Tether launches a dataset of 41 billion AI tokens to democratize STEM intelligence training

The world's largest stablecoin issuer is making a bold move to redesign artificial intelligence by putting computational power directly into the hands of users.

The company behind the USDT stablecoin of $150 billion announced on Friday the launch of QVAC Genesis I, a synthetic dataset of 41 billion tokens designed to train artificial intelligence models focused on science, technology, engineering, and mathematics.

The dataset represents the largest publicly available synthetic training resource ever created for AI development, marking a significant escalation in the company's diversification beyond digital payments.

The announcement comes alongside QVAC Workbench, an application for consumers that allows users to train, run, and interact with AI models directly on their personal devices - from smartphones to laptops - without relying on cloud infrastructure or corporate servers.

The dual launch reflects CEO Paolo Ardoino's vision of “returning intelligence to the people” by breaking the control that large tech companies currently have over computing and AI data.

Breaking free from cloud dependence

“Intelligence should not be centralized,” Ardoino said in a statement. “It should belong to the individual, not to the institution.”

Philosophy represents a direct challenge to the current paradigm of AI, where companies like OpenAI, Google, and Anthropic dominate the development of models through massive centralized data centers and cloud computing infrastructure. QVAC (QuantumVerse Automatic Computer), revealed for the first time in May 2025, aims to change this power dynamic by enabling AI to operate locally on user devices while maintaining privacy and autonomy.

The Genesis I dataset specifically targets reasoning and precision in scientific domains. According to the announcement, the 41 billion text tokens have been validated through educational and scientific standards in mathematics, physics, biology, and medicine - fields where obtaining accurate and specialized training data remains challenging.

The rise of synthetic data

Synthetic datasets have emerged as a critical solution to the AI data scarcity problem. Unlike real-world data collection, which can be costly, time-consuming, and fraught with privacy concerns, synthetic data is artificially generated to mimic statistical properties and patterns of authentic information while eliminating exposure of sensitive personal details.

Industry research suggests that 60% of AI project data will be synthetic according to current projections, driven by privacy regulations such as GDPR and the need for domain-specific training materials. High-quality synthetic data enables AI models to achieve 90-95% of the performance of models trained on real data while reducing acquisition costs by 60-80%, according to synthetic data generation platforms.

The Genesis I dataset joins this rapidly expanding ecosystem, but with a crucial difference: it is explicitly designed for STEM applications and is available as the first publicly accessible synthetic dataset built specifically for educational and scientific content.

Intelligence on the device: The QVAC workstation

The QVAC Workbench application complements the dataset by providing practical infrastructure for decentralized AI development. The application supports major open-source models including Meta's Llama, Google's Medgemma, Alibaba's Qwen, and OpenAI's Whisper - all while keeping data processing private and on-device.

“With QVAC Workbench and Genesis I, we are opening the door to infinite intelligence, an AI that lives, learns, and evolves locally on your own device,” Ardoino stated. “Whether it's a phone, a robot, or a wearable device, intelligence should belong to the individual, not to the institution.”

The modular architecture of the platform allows developers to build applications using small and composable components, while the peer-to-peer network facilitates direct device-to-device communication without centralized servers. This design theoretically enables the QVAC ecosystem to scale to trillions of agents and AI applications without single points of failure.

Integrated payments through the Wallet Development Kit (WDK) also allow AI agents to autonomously transact in Bitcoin and USDT, opening possibilities for self-sufficient AI systems that can trade independently.

Beyond Stablecoins: Aggressive Diversification

The AI initiative represents the latest chapter in the company's aggressive diversification strategy. The company, which reported $13 billion in profits during 2024, has systematically expanded beyond its core stablecoin business into energy, agriculture, media, and technology infrastructure.

Recent investments include a 10 million euro stake in the Italian media company Be Water, an investment of $775 million in the video platform Rumble, and a 70% stake in the agricultural company Adecoagro. The company also holds over $120 trillion in U.S. Treasury bonds, ranking as the 19th largest holder globally – surpassing Germany's holdings.

The company previously presented its open-source Wallet Development Kit last year, allowing humans, machines, and AI agents to build secure and self-custodied wallets and transact using USDT and Bitcoin. The strategic positioning places the company at the intersection of cryptocurrency infrastructure and decentralized AI development - two of the fastest-growing domains in the tech sector.

The competitive landscape

The movement comes at a time when the generation of synthetic data is projected to exceed $2 billion by 2027, with major players like Mostly AI, Synthesis AI, and DataGen leading the commercial development. Companies like Waymo and Tesla are already extensively using synthetic data to train autonomous vehicle systems in rare and dangerous scenarios that rarely appear in real-world datasets.

In healthcare, where patient data privacy creates substantial barriers for AI development, synthetic data offers solutions by providing simulated patient records that maintain statistical patterns without exposing real identities. Financial institutions have similarly adopted synthetic data to train fraud detection systems while maintaining compliance with privacy regulations.

The competitive advantage it provides lies in its integration with cryptocurrency infrastructure and its commitment to open-source development. The company plans to launch a comprehensive Software Development Kit in the coming months, making it easier for developers to build, deploy, and scale their own QVAC agents across any type of device.

Final reflections

“If you need an API key to use your AI, it's not really yours,” emphasized Ardoino, articulating the underlying philosophical basis for the QVAC initiative. The statement reflects growing concerns about centralized AI infrastructure, where a handful of corporations control both the computational resources and the data pipelines that power modern artificial intelligence.

By allowing AI models to operate entirely offline on personal devices, QVAC addresses fundamental privacy vulnerabilities inherent in cloud-based systems. This approach ensures that users retain complete control over their data, computing, and autonomy – values that closely align with the fundamental principles of cryptocurrency decentralization and self-sovereignty.

QVAC Genesis I and Workbench represent more than just new products; they constitute a philosophical statement about the future of artificial intelligence. While it remains to be seen whether the initiative can successfully challenge the dominance of Big Tech, the dataset of 41 billion tokens… Device infrastructure provides developers with unprecedented tools to build AI systems that prioritize user control over corporate convenience.

The era of “Infinite Intelligence,” as proposed by Ardoino, may have begun - and this time, it could truly belong to everyone.