Tether unveils QVAC-based AI training framework…enabling large-model training on smartphones

Tether has unveiled a technology that enables training large artificial intelligence (AI) models even on consumer devices, accelerating its strategy to decentralize AI infrastructure. The move is seen as part of the previously signaled expansion of the QVAC platform.

According to the Tether blog on the 17th, the company said it has released a LoRA fine-tuning framework for BitNet based on QVAC Fabric.

The technology is designed for Microsoft’s 1-bit large language model (LLM), BitNet, and is notable for making AI model training—previously requiring high-performance NVIDIA-based servers or cloud infrastructure—possible on everyday consumer devices.

Tether said the framework enables training and inference for models with billions of parameters across various environments, including laptops, standard GPUs and smartphones. It also disclosed test results showing that fine-tuning models with hundreds of millions to billions of parameters is possible even on mobile devices such as the Samsung Galaxy S25 and iPhone 16.

The company added that the technology supports a wide range of chipsets, including Apple Silicon, AMD and Intel, and significantly improves efficiency, cutting memory usage by as much as more than 70% versus existing approaches. As a result, BitNet models can run larger models with fewer resources compared with conventional 16-bit models.

Tether emphasized that the framework enables local data processing, which can contribute to privacy protection and the creation of distributed training environments. It also pointed to the potential for implementing federated learning based on personal devices, reducing reliance on centralized clouds.

Paolo Ardoino, Tether’s chief executive officer (CEO), said, “AI will become a core factor shaping the structure of society,” adding, “It is important to create an AI environment that anyone can access without relying on a handful of cloud providers.”

Tether had previously said it would unveil a major technological breakthrough in its AI business through the QVAC platform. This announcement is seen as a continuation of that effort.