Tether Launches AI Framework for Smartphones and Consumer GPUs

Tether, issuer of the USDT stablecoin, has launched a new AI training framework that enables fine-tuning of large language models on consumer devices-including smartphones and non-Nvidia GPUs.

The system, part of Tether’s QVAC platform, leverages Microsoft’s BitNet architecture and LoRA techniques to slash memory and compute demands. According to the company, it reduces VRAM usage by up to 77.8% compared to standard 16-bit models.

The framework supports AMD, Intel, Apple Silicon, and mobile GPUs from Qualcomm and Apple. Tether engineers reportedly fine-tuned 1-billion-parameter models on smartphones in under two hours, with smaller models completing in minutes. It also enables on-device training and federated learning-updating models across distributed devices without centralized data uploads.

Inference performance sees significant gains too: mobile GPUs run BitNet models several times faster than CPUs, the company claims.

This move aligns with a broader trend of crypto firms pivoting into AI infrastructure, as Bitcoin miners and blockchain platforms increasingly deploy high-performance computing resources for machine learning workloads.