How Two Harvard Dropouts Are Taking on Nvidia with a Chip That Could Change AI Forever

In a world dominated by tech giants like Nvidia, a daring startup founded by two Harvard dropouts is making waves with a bold bet on specialized AI chips. Etched AI, a company that barely existed two years ago, has already attracted the attention of top investors and tech enthusiasts, thanks to its groundbreaking approach to AI chip development. But this isn’t just another chip startup—it’s a David vs. Goliath story, with Etched AI hoping to redefine the future of artificial intelligence.

From Harvard Dropouts to AI Chips

Founded in 2022 by Gavin Uberti and Chris Zhu, Etched AI was born out of a realization that the world was on the verge of a new AI era. Both co-founders were Harvard students at the time, working summer internships that sparked their interest in chip development. They saw a future where AI models, specifically transformer-based models, would dominate the landscape.

Uberti recalls the defining moment of their journey: “In 2022, we made a bet that transformers would take over the world. We’ve hit a point in the evolution of AI where specialized chips that can perform better than general-purpose GPUs are inevitable — and the technical decision-makers of the world know this.”

With that vision, they made the risky decision to drop out of Harvard and devote themselves fully to Etched AI. The duo quickly assembled a team of engineers and AI experts, and soon the Sohu chip, Etched AI’s flagship product, began to take shape.

The Sohu Chip

At the heart of Etched AI’s ambitions is the Sohu chip, a specialized AI chip designed to run transformer models. Transformers, introduced by Google in 2017, revolutionized how AI processes data, allowing models to consider context—whether it’s the meaning of a sentence, the pixels in an image, or even the notes in a piece of music. These models power today’s most advanced AI systems, including OpenAI’s ChatGPT.

Etched AI’s Sohu chip is unique in its laser focus on transformer models. Unlike Nvidia’s general-purpose GPUs, which are designed to handle a variety of AI tasks, the Sohu chip is an Application-Specific Integrated Circuit (ASIC) designed solely for transformer models. This hyper-specialization allows the chip to outperform Nvidia’s GPUs in terms of efficiency, speed, and raw computing power.

According to Uberti, Sohu offers performance gains that are “an order of magnitude faster and cheaper” than Nvidia’s Blackwell GB200 GPUs. In fact, one Sohu server, with just eight chips, can reportedly handle more than 500,000 Llama 70B tokens per second, compared to just 23,000 tokens per second for an eight-GPU Nvidia H100 server.

But the most jaw-dropping claim is that eight Sohu chips can replace 160 Nvidia H100 GPUs, highlighting the chip’s exceptional efficiency and transformative potential.

The David vs. Goliath Battle

Taking on Nvidia, the undisputed king of AI chips, is no small feat. Nvidia commands a staggering 80% of the AI chip market, and its market cap recently soared beyond $3 trillion. In comparison, Etched’s recent $120 million Series A funding round—led by Primary Venture Partners and Positive Sum Ventures—might seem like a drop in the bucket. But Etched’s founders are undeterred.

“We’re making the biggest bet in AI,” says Uberti. “If transformers go away, we’ll die. But if they stick around, we’ll be the biggest company of all time.”

Etched’s operating chief, Robert Wachen, shares Uberti’s confidence. “The reason we were so excited about what we’re doing, why we dropped out of school and we’ve convinced so many people to leave these chip projects — this is the most important thing to be working on. The entire future of technology is going to be shaped by whether the infrastructure can handle the scale.”

Wachen describes Sohu as the steak knife to Nvidia’s Swiss Army knife—optimized for one task, and doing it far better than a general-purpose tool.

Manufacturing Powerhouse: TSMC and the 4nm Process

To bring the Sohu chip to life, Etched partnered with Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s leading semiconductor foundry. TSMC is fabricating the Sohu chip using its advanced 4nm process, which ensures top-tier performance and efficiency.

With TSMC’s help, Etched AI is set to start production later this year, and the company already has contracts worth tens of millions of dollars lined up. Unnamed customers, including AI companies and cloud platforms, have reserved significant hardware ahead of the chip’s release.

Energy Efficiency and Environmental Impact

One of the most pressing concerns in AI today is energy consumption. Large-scale AI models consume vast amounts of power, leading to increased environmental impact. Etched AI aims to address this problem by designing the Sohu chip to deliver exceptional inferencing performance while consuming less energy than traditional GPUs. This focus on energy efficiency positions Etched AI as a solution for companies looking to scale their AI operations without further straining the planet’s resources.

Betting the Future on Transformers

Etched AI’s entire business model revolves around transformer models. While this gives the company a clear advantage in performance and efficiency, it also presents a significant risk. As Uberti acknowledges, “If transformers go away, our company will die.” But the founders are prepared to pivot if necessary, and Uberti has stated that they would design a new chip if transformer models fall out of favor.

A Growing Team with a Bold Culture

In just two years, Etched AI has grown to a team of 35 employees, many of whom left high-profile projects to join the startup. To attract top-tier talent, the company offers housing stipends and fosters a startup culture focused on speed and innovation. At their headquarters, a countdown timer serves as a constant reminder of their mission to bring the Sohu chip to market as quickly as possible.

As the team continues to push boundaries, Etched AI has attracted an impressive roster of investors, including Peter Thiel (PayPal co-founder), Thomas Dohmke (GitHub CEO), and Charlie Cheever (Quora co-founder). The company’s $120 million Series A funding brings its total raised to $125.36 million, a testament to the belief investors have in the future of transformer-specialized AI chips.

Etched AI’s Sohu Developer Cloud

Looking to the future, Etched AI is preparing to launch the Sohu Developer Cloud, an interactive platform where customers can preview the chip’s capabilities before committing to large orders. This online platform could be the key to driving further interest and sales, particularly for companies eager to push the limits of transformer-based AI applications.

The Stakes Are High

Etched AI’s story is one of incredible ambition and risk. The company has bet everything on the success of transformer models, which now power the most advanced generative AI applications in the world. But as with any bet, the stakes are high. The specialized nature of the Sohu chip means that if transformers lose relevance, Etched could face extinction. However, if they remain central to AI’s evolution, Etched AI might just become the next big player in tech—a company that took on Nvidia and won.

As Uberti puts it: “We’re either going to die, or we’re going to be the biggest company of all time.”

Only time will tell if Etched AI’s bold bet will pay off, but one thing is certain: the AI chip industry has a new contender, and they’re not backing down from the fight.