archy13 - stock.adobe.com
SambaNova Systems, a startup that makes chips for AI workloads, raised $676 million in Series D funding. The funding round, led by Softbank Group's Vision Fund 2, indicates a sign of confidence in the AI hardware market, and SambaNova's ability to take a share of it.
"There are not a lot of pure-play AI hardware vendors," said R "Ray" Wang, founder of Constellation Research. "Investors see a winner-takes-all market here, and SambaNova is one of the potential winners."
Founded in 2017 by Stanford professors Kunle Olukotun and Christopher Ré with Rodrigo Liang, previously a vice president at Oracle, SambaNova didn't come out of stealth until December of last year. The startup has raised more than $1 billion to date.
Its flagship chip, Cardinal, contains 40 billion transistors. It's built on Taiwan Semiconductor Manufacturing Company's N7 process and designed for AI training workloads.
Unlike competing AI chipmakers, like Nvidia, SambaNova doesn't sell its AI chips directly to enterprises. Instead, the startup has incorporated the chip its SambaNova DataScale server, which enterprises can lease.
The pre-built servers include SambaFlow, a development interface designed specifically for DataScale. According to SambaNova, the interface automatically extracts, optimizes and executes the optimal dataflow graph for user's models running on SambaNova's AI chips.
R 'Ray' WangFounder, Constellation Research
In a release, SambaNova said it will use its latest funding round to "aggressively challenge" legacy competitors through its AI hardware and software. How that will play out remains to be seen -- Nvidia, likely SambaNova's biggest competitor, is a powerhouse in the AI hardware market with powerful GPUs and a new CPU designed for AI.
Still, according to Wang, AI hardware isn't easy to pull off.
SambaNova is a growing player in a largely unoccupied market. Its latest funding round indicates that the company is growing quickly.