BreakingPremium
Aperture Compute's new accelerator claims 3x performance per watt
The chip targets inference costs, the fastest-growing line item for companies deploying large models.
CompGenome WirePublished: Jun 22, 2026 at 10:15 AM Last updated 1 month ago 5 min read
Aperture Compute revealed its next-generation inference accelerator, claiming a 3.1x improvement in performance per watt over the previous generation. The company is betting that inference efficiency, not raw training power, will define the economics of AI at scale.
Data center operators have struggled with the energy footprint of serving large models to mi
This is a premium analysis
Upgrade to a Professional plan or higher to read the full piece and unlock all premium coverage.
Upgrade to read