VMware has announced three expanded partnerships for its private AI service, with IBM Consulting and Intel all set for long-term collaboration on generative AI.
The partnership with Intel will bring Intel hardware and its software suite to VMware Cloud Foundation as well as fourth-generation Intel Xeon processors, which the firm stated could provide inference response times as low as 100m/s.
Through its IBM partnership, VMware will bring the IBM watsonx data and AI platform to VMware Cloud Foundation, in conjunction with Red Hat’s OpenShift container program.
“This is an important use case we’ve seen from a number of customers where they like what they have seen so far with watsonx, but they have this desire to bring the AI services available through watsonx to their data on-premises or wherever they do business.”
VMware aims to improve the availability of GPUs and CPU capacity for customer applications, and in a presentation highlighted the ability of its platform to intelligently provision compute resources to applications as and when they are needed as one of the benefits of running Intel’s offerings on its platform.
“Oftentimes organizations are having real challenges with being able to intelligently manage access to compute and the result for that is often cost overruns. So this is an area in which we can really shine.”
The announcement was made at VMware Explore Barcelona, the firm’s annual EU conference. It came alongside sweeping new commitments for data management to empower customers who want to implement new generative AI workloads.
VMware announced its new Private AI service in August at VMware Explore in , Las Vegas. The event saw VMware CEO Raghu Raghuram joined onstage by Jensen Huang, CEO at Nvidia, who explained that Nvidia, Dell, Lenovo, and HPE would provide 100 dedicated servers for VMware Private AI.
Discover how IBM watsonx.data supports managing data used for AI and analytics
DOWNLOAD NOW
At the time, Raghuram stated that firms such as Hugging Face and Anyscale would work with it on generative AI to expand options for customers and enable more secure operations.
Since then, the firm says that it has been able to realize lower AI running costs than alternatives and that performance on its hypervisor has ranged from at the level of bare metal to 5% better than this option.
While private AI workloads could help to alleviate c-suite concerns over data leaks connected to generative AI, leading voices in the space have warned that the cost of implementing AI models on-prem could be ‘eye watering’.
VMware’s proposition, which includes the data safety and fine-tuning benefits of running models on-prem using virtualized GPUs, could also help customers embrace a hybrid cloud approach to AI.
“A core challenge that we have seen has been centered around privacy and that's privacy and control of organizations, intellectual property, wanting data to remain oftentimes where it is created, processed, or consumed and having the ability to bring AI to their choice of where they would like to see data,” said Chris Wolf, VP, VMware AI Labs
“That also includes access controls, not just to the AI models but to the data that might be used for inferencing with the model or the data that's being used to train the model.”
