Etched co-founders Robert Watchen, Gavin Uberti and Chris Zhu
etching
Nvidia The stock has been the hottest topic on Wall Street of late, growing its market value eightfold since the end of 2022, surging to more than $3 trillion this month.
The two-year-old startup founded by a Harvard dropout just raised $120 million in venture capital in an attempt to build a competitive chip and compete with Nvidia in artificial intelligence.
Headquartered in Cupertino, California—home of Apple— etching It is developing a chip called Sohu, which the company says will be used to train and deploy artificial intelligence models using “Transformers,” the core architecture that underpins advances such as OpenAI’s ChatGPT.
Co-founder and CEO Gavin Uberti said that as artificial intelligence develops, most of the technology’s power-hungry computing needs will be met by custom hard-wired chips called ASICs. Their efficiency lies in AI models that only perform what they were designed to do, whereas Nvidia’s general-purpose graphics processing units (GPUs) are more capable but also more expensive.
“We have made the biggest bet in artificial intelligence,” Uberti said in an interview. “If the Transformers disappear, we’ll be dead. But if they continue to exist, we’ll be the biggest company ever.”
Uberti and his co-founders realized it was a high-stakes bet, taking on some of the most well-capitalized and competitive companies on the planet. While raising $120 million in Series A funding is a huge amount of money, it’s also about how much revenue Nvidia can generate in half a day. Nvidia’s sales have more than tripled annually for three consecutive quarters, most recently topping $26 billion.
It is estimated that Nvidia owns more than 80% of the AI chip market. Etched is one of many startups attracting capital to pursue this emerging opportunity. Primary Venture Partners led the round. Peter Thiel, Stanley Druckenmiller and Cruise founder Kyle Vogt are also backers.
Despite Nvidia’s lead, as some developers describe it, new chipmakers are still forging ahead, largely because the opportunity is so great. Other chip startups competing with Nvidia include Cerebras Systems, which is building larger AI chips, and Tenstorrent, which is building AI chips using a popular technology called RISC-V.
“We were so excited about what we were doing, and why we dropped out, and we convinced so many people to leave these chip projects, was the most important thing we had to do,” said Robert Wachen of Etched. Manager. “The entire future of technology will depend on whether the infrastructure can handle scale.”
Semiconductors have traditionally been one of the most difficult industries for new companies to start in due to long development cycles, the large capital required to manufacture wafers, and the need to work with a limited number of manufacturing partners, e.g. British SemiconductorEtched’s wafers are being built.
According to PitchBook data, venture capitalists invested $6 billion in artificial intelligence semiconductor companies in 2023, slightly higher than the $5.7 billion in 2022.
hardcode
Uberti and Chris Zhu Uberti started working at a chip company after a summer internship in the compiler department. This exposed him to low-level hardware creativity, and eventually Etched was born.
In 2022, the two dropped out of Harvard and joined Uberti’s college roommate, Wachen. They quickly began hiring chip industry veterans. The company is based in Cupertino and currently has 35 employees. It provides housing subsidies for new employees.
“When ChatGPT launched, Nvidia’s stock exploded, especially when all the other models launching were also Transformers, and we found ourselves in the right place at the right time,” Uberti said.
Etched is preparing to bring Sohu to market, and the founders say they will have something ready to show off later this year. The startup is also committed to ensuring customer safety and says technology companies are eager to try new artificial intelligence chips.
To keep their business running smoothly, companies spending billions on GPUs need to save significant costs in building custom chips designed specifically for their specific AI models, and be willing to make trade-offs in flexibility.
Uberti says that by focusing on transformers, which move data from the chip to memory in a predictable manner, Etched’s Sohu chips can allocate less space to memory and more to defining the raw operations of the chip. capabilities of the transistor.
Another aspect of Eteched’s efficiency is that the chip has a large core. This results in a reduction in inefficient computations performed by the streaming multiprocessor by coordinating operations on different cores.
Uberti said the impact of dedicated AI chips could be similar to ASIC custom chips, first introduced in 2013 specifically for mining Bitcoin or Ethereum, reducing the need for Nvidia GPUs.
Etched’s founders expect demand for chips to run these models will increase, especially once they are used to power artificial intelligence software millions of times per minute.
They also say that by hardcoding the AI architecture into the chip, their devices can reduce the latency in returning answers, unlocking new use cases such as artificial intelligence agents or real-time voice conversations. Etched says its chips are more than 10 times faster than Nvidia’s GPUs due to their simpler architecture and single use case.
But Etched is competing with some of the most valuable companies in the world, including Nvidia, which have large development teams and access to the capital needed to ensure production and improve the wafers every year.
Etched headquarters has a countdown timer, so the pace has to be increased.
“For us to win in this specialized AI chip market and beyond, we have to be first to market,” Uberti said.
About The Author
