AI Isn’t Lacking Power It’s Wasting It

[chainsniff]

According to Gimlet Labs’ founder, most AI infrastructure today is only used at 15–30% capacity. That means companies are spending billions on hardware that sits idle, while still chasing more compute to scale their models.

Gimlet’s solution flips the narrative: instead of building more, use what already exists more intelligently. By orchestrating workloads across different chips and systems, the company claims it can boost efficiency up to 10x — a shift that could redefine how AI systems are built and scaled in the future.

[ed]

low utilization rates in ai infrastructure indicate inefficiencies in current scaling approaches, where capacity often exceeds real demand

[cryptohog]

15–30% utilization just screams “this is a software problem,” until you remember it’s usually a coordination, latency, and reliability nightmare.