The advent of generative AI has turbocharged demand for cloud computing services, and data centers need a boost of their own to meet the rising demand. MangoBoost, a Seoul- and Seattle-based chip design startup, says it can give that boost.
Founded in February last year, MangoBoost develops specialized chips designed for data centers, called data processing units, or DPUs. The chips can boost the efficiency of existing server equipment, says Jangwoo Kim, cofounder and CEO of MangoBoost, in a video interview.
“Everybody believes that DPUs are going to be equipped to every server,” says Kim, who holds a doctorate in computer engineering from Carnegie Mellon University and was a senior engineer at Oracle. “So that’s a huge market.”
Some of South Korea’s top tech investors agree. Last month, MangoBoost scored $55 million in a Series A round, bringing its total funding to date to $65 million. The round was led by IMM Investment, which has backed the likes of billionaire Bom Kim’s Coupang and Chang Byung-gyu’s Krafton, as well as Shinhan Venture Investment, an arm of banking giant Shinhan Financial Group whose portfolio companies include billionaire Bang Jun-hyuk’s Netmarble and Indonesian aquaculture unicorn eFishery.
MangoBoost’s Kim declined to share the valuation, but a source with direct knowledge of the deal said the 21-month-old startup was valued at around $300 million in the latest round.
Other backers in the Series A round include all of its seed investors and new investors such as IM Capital Partners, KB Investment, Korea Development Bank and Premier Partners. MangoBoost raised $10 million in a seed round in last May from Korean investors including DSC Investment (which backed Forbes Asia 100 to Watch alum WAD, a restaurant reservation platform), Must Ventures and Stonebridge Ventures (portfolio companies include autonomous drone startup Nearthlab, which made this year’s Forbes Asia 100 to Watch).
Kim says the startup hasn’t started generating revenue, but expects it will “soon,” without providing more details. He adds that its potential customers include cloud computing providers and server makers such as Dell and HP that want central processing units, or CPUs, in their servers to focus more on AI processing by offloading other tasks to MangoBoost’s DPUs.
“So what’s happening in data centers is that they run out of CPUs, so they have to buy more servers just to get the CPU. Then they can accelerate AI or other big data applications, but your CPU becomes busy again, then they buy another server,” explains Kim. “So that’s how the number of servers is increasing and then the data center cost becomes too high.”
“Basically everybody’s nightmare is CPU now. So we are trying to reduce the CPU bandwidth so that Intel’s and AMD’s CPUs can focus on the more important things,” says Kim. “By taking all the important CPU software from CPU to DPU, the CPU becomes free.” In one configuration, MangoBoost’s DPU can reduce CPU usage by up to 95%, he says.
Of course, MangoBoost is not the only company designing DPUs. It’s facing competitions from some of the world’s largest tech companies, including Nvidia’s BlueField, Intel’s Oak Springs Canyon and Mount Evans, AMD’s Pensando (which AMD acquired last year for about $1.9 billion), Microsoft’s Fungible (acquired in January this year for around $190 million) and Amazon’s Nitro Cards.
But Kim is confident that MangoBoost will succeed. One reason is that MangoBoost can customize DPUs for each customer, he says.
Another is its human capital. “Our DPU is outperforming [competitors], and I’m going to tell you why. We are not nobody,” says Kim, showing more than a dozen papers published by MangoBoost’s management team, which includes former principal engineers from Intel, Microsoft and Samsung.
“Basically, we know what we are doing,” he says.
MORE FROM FORBES
Read the full article here