Laurent - stock.adobe.com
Nvidia has closed on the $7 billion acquisition of Israeli-American Mellanox Technologies Ltd. in a bid to enhance its high-performance computing products, as well as Nvidia AI-focused products, such as its services for AI in the cloud.
The purchase, initially announced more than a year ago, is a strategic move, but it doesn't substantially improve their ability to compete in the AI infrastructure market, according to at least one analyst.
Mellanox Technologies, founded in 1999 and based in Sunnyvale, Calif., develops and sells an array of end-to-end networking products based on its InfiniBand and Ethernet technologies. Its interconnected platforms, which aim to provide low latency, high bandwidth and message rate, and smart offloads, can help accelerate AI infrastructures, and power large-scale machine learning training and inferencing systems.
Mellanox will strengthen Nvidia's ability to compete in the high-performance computing market, enabling the vendor to optimize datacenter-scale workloads across computing, networking and storage stacks to lower operating costs and achieve higher performance.
From an AI perspective, the Mellanox acquisition, which was finalized April 27, can potentially enhance Nvidia's own AI offerings, such as its services for AI in the cloud and its AI systems.
Still, the deal isn't likely to significantly boost Nvidia's position in the crowded AI infrastructure market.
Nvidia already held a strong position in the AI infrastructure market, said Forrester analyst Mike Gualtieri. The deal won't necessarily enhance Nvidia's GPU hardware, but rather its AI systems.
"A case can be made that networking chops under Nvidia can improve its ability to deliver AI systems, but it is more likely that Nvidia will compete in the AI infrastructure market by supplying systems and cloud vendors with components rather than AI systems and/or AI cloud services directly," he said.
Competition in AI infrastructure sector
That's not to say that Nvidia doesn't face strong competition in the AI infrastructure market. While it was one of the first vendors to develop hardware specifically for developing and running AI systems, and still remains one of the only companies that produces AI-focused GPUs, it now faces competition from major technology vendors and startups alike.
Intel, for example, boasts a portfolio of AI chips, which it bolstered with the recent acquisition of deep learning accelerator startup Habana Labs. Meanwhile, Google sells the widely used Tensorflow-specific Tensor processing units, as well as the super-powered Cloud TPU Pods. Amazon, AMD and Huawei are also major players in the market.
Mike GualtieriAnalyst, Forrester
"Vendors like Intel will give Nvidia a run for their money literally for both AI training and inferencing solutions," Gualtieri said. "Nvidia is much less king of AI infrastructure than it once was, and it is not difficult to see how Intel races to the top with its broad rollout of AI chips."
The Mellanox deal, first announced in March 2019, was delayed by competing vendors and regulators, particularly regulators in China, the world's largest market for semiconductors. Mellanox's technology helps power numerous vendors' GPUs and chips, and competitors and regulators feared Nvidia would attempt to close off that technology once the acquisition came through.
The deal was approved after Nvidia offered assurances that would not happen.