In a striking departure from its core business as a stablecoin issuer, Tether announced the release of QVAC SDK, a fully open-source framework designed to democratize on-device artificial intelligence and free developers from cloud-dependent AI providers. The move signals a fundamental shift in how the $186 billion cryptocurrency company plans to deploy the $10 billion it earned in 2025.
QVAC, which stands for Quantized Verified Agentic Compute, is a cross-platform software development kit that enables machine learning models to run directly on smartphones, laptops, tablets, and embedded devices, without requiring cloud connectivity, GPU servers, or dependency on OpenAI, Google, or other centralized AI providers. The framework is available under a permissive open-source license on GitHub and targets developers who want to build AI applications with privacy, sovereignty, and independence from corporate data centers.
Why a Stablecoin Company Built an AI Platform
At first glance, Tether's foray into AI infrastructure seems incongruous. The company's core business, issuing USDT, the world's most widely circulated stablecoin, operates as a decentralized financial rail, holding reserves in U.S. Treasuries (now $141 billion) and leveraging blockchain networks for instant settlement. Yet Tether's CEO Paolo Ardoino has signaled this expansion reflects a deliberate strategic vision.
"The world is approaching a moment where billions of humans share the planet with billions of autonomous machines and trillions of AI agents," Ardoino said in announcing QVAC. "The current model, routing every decision through a centralized server, won't scale to meet that reality."
Tether's logic is straightforward: if AI agents will eventually outnumber humans and execute trillions of transactions daily, those agents will need programmable money. USDT is already global, interoperable across multiple blockchains, and supported by the company's infrastructure. QVAC provides the AI runtime to orchestrate those agents locally, moving computation to the edge rather than concentrating it in cloud provider data centers.
In essence, Tether is betting that the future of AI requires both local intelligence (QVAC) and global payments (USDT). The company has begun channeling a portion of its record $10 billion annual profit into AI, renewable energy, and decentralized communication platforms.
Technical Ambitions: From Phones to 13B Parameters
QVAC's engineering team has achieved several notable milestones that demonstrate the viability of running large language models on consumer devices. The SDK leverages QVAC Fabric, a fork of the popular llama.cpp framework, maintaining compatibility with the llama.cpp model ecosystem while adding proprietary optimizations.
The most striking achievement involves Microsoft's BitNet 1.58-bit quantization, a radical compression technique that reduces model weights to ternary values, cutting memory footprint by over 70% compared to standard 8-bit quantization. Using this approach, Tether engineers have successfully fine-tuned BitNet models up to 13 billion parameters on iPhone 16 devices. A 1-billion-parameter BitNet model can be fine-tuned on a Samsung Galaxy S25 in approximately 1 hour and 18 minutes, a feat that would have been impossible on consumer hardware just months ago.
“The world is approaching a moment where billions of humans share the planet with billions of autonomous machines and trillions of AI agents. The current model, routing every decision through a centralized server, won't scale to meet that reality.”— Paolo Ardoino, CEO, Tether
The SDK supports multiple hardware backends including Apple Silicon, NVIDIA GPUs, and AMD processors, enabling developers to write once and run across heterogeneous devices. QVAC also bundles tools for speech recognition (whisper.cpp), machine translation (Bergamot), and embeddings, providing a unified toolkit rather than requiring developers to integrate disparate libraries.
Positioning Against Cloud AI Monopolies
QVAC enters a crowded ecosystem of on-device AI frameworks. Competitors include llama.cpp (the lightweight inference engine powering much of the local AI movement), Ollama (which now integrates Apple's MLX framework for optimized Mac inference), MLX (Apple's machine learning framework), and Apple Intelligence (Apple's proprietary 3-billion-parameter foundation model built into iPhones and Macs).
Yet QVAC differentiates itself in three ways. First, it emphasizes decentralized model distribution, allowing models to be served peer-to-peer without centralized infrastructure. Second, it is built on BitNet, supporting extreme quantization that competitors don't yet offer at scale. Third, it includes an integrated agent runtime, positioning QVAC not just as an inference engine but as a platform for deploying autonomous AI agents on consumer devices.
Apple Intelligence, while impressive, remains proprietary and constrained to Apple devices. Ollama and llama.cpp are excellent inference tools but lack the agent orchestration layer and decentralized distribution primitives that Tether is adding.
The Broader AI Sovereignty Movement
Tether's push for decentralized, on-device AI reflects a growing skepticism toward centralized cloud AI providers. Ardoino has stated that Tether believes AI should empower the next wave of growth for society and humanity, not delegate even more control to corporations that own servers and access keys. This philosophy, sometimes called AI sovereignty, argues that just as blockchains distributed financial infrastructure, local AI should distribute machine intelligence.
QVAC's GitHub repository and permissive licensing suggest Tether views this not as a proprietary product but as infrastructure for an ecosystem. The company is betting that if developers build with QVAC, they will settle transactions in USDT, creating a synergistic flywheel: more local AI adoption drives stablecoin utility, and more stablecoin integration makes QVAC more attractive to developers.
Implications for Developers and Users
For developers, QVAC removes the cost of deploying cloud infrastructure for AI features. For users, it means AI services that function without internet connectivity, don't leak data to remote servers, and can operate on devices they own without permission from platform providers. For enterprises, it offers a path to deploy AI without relying on OpenAI, Google Cloud, or AWS.
The release of QVAC SDK represents Tether's largest bet on AI infrastructure to date. With $6.3 billion in excess reserves and a track record of profitable operations, the company has the financial capacity to sustain a multi-year investment in developer adoption and model optimization.