The company said South Korean chip startup Panmnesia had raised a seed round that values it at $81.4 million. The amount of money that startups raise in their first rounds varies widely. Still, the amount is usually enough to buy credibility with investors and avoid too much dilution of the company’s equity.
Panmnesia’s business is focused on developing intellectual property around a technology it calls Compute Express Link (CXL), which allows big data center operators to pool devices such as AI accelerator chips, processors, and memory. Memory is often a bottleneck for crunching data for AI applications, and pooling it could help free up processors to perform more complex calculations.
CXL enables devices such as GPUs to access large pools of memory using a high-speed interconnect that makes them more efficient than they might be with only a tiny amount of internal storage. The technology is also being used for accelerating other types of computing workloads.
A group of startups based in Korea is looking to challenge global chip leader Nvidia Corp in the area of artificial intelligence (AI). Rebellions Inc., which designed its ATOM chip to run Microsoft-backed chatbot ChatGPT, beat the latest chips from Qualcomm Technologies Inc. and Nvidia in a benchmark test conducted by an international machine learning engineering organization. The ATOM chip performed 1.8 times and up to 2 times faster in the MLPerf Inference v3.0 test for processing significant language and vision models.
Other companies are trying to get in on the action, too. FuriosaAI, backed by top search engine Naver Corp, and Sapeon Korea Inc, a subsidiary of SK Telecom Co, are both developing AI accelerator chips that can compete with Nvidia’s GPUs. Both seek government contracts for two so-called AI farms that only allow domestic companies to bid on the project, aiming to wean data centers off Nvidia.
The government has earmarked $800 million to promote the development of its homegrown AI chip makers and hopes to boost their share of domestic data center markets from essentially zero today. This month, it will put out a tender for constructing two AI farms, with only domestic companies allowed to bid.
Investors believe that a better, more efficient way to run AI algorithms will make it easier for enterprises to use the technology and generate insights from the vast amounts of data they gather. Those insights can include customer patterns and trends that help businesses optimize their operations. In some cases, the technology is even being used to pore vast volumes of data for potential terrorist activity and other forms of crime. In other cases, the technology is helping to drive new products and services, such as self-driving cars and medical diagnostics. The need for more efficient computing is growing, especially as AI becomes a more significant part of many business processes. And that means more companies will need to invest in better chips.