still too big. I’ve been doing various types of artificial intelligence work for about a decade at this point. And when Mitch showed me what he had been working on, it was immediately apparent to me that if we could pull off anything remotely close to what he had been working on, it would be the step change required to make that next step in artificial intelligence. ,” says Gomez.
Nahmias grew up in the heart of Silicon Valley in Menlo Park, California, but left California to attend college at Princeton in New Jersey to walk the same halls as the college’s most famous lecturer and scientist, Albert Einstein. “I came to Princeton to learn how to build the next supercomputer. I actually thought at the time that the solution to building the next AI supercomputer was physics to come up with a way of using physics to build new devices to do build new technologies,” says Nahmias. He would go on to earn his Ph.D for his pioneering work in the field of Neuromorphic Photonics. When he met Gomez, he realized that it’s not just physics. It’s supply chains, economics and usability that matter as much as the physics.
Expectations for the company are running high, though whether or not Luminous can make good on its promise, only time will tell. “The minimum expectation that we’ve set for ourselves is that we want to be the hardware provider that every AI practitioner on Earth uses. But the thing that takes you from that to a trillion-dollar opportunity is if you have the best AI hardware on Earth, the best thing you can do with it isn’t just selling AI hardware, it’s building the best artificial intelligence on it. And so the long term strategy of luminous computing is to move up the stack to use our own hardware to build the best artificial intelligence to solve these great problems,” concludes Gomez.
.