Empire AI Beta multiplies inference capacity 40x

New York brought the $40 million Empire AI Beta supercomputer online with 40 times Alpha’s inference capacity, 11 times its training capacity, and more than 300 research projects queued. For academic builders, the 10-member consortium is a public alternative to the private capacity market described in Nvidia’s research-compute investment model, but access, queue time, and usable accelerator hours—not the superlative—will determine whether the machine changes research velocity.