CEO Satya Nadella said he wants to make Google “dance” when asked about Microsoft’s search business objectives. Microsoft sees itself in AMD and wants to disrupt Nvidia’s monopoly.
Just a few days ago, news surfaced that Microsoft had been developing its own AI processors, codenamed Athena, since 2019 to compete with Google and AWS, who have their own chips for training and inference. On top of that, recent sources claim that Microsoft is working with AMD on this secret project.
Microsoft may be funding AMD’s AI chip push, regardless of Athena. Microsoft’s strong cooperation with Nvidia, which trains OpenAI’s huge language models on Azure, is startling. AMD and Microsoft may have conspired to overthrow Nvidia. The report lowered Nvidia’s stock.
Microsoft used AMD before. AMD’s technology powers Azure’s secure AI infrastructure and Xbox Series X and Series S consoles.
Nvidia dominates the GPU market, thus the industry wants an alternative. Intel and AMD are competing over crumbs in the data center, while Nvidia dominates sales. Microsoft may desire a cheaper Nvidia option to manage their burden.
“Intel is still failing Microsoft. According to semiconductor researcher Sravan Kundojjala, AMD has great GPU hardware but poor software optimization. AMD will release the Instinct MI300, its first data center/HPC-class APU, later this year.
AMD created ROCmTM, an open software platform that provides HPC and AI communities with open compute languages, compilers, libraries, and tools, to compete with Nvidia. It’s immature. Kundojjala said Nvidia’s CUDA works across sectors, while AMD doesn’t. AMD is targeting verticals, clients, and workloads in the data center to position itself. Microsoft is a client.
Nvidia cannot satisfy all use cases or customers. “It’s a growing tide, lifts all the boat kind of situation,” Kundojjala adds. Nvidia can win if AMD wins. Both can expand in the huge market. He also believes Microsoft’s AMD campaign won’t affect Nvidia’s GPU price.
Data centre clients are price insensitive, unlike consumer applications. Pricing is irrelevant to enterprise cloud. He continued, “What they need is that the product, the roadmap, and the maturity of the software framework that they provide along with the hardware product are good.”
Generative AI goldrush
AMD CEO Lisa Su mentioned semi-custom options with larger volume potential in the latest results call.
AMD is replicating its semi-custom gaming console business model in data centers with hyperscalers like Microsoft. Kundojjala believes Microsoft may help AMD speed its AI software.
Kundojjala notes that AMD’s data center GPU chip sales is only a few hundred million dollars, compared to Nvidia’s $15 to $16 billion. AMD arrived late because it never focused on it. ‘They were eating Intel’s lunch,’ he claimed.
AMD is still making money despite a PC inventory correction and data center sales slowdown, while Intel is losing money and investing billions in new fabs. AMD also claims that its new Ryzen 7040U series laptop CPUs outperform Intel and the MacBook M2. They’re progressing well.
AMD realizes data centers need generative AI. Since their MI300 device won’t launch until 2024, nothing will change. AMD won’t dominate data center GPUs for three years.