IBM Research has upgraded its Vela AI Supercomputer, part of the IBM Cloud, to handle the growing demand for AI models. The company has doubled the capacity of Vela and plans to continue expanding it to meet the demands of training larger foundation models. IBM Research has shared details of the upgrades, which include increasing the density of GPUs and improving networking bandwidth. The company is also developing its own inference accelerator, the AIU, which is expected to provide more cost-effective processing infrastructure. IBM’s focus on AI has been successful, with the watsonx pipeline doubling in size over the past quarter and numerous AI projects in the pipeline. As the company deploys the AIU at scale, it will have a competitive advantage over other cloud providers in terms of efficiency and cost-effectiveness.