Oracle has penned a deal with Microsoft that will see its cloud infrastructure used to power the tech giant’s Bing search engine. Microsoft claims growing demand for Bing’s AI-infused services is behind the move, but it also reflects the high demand for GPUs used to train and operate artificial intelligence models.
The agreement between the two US tech companies, which feels like a form of cloud computing Inception, was announced on Tuesday. No financial details were released.
Microsoft and Oracle are nominally rivals in the public cloud market, though the latter’s 2% market share is dwarfed by Microsoft’s 23% slice. It is not the first time they have teamed up, as in September it was revealed Oracle databases were being made available on Microsoft’s Azure cloud platform.
Oracle and Microsoft reach GPU agreement
Under the deal, Microsoft will use Oracle Cloud Infrastructure (OCI) AI infrastructure, along with AI infrastructure from Azure, for inferencing of AI models that are being optimised to power Microsoft Bing conversational searches daily.
It will use the Oracle Interconnect for Microsoft Azure function to access managed services like Azure Kubernetes Service. This, a statement said, will allow it “to orchestrate OCI Compute at massive scale to support the increasing demand for Bing conversational search”.
Microsoft gave Bing an AI make-over earlier this year after its multi-billion dollar investment in OpenAI, the developer behind ChatGPT, giving it what the company describes as a “conversational” interface, where users can pose questions in natural language. This requires powerful clusters of computing infrastructure that support the evaluation and analysis of search results that are conducted by Bing’s inference model.
Karan Batta, senior vice president of Oracle Cloud Infrastructure, said: “Generative AI is a monumental technological leap and Oracle is enabling Microsoft and thousands of other businesses to build and run new products with our OCI AI capabilities.
“By furthering our collaboration with Microsoft, we are able to help bring new experiences to more people around the world.”
Deal reflects high GPU demand
Inference models such as Bing’s use require thousands of compute and storage instances and tens of thousands of GPUs that can operate in parallel as a single supercomputer over a multi-terabit network.
Microsoft will be able to access OCI superclusters of chips, which can apparently contain up to 32,768 Nvidia A100 GPUs or 16,384 of the company’s more advanced H100 model, as well as petabytes of high-performance clustered file system storage to efficiently process massively parallel applications.
Nvidia’s GPUs have been used to power the AI revolution that has taken place over the past year, and as a result, those buying the chips face potentially long lead times. The vendor says it will manufacture 550,000 H100s this year and hopes to triple its capacity next year, but in August the FT reported that orders placed now will not be delivered until 2024.
With this in mind, it is perhaps no surprise that Microsoft is looking to partner with other cloud providers to ensure it has enough processing power for its many AI services.
Divya Kumar, global head of marketing for search and AI at Microsoft, said: “Our collaboration with Oracle and use of Oracle Cloud Infrastructure along with our Microsoft Azure AI infrastructure, will expand access to customers and improve the speed of many of our search results.”