Tether launches AI training framework for smartphones and consumer GPUs

Part of the QVAC platform, the framework can use non-Nvidia hardware, expanding support beyond the dominant GPUs typically used for AI training.

Tether, issuer of the world’s largest stablecoin by market cap, USDT, has released a new AI training framework that it says allows large language models to be fine-tuned on consumer hardware, including smartphones and non-Nvidia GPUs.

According to Tuesday’s announcement, the system, part of its QVAC platform, uses Microsoft’s BitNet architecture and LoRA techniques to reduce memory and compute requirements, potentially lowering the cost and hardware barriers to developing AI models.

The framework supports cross-platform training and inference across a range of chips, including AMD, Intel and Apple Silicon, as well as mobile GPUs from Qualcomm and Apple.

Tether said its engineers were able to fine-tune models with up to 1 billion parameters on smartphones in under two hours, and smaller models in minutes, with support extending to models as large as 13 billion parameters on mobile devices.

Built on BitNet, a 1-bit model architecture, the framework can cut VRAM requirements by up to 77.8% compared with similar 16-bit models, according to the company, allowing larger models to run on limited hardware. It also enables LoRA fine-tuning on non-Nvidia hardware for 1-bit models, expanding support beyond the GPUs typically used for AI training.

The company said the performance gains extend to inference, with mobile GPUs running BitNet models several times faster than CPUs. It also pointed to use cases such as on-device training and federated learning, where models can be updated across distributed devices without sending data to centralized servers, potentially reducing reliance on cloud infrastructure.

Related: Messari’s new CEO is doubling down on AI as firm cuts staff

Tether’s move into AI infrastructure comes as crypto companies have been expanding into compute and machine learning, with activity accelerating across Bitcoin mining and the rise of AI agents.

In September, Google took a 5.4% stake in Cipher Mining as part of a $3 billion, 10-year deal tied to AI data center capacity. In December, Bitcoin miner IREN said it planned to raise about $3.6 billion to fund AI infrastructure.

The trend has continued into 2026. In February, HIVE Digital Technologies reported record revenue of $93.1 million, fueled by growth in its AI and high-performance computing (HPC) operations, while Core Scientific secured a $500 million loan facility from Morgan Stanley in March, with the option to expand it to $1 billion.

The mining sector’s pivot to AI and HPC comes as AI agents, autonomous programs that can transact, interact with services and execute tasks, are gaining momentum across the crypto sector.

In October, Coinbase introduced wallet infrastructure enabling AI agents to conduct onchain transactions. Last month, Alchemy launched a system allowing agents to access blockchain data services using USDC on Base. Also in February, Pantera and Franklin Templeton joined Arena, a platform from Sentient for testing enterprise AI agents.

On Tuesday, World, the identity network co-founded by OpenAI’s Sam Altman, launched AgentKit, a toolkit that allows AI agents to verify they are linked to a unique human using World ID capabilities while making payments via the x402 micropayments protocol.

Magazine: All 21 million Bitcoin is at risk from quantum computers

Source: CoinTelegraph