Echo verifies claimed GPU capacity by combining fresh computational challenges with independent power measurements. We send unpredictable workloads to a data center, verify the resulting compute, and correlate it with the facility’s physical electrical response. This gives lenders and compute buyers independent evidence that the compute they are financing or purchasing is actually installed and operating where claimed. Our prototype demonstrates both legitimate local execution and simulated remote-compute fraud, creating a fast, repeatable proof-of-capacity layer for AI infrastructure. Long term we aim to make compute verifiable and trackable, in order to prevent leakages of american tech to our adversaries.
