In a bold move that has sent ripples through the tech industry, Etched, a startup founded by Harvard dropouts, has secured a staggering $120 million in funding to develop a specialized AI chip aimed at rivaling the products of industry giant Nvidia. The Series A funding round, led by Primary Venture Partners and Positive Sum Ventures, has attracted the attention of prominent investors, including Peter Thiel, Thomas Dohmke (CEO of Github), and Balaji Srinivasan (former CTO at Coinbase Inc.).
Etched, a mere two years old, is the brainchild of Gavin Uberti and Chris Zhu, who made the audacious decision to leave Harvard to pursue their vision of revolutionizing the AI chip market. The young entrepreneurs have set their sights on creating a chip that excels at running transformer models, the architecture that powers cutting-edge AI systems like OpenAI's ChatGPT and Google's Gemini.
The Sohu chip, as it has been dubbed, is an application-specific integrated circuit (ASIC) tailored to handle the unique demands of transformer models. By focusing on this specific task, Etched claims that Sohu will deliver unparalleled performance, outpacing even Nvidia's upcoming Blackwell GB200 GPUs. Uberti boldly asserts, “One Sohu server replaces 160 H100 GPUs, making it a more affordable, efficient, and environmentally friendly option for businesses that require specialized chips.“
Etched's decision to target transformer models is a calculated gamble, as these models have demonstrated an unrivaled ability to scale and handle increasingly complex AI tasks. Uberti explains, “By scaling AI models and feeding them more training data, they become significantly smarter. We believe this trend will continue, with major companies investing billions in AI data centers.“
The startup argues that as AI models grow in complexity, the demand for specialized chips will surge. While general-purpose chips like GPUs have been the go-to solution, they face limitations in efficiently handling transformer inference. Etched is confident that the future belongs to specialized chips, drawing parallels to the evolution of other computing markets, such as Bitcoin mining, where specialized hardware has become the norm.
Etched has forged a strategic partnership with Taiwan Semiconductor Manufacturing Co. (TSMC) to manufacture its chips using the cutting-edge 4-nanometer process. The substantial funding raised will be crucial in covering the high costs associated with chip design and production, a process known as “taping out.”
The AI chip market is currently dominated by Nvidia, which commands an estimated 80% of server AI chip sales. However, Etched is undeterred by the challenge, believing that its laser-focus on transformer models and specialized hardware will give it a competitive edge. As the AI landscape consolidates around transformer-based architectures, the demand for purpose-built chips is expected to soar.
If Etched's bet pays off, the implications could be far-reaching. The startup envisions a future where real-time video, audio, agents, and search become possible thanks to the efficiency and speed of its Sohu chips. This could democratize access to AI technologies, making them more affordable and widely available.
However, the road ahead is not without obstacles. Nvidia's position as the world's most valuable company, with a market capitalization surpassing $3.3 trillion, is a testament to its dominance and resources. Other contenders, such as Intel's Gaudi 3, are also vying for a piece of the AI chip market.
The AI chip market is set for a thrilling ride as Etched and other innovative startups challenge the status quo. The potential for specialized chips to unlock new frontiers in artificial intelligence is immense, and the battle for supremacy is only just beginning. As the world watches with bated breath, one thing is certain: the future of AI will be shaped by the daring visions of trailblazers like Etched.
