Tether, the world’s largest stablecoin issuer, has entered the large language model (LLM) arms race with the launch of QVAC Fabric LLM. Announced on December 2, the system allows for full LLM execution, LoRA training, and instruction-tuning directly on laptops, desktops, consumer GPUs, and even smartphones. Tether states its objective is to decentralize access to AI development and eliminate the privacy trade-offs associated with cloud-based models.
Shift Away from Cloud Dependence
High-performance LLM inference has traditionally required expensive servers or NVIDIA-grade infrastructure. QVAC Fabric LLM aims to challenge these assumptions. The company indicated that this release removes vendor lock-in and empowers organizations to build and deploy AI systems using their existing hardware. This includes support for GPUs from AMD, Intel, NVIDIA, and Apple Silicon, as well as mobile GPUs such as Qualcomm’s Adreno and ARM Mali.
Tether just open-sourced its edge-first generalized LLM inference and LoRA fine-tuning framework for heterogeneous GPUs.
— Paolo Ardoino 🤖 (@paoloardoino) December 2, 2025
AI ubiquity era is beginning. https://t.co/e48XLZo6GI
A significant advancement is the capability to fine-tune models directly on smartphone-class hardware. Tether explained that this enables “personalized AI that can learn directly from users,” thereby preserving privacy and facilitating offline operation in areas with limited connectivity. QVAC Fabric LLM is available under the Apache 2.0 license, accompanied by multi-platform binaries and prebuilt adapters on Hugging Face. Tether stated that developers can commence fine-tuning “in only a few commands,” making customization accessible to a considerably wider audience.
Tether Faces 'Insolvency' Doubts and S&P Downgrades
The launch of the LLM system occurs amidst ongoing questions regarding the company's financial stability. Billionaire and BitMEX co-founder Arthur Hayes recently contended that the USDT issuer is “in the early innings of running a massive interest rate trade.” Hayes referenced Tether’s latest attestation, asserting that its reserves now have a greater allocation towards Bitcoin (BTC) and gold. He believes these assets could face downward pressure if interest rates decrease, and suggested that a reduction of approximately 30% in these holdings “would wipe out their equity,” rendering USDT “theoretically insolvent.”
This critique follows closely behind S&P Global Ratings’ downgrade of Tether to “5 (weak)” on its five-point stablecoin risk scale on November 26. The agency cited insufficient transparency and what it described as an increasing exposure to higher-risk investments, including Bitcoin, gold, corporate debt, secured loans, and other instruments that introduce credit, market, foreign exchange, and interest-rate risks. Tether CEO Paolo Ardoino has publicly refuted both the downgrade and Hayes’ assertions, criticizing what he termed inaccurate assumptions about the company’s reserves.

