Microsoft Rents Bing Chat GPU Power From Oracle

Microsoft has agreed to cooperate with the software company Oracle, in which the Redmond-based company rents capacity for calculating AI tasks on the servers of the actually competing company. You make a pact with the devil, so to speak.

Microsoft Bing Chat Is GPU Power Hungry

Apparently, Microsoft is having problems meeting demand for its own artificial intelligence services with its own server capacity. The software giant has therefore announced that it also wants to use competitor Oracle’s cloud infrastructure in order to offer Microsoft AI services.

They want to use the Oracle cloud and Microsoft’s own Azure AI infrastructure in parallel to improve access options for customers in order to increase the speed of processing many search queries, explained Microsoft’s head of the search and AI team, Divya Kumar, according to The Register colleagues in a statement.

Oracle Have It

Oracle operates its own “supercluster” of numerous servers with Nvidia A100 and H100 graphics processors, which are apparently available for “rental”. It is still unclear whether Oracle cannot use these capacities itself or whether it has no other customers for it. In any case, this gives Microsoft the opportunity to avoid possible bottlenecks in the availability of its “explosively growing” AI services.

In their cooperation, Microsoft and Oracle use the so-called Oracle Interconnect for Microsoft Azure, with which services from the Azure cloud can access resources from the Oracle Cloud Infrastructure (OCI). The main thing is to let the AI ​​models behind Windows Copilot and Bing Chat run on a large scale as part of so-called inference in order to process the increasing number of user requests.

As is well known, AI is playing an increasingly important role for Microsoft, as the company wants to upgrade all of its products with corresponding functions. It was recently announced that Windows Copilot would also be made available for Windows 10 and thus for a billion more potential users. This of course increases the need for AI capacities accordingly…