Nvidia, the undisputed leader in AI chips with a market share exceeding 80%, has recently seen its stock price soar. However, a new challenger has emerged – Etched, a 2-year-old startup founded by Harvard dropouts, has secured $120 million in funding to develop a competing AI chip called Sohu.
This raises a crucial question: can Etched carve out a significant space in the market dominated by a tech giant like Nvidia?
Etched’s niche: Specialisation in transformers
Etched’s Sohu chip focuses on a specific type of AI architecture called “transformers,” a core technology underlying advancements in large language models such as OpenAI’s ChatGPT. Etched believes that as AI evolves, specialised, hard-wired chips (ASICs) will become the preferred solution for training and deploying these models due to their superior efficiency and lower cost compared to general-purpose GPUs like Nvidia’s offerings.
“We’re making the biggest bet in AI,” reportedly said Gavin Uberti, co-founder and CEO of Etched. “If transformers go away, we’ll die. But if they stick around, we’re the biggest company of all time.”
This “all-or-nothing” approach highlights the inherent risk involved in Etched’s strategy. While focusing on a specific technology can lead to superior efficiency, it also exposes the company to market fluctuations if transformer architectures lose favour.
The challenges of disruption
While $120 million might seem like a significant amount, it pales in comparison to Nvidia’s daily revenue stream. Additionally, Nvidia possesses a large and experienced development team, extensive resources, and established industry relationships – all factors that give them a significant competitive edge.
However, Etched is not alone in challenging Nvidia’s dominance. Other startups like Cerebras Systems and Tenstorrent are also pursuing specialised AI chip development, each with their own technological approaches. This growing interest in AI-specific hardware reflects the immense potential of the market, motivating investors to explore alternative solutions.
Etched’s path forward
Etched recognizes the need for speed to be successful. They are aiming to be the first company to market with a transformer-specific AI chip. Their strategy hinges on the belief that customised chips designed for specific workloads can offer significant cost and efficiency advantages over general-purpose GPUs.
“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,” reportedly said Robert Wachen, Etched’s operating chief. “The entire future of technology is going to be shaped by whether the infrastructure can handle the scale.”
Etched claims their Sohu chip can be more than 10 times faster than Nvidia’s GPUs due to its simpler, single-use architecture. Additionally, by partnering with Taiwan Semiconductor Manufacturing Co. (TSMC), a leading chip manufacturer, Etched aims to ensure production capacity for their product.
What we think about this development
The venture capital landscape is showing renewed interest in semiconductor companies. After a dip in 2023, funding for AI chip startups has rebounded in 2024, with companies like Black Semiconductor, Celestial AI, and Recogni securing significant investments. This trend reflects the growing recognition of the need for specialised hardware solutions to handle the ever-increasing demands of AI applications.
The success of Etched’s endeavour will likely depend on their ability to demonstrate the cost-effectiveness, efficiency, and performance advantages of their transformer-specific chip against established players like Nvidia. While overcoming Nvidia’s dominance might seem like a daunting task, the growing demand for AI hardware and Etched’s strategic focus create a potentially exciting David vs. Goliath battle that will shape the future of AI chip development.
