How Tether's Paolo Ardoino Is Betting on Local AI to Close the Intelligence Gap

The same morning Paolo Ardoino joined DevNTell, Tether published MedPsy, a medical AI model with 4 billion parameters that the company says outperforms Google's MedGemma, a 27-billion parameter model, on benchmark evaluations. The model is small enough to run on a smartphone. It is the kind of announcement that makes more sense once you understand what Ardoino, CEO of Tether, is actually trying to build.

Intelligence Inclusion

Tether is best known for USDT, the stablecoin it launched in 2014. Today USDT is used by 570 million people, many of them in countries where local currencies have shed most of their value. Ardoino described the Turkish Lira losing 80% of its value against the US dollar in five years and the Argentinian Peso losing 94% over the same period. For people living inside those economies, USDT became what Ardoino called a lifeline.

The logic behind QVAC follows a similar line of reasoning. More than 4 billion people live in countries with high inflation or without access to basic financial services. If AI tools require paid subscriptions to services like Anthropic or OpenAI, the people who can afford them will get substantially smarter while the people who cannot will fall further behind. Ardoino put it directly:

"If half of the population of the world that has access to financial services plus has access to AI because they can pay for Anthropic or OpenAI subscription, this part of the population will become 100 times smarter. And on the other side, you have the half of the population that does not have access to basic financial services and cannot afford to pay for an Anthropic or OpenAI subscription. The gap will become not twice bigger, but will become 100 times bigger."

The answer, in Ardoino's framing, is open-source, local-first AI that runs without a cloud subscription and without permission from any centralized provider.

"The Last Question"

Ardoino traces the philosophical foundation of QVAC to Isaac Asimov. He cited a 14-page short story called "The Last Question" as one of his central touchstones. In the story, humanity builds progressively more capable AI across billions of years, and at every stage asks the same question: how can entropy be reversed? Ardoino described the story as combining "philosophy, science, physics, AI, religion and a lot more" in a very small number of pages.

His reading of the story led him to a more grounded question: how do you stop society from dying? USDT, in his analysis, addressed one version of that by giving people in collapsing currency environments access to a stable store of value. QVAC is meant to address another version, by giving people access to AI tools regardless of their income or geography.

This kind of long-range thinking sits alongside a more immediate engineering philosophy that Ardoino developed earlier in his career. Before joining Tether, he worked on communication systems designed for battlefield and distress situations. That experience changed how he thought about what software is for. "Whenever I still today design anything in technology, I always think how technology should be designed to support humans in their worst moments, not in their best moments," he said. "It's easy to design technology that works perfectly when everything is fine, but technology should be the thing that helps us humans in the moments where everything goes to hell."

QVAC Fabric

The technical foundation of QVAC is an inference engine called Fabric, which Tether forked from Llama.cpp. Fabric adds capabilities that the upstream project does not include, among them support for Microsoft's BitNet 1-bit models on consumer hardware. The original BitNet implementation from Microsoft targeted high-end Nvidia GPUs, but Tether adjusted it to run on Snapdragon GPUs found in Android phones and on Apple iOS devices.

On top of Fabric, Tether built the QVAC SDK, which exposes a unified interface for a range of AI tasks: text chat, text-to-speech, speech-to-text, OCR, image generation with diffusion, and general model fine-tuning. The SDK has a JavaScript interface, which means web developers can integrate it without switching languages. Ardoino noted that Tether demonstrated fine-tuning of 1 and 3-billion parameter models directly on an iPhone 16, a result he described as "truly unprecedented."

The practical consequence is that a developer building an application can run AI inference locally on the device without sending data to an external server. This reduces latency and keeps user data on the device. Ardoino used the example of a car that needs to brake based on what is directly in front of it: the decision should not depend on a round trip to a data center. MedPSI is an early demonstration of what the Fabric infrastructure can produce. Tether is also working on translation models for African languages, with the stated goal of providing free education in villages the company already reaches through USDT distribution.

Agents Will Not Have a Morgan Stanley Account

The connection between Tether's AI work and its stablecoin business becomes clearer when Ardoino describes what he expects AI agents to need. Autonomous agents executing complex tasks will need to transact, and they will not be opening traditional bank accounts to do it.

"If we believe in a future where you have 10 billion humans, 10 billion machines or robots, and a trillion of AI agents, these AI agents will not have a Morgan Stanley account. They will not probably use PayPal. They will need programmable money because agents are all about programmability, right? They are all about executing complex actions."

USDT already reaches hundreds of millions of users and services, which means an AI agent using USDT can interact with a large existing network without additional infrastructure. Ardoino sees the two product lines converging around this eventually but only time will tell