AI Chipmaker Etched Raises $120M Claiming 20x Speed-up Over NVIDIA H100

Etched Secures $120M to Challenge Nvidia with ASIC for Transformer Models

Short Summary:

  • Etched has raised $120 million in a Series A funding round.
  • The company develops specialized ASICs for transformer AI models.
  • Etched aims to outperform Nvidia in cost and energy efficiency.

In a significant move that signals a shift in the competitive landscape of AI chip manufacturing, Etched has announced it has raised $120 million in a Series A funding round. Founded merely two years ago by Harvard dropouts Gavin Uberti and Chris Zhu, along with Robert Wachen and former Cypress Semiconductor CTO Mark Ross, the company is making waves with its application-specific integrated circuit (ASIC), named Sohu. These specialized chips are designed exclusively to run transformer models, which have become the backbone of modern generative AI technologies.

Transformers are not just an emerging trend; they dominate the generative AI landscape, underpinning models from OpenAI’s video-generating Sora to Google’s text-creating Gemini. “In 2022, we made a bet that transformers would take over the world,” said Uberti, CEO of Etched, in an interview with TechCrunch. “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.”

The Sohu chip is manufactured using TSMC’s 4nm process and claims to deliver vastly improved inferencing performance compared to traditional GPUs, all while consuming less energy. “Sohu is an order of magnitude faster and cheaper than even Nvidia’s next generation of Blackwell GB200 GPUs when running text, image, and video transformers,” remarked Uberti. “One Sohu server replaces 160 H100 GPUs. We believe Sohu will be a more affordable, efficient, and environmentally friendly option for business leaders that need specialized chips.”

sohu

What sets Sohu apart is its streamlined inferencing pipeline, designed exclusively for transformer models. This allows the hardware to discard unnecessary components and trim software overheads that would otherwise slow down or complicate the inferencing process. As Uberti elaborated, “Companies are willing to take a bet on Etched because speed and cost are existential to the AI products they are trying to build.”

“In short, our future customers won’t be able to afford not to switch to Sohu,” Uberti said.

Moreover, this advancement comes at a critical time. Current AI infrastructure, primarily powered by GPUs, is notoriously resource-intensive. Goldman Sachs has predicted that AI will drive a staggering 160% increase in data center electricity demand by 2030. Researchers at UC Riverside warn that global AI usage could necessitate up to 1.7 trillion gallons of fresh water for data center cooling by 2027, impacting local water resources significantly. Uberti posits that Sohu could be the answer to this looming crisis, providing a more sustainable and economical solution.

While Etched is currently one of the few companies focusing on transformer-specific chips, competition is on the horizon. AI chip startup Perceive recently teased a processor with hardware acceleration for transformers, and Groq is investing heavily in similar optimizations. Despite this, Etched appears confident in their unique approach and technical advantages.

This confidence is shared by a cadre of influential investors. The funding round, co-led by Primary Venture Partners and Positive Sum Ventures, brings Etched’s total raised to $125.36 million. Notable angel investors include Peter Thiel, GitHub CEO Thomas Dohmke, Cruise co-founder Kyle Vogt, and Quora co-founder Charlie Cheever. These investors believe in Etched’s potential for scaling up their business, substantiated by “tens of millions of dollars” in hardware reservations, according to Uberti.

Looking forward, Etched plans to ramp up production and bring Sohu to market soon. The forthcoming launch of the Sohu Developer Cloud, an online interactive playground to test-drive Sohu, is expected to generate additional sales and interest. This move underscores Etched’s ambition to become a critical player in the AI infrastructure market.

“Video generation, audio-to-audio modalities, robotics, and other future AI use cases will only be possible with a faster chip like Sohu,” Uberti said. “The entire future of AI technology will be shaped by whether the infrastructure can scale.”

Despite the optimism, the AI chip segment remains highly competitive and unforgiving. Previous attempts by startups like Mythic and Graphcore to disrupt the market saw limited success. The sector is also witnessing a decline in investments, posing additional risks. Still, Etched’s innovative approach and significant backing position it well to challenge the industry giants.

The tech world watches with bated breath as Etched gears up for its market debut. For PC hardware enthusiasts and tech aficionados, the success of Etched could mean more efficient, affordable, and specialized components in their future builds, pushing the boundaries of what consumer tech can achieve. As we move deeper into the era of AI, the stakes couldn’t be higher, and Etched seems ready to rise to the challenge.

About Author