It’s been a big week for the Big 3 advanced chip companies:
- AMD chips will drive upcoming leadership-class supercomputers contracted by the U.S. Department of Energy (see coverage).
- NVIDIA is investing and partnering and everywhere while releasing its customary storm of new product announcement at its Washington, DC GTC conference this week (coverage)
- Intel, the target of rumors earlier that it might split its chip design business from its foundry business, is now rumored to be on the verge of acquiring AI chip company SambaNova.
Starting with Intel, the company has had a favorable run of financial news of late, including the U.S. government taking a 10 percent stake in the company in August followed by better-than-expected quarterly earnings announced in September and, lastly, NVIDIA’s decision to invest $5 billion in the company, with the implied expectation of integrations of NVIDIA GPUs and Intel CPUs for AI workloads.
In addition, Intel announced earlier this month new 2nm CPU chips produced at its newest and most advanced chip foundry, Fab 52, using its 18a process.
Now there are reports that Intel, which has been hamstrung in its years-long effort to compete in the GPU arena, may be interested in acquiring AI compute company SambaNova. According to a Bloomberg story, the two companies are in talks, though neither company would confirm the report.
SambaNova has been rumored to be open to an acquisition, and there have been reports that its valuation has declined this year.
The Bloomberg article reports that the Intel-SambaNova discussions are in the initial phase and that a deal is not assured. In fact, the articled stated that a different company make come on the scene interested in making the acquisition.
SambaNova is located in Palo Alto, CA, and was formed in 2017. It has often been grouped with other AI-compute startups, such as Cerebras and Groq, that have developed alternative architectures and posted impressive model training and inference benchmark numbers.
SambaNova has developed a “reconfigurable dataflow architecture” designed to speed up AI tasks by dynamically maximizing compute resources. The company says its processors deliver 10x performance increases while consuming a tenth of the power of mainstream AI chips.
Meanwhile, the undisputed leader in the AI compute market, NVIDIA, capped a week of new product announcements with news of major activity in South Korea’s AI industry.
At the APEC Summit in South Korea, NVIDIA announced it is working with that country to expand the nation’s AI infrastructure with over a quarter-million NVIDIA GPUs across its sovereign clouds and AI factories, built with public- and private-sector deployments.
- The Korean government is investing in sovereign AI infrastructure with over 50,000 of the latest NVIDIA GPUs to be deployed across the National AI Computing Center and Korean cloud service and IT providers NHN Cloud, Kakao Corp. and NAVER Cloud.
- Samsung Electronics is building an AI factory with more 50,000 GPUs to accelerate its AI, semiconductor and digital transformation roadmap.
- SK Group is building an AI factory featuring over 50,000 NVIDIA GPUs and Asia’s first industrial AI cloud featuring NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for physical AI and robotics workloads.
- Hyundai Motor Group is collaborating with NVIDIA and the Korean government in building an NVIDIA AI factory with 50,000 NVIDIA Blackwell GPUs to enable integrated AI model training, validation and deployment for manufacturing and autonomous driving.
- NAVER Cloud is expanding its NVIDIA AI infrastructure with over 60,000 GPUs for enterprise and physical AI workloads.
- NAVER Cloud, LG AI Research, SK Telecom, NC AI, Upstage and NVIDIA are developing Korean foundation LLMs to accelerate Korean AI applications through public-private partnerships.
- The Korea Institute of Science and Technology Information is establishing a Center of Excellence for the advancement of quantum computing and science.
In addition, a Bloomberg story reports that NVIDIA may invest up to $1 billion in AI start-up Poolside, a developer of products that automate software coding, with a focus on government and defense applications.
The Bloomberg story reports that NVIDIA’s stake in Poolside could start at $500 million and climb to $1 billion if Poolside hits certain fundraising goals.
The Poolside investment, if it happens, could be added to other deals NVIDIA has put together in recent years in which they invest in companies that then buy NVIDIA products.
