Deep Silicon: From Dartmouth to Startup Success

Many Dartmouth students aspire to start their own business. Two Dartmouth 27’s, Abhinav Reddy and Alexander Nanda, took the leap to make that aspiration a reality.

To pursue this vision, they applied to the summer batch of the highly competitive startup accelerator, Y Combinator—a prestigious program known for launching companies like Airbnb, DoorDash, and Dropbox—and earned a coveted spot.

Abhinav (CEO) and Alex (CTO) are the co-founders of Deep Silicon, a startup working on reducing the size of transformer models—a core element of Artificial Intelligence systems. Deep Silicon is creating technology that allows large language models, which understand and generate human language, to run on smaller, local devices. 

The inspiration behind Deep Silicon came from several sources. One was a BitNet paper from Microsoft, which expressed frustration over the inability to run state-of-the-art (SOTA) models on consumer hardware. Another was a graduate machine learning class that Abhinav and Alex took, which deepened their understanding of advanced AI models.

The most pivotal moment, however, came during a talk Abhinav gave at Dartmouth’s DALI Lab, a student-powered design and innovation hub that supports technology projects and startups. While discussing a new series of chips that included extra cache—memory used to store frequently accessed data to reduce processing time—a student asked, “Why don’t you just make a chip that’s all cache?”

This question sparked the idea that would eventually become Deep Silicon.

The founders realized that by using a small cache and fitting all the data within it, large language models could run significantly faster. “It’s a fusion of hardware design and existing solutions that really inspired this approach,” Abhinav explained.

The path from idea to startup wasn’t straightforward, Abhinav acknowledged, but his background in chip technology gave him confidence.  “I know what I’m talking about for chips. I knew every competitor in the chip market [working on] large language models and how they stacked up against ours,” he explained. With his experience and the clear potential of their idea, pursuing a startup felt like the natural next step. “All paths led to it.”

Another source of inspiration was Dartmouth student Edward Zhang, a member of the Magnuson Center and DALI. “He was really influential because he was doing a startup before [I was],” Abhinav said. The summer after their freshman year, Abhinav and Alex reached out to Edward to join their team, making him their first hire. “He was helpful because I was wondering, ‘Are we doing a startup because we want to do a startup? Or are we doing a startup because this is the fastest way to make this vision a reality?’ He helped us figure that out.”

The pair originally applied for grant funding, both from external sources and within Dartmouth, before Alex had the idea to apply to Y Combinator: a startup accelerator and venture capital firm that supports early-stage companies. 

They applied in April of their freshman year at Dartmouth. “We applied about one or two days before the deadline for the summer batch,” Abhinav explained. “We didn't really anticipate anything happening, but it looks like YC saw something that we didn't see at the time.”

Working with Y Combinator catapulted Deep Silicon’s growth. Within the first 3 months, Abhinav and Alex received $500,000 in funding and $1.2 million in compute credits, providing crucial support for their research. Simply being associated with YC boosted their company evaluation, Abhinav explained. YC also gave the pair opportunities to make connections with major companies. “A great example is Replit”, Abhinav said. At a dinner for YC founders, Abhinav met Amjad Masad, the CEO of Replit, a fellow YC company. After hearing about Deep Silicon’s work, Replit agreed to try out their product. Reflecting on YC’s impact, Abhinav remarked, “They say this jokingly, that they’re the Illuminati, but they kind of are.”

At the beginning, Deep Silicon didn’t focus on profitability. What mattered was growth. “You can just kind of eat venture capital money until you're profitable…but you need to be growing month-on-month, year-on-year, 200 to 300%,” Abhinav said.

According to Abhinav, YC focuses on founders, not startup ideas themselves. “They told one of our close friends in the batch that, your idea sucks, but we like you. So we'll let you in, but you'll have to change your idea,” he said. “They really care about the founders…I don't think there's any possible way that anyone would have taken a twenty and nineteen year old who want to do state of the art research by themselves seriously. It's YC that decides to take that chance, and that's the only reason we're in the position we’re in.”

“Everyone there is basically a genius,” Abhinav said regarding YC. “I don't think I've met a smarter group of people at any point in my life…I think I've made more friends at YC in three months than I have in the rest of my life combined”. 

Turning Deep Silicon into a startup involved unlearning some “bad habits”. While the temptation for engineers and tech-minded individuals is to assume good technology leads to customers, “selling technology doesn’t work”, Abhinav explained. “YC's whole mantra is product-market-fit. No matter how good the technology is fundamentally, you have to find that product-market-fit. Building your product for customers matters a lot…the user experience is way more critical than building the best technical product.”

Deep Silicon is currently working with enterprises—particularly defense companies, like Replit. They offer these companies faster language models and lowered operating costs. But startups pivot multiple times before they find something that works, according to Abhinav. Deep Silicon’s business model could change in a matter of months.

Deep silicon is focusing on software at the moment. “Hardware is pretty expensive,” Abhinav explained. “You can still run these models on existing hardware, so we still try to integrate and work with the community to get it to run. Down the line if there's enough traction, it would make sense to explore hardware.”

The founders are emphasizing fitting their technology into existing company systems. “They can just add two or three lines of code and get everything that we have to offer. The real focus here is the customer experience—making sure the customers have to do as little work as possible,” Abinav said. 

Looking ahead, Abhinav and Alex envision making their technology accessible to everyday consumers. “The dream is to allow consumers to run models,” Abhinav said. In fields like healthcare, law, and finance—or for individuals concerned about privacy—running language models locally ensures that sensitive information remains secure and isn't intercepted or harvested. “Our goal is to have our own compute cluster, run experiments 24/7, and ship to customers,” Abhinav added. “Over the next few months, people might hear about us building something for consumers.” 

The biggest piece of advice the founders have for students looking to pursue a startup is to learn as much as possible. They emphasize the importance of gaining in-depth knowledge about the product, market, customers, and competitors. “I was talking to our investors, and the biggest reason they decided to invest is because I know what I'm talking about,” Abhinav explained. “Let's say you have an idea. Look for as many red flags as you can about why you shouldn't do it. If you can't find enough red flags—if there isn't a company that's doing exactly what you're doing, or if there is, they don't have a central insight that you have—then that's a pretty good startup idea, and you should go for it.”