Researchers at the Technical University of Munich (TUM) have developed a groundbreaking AI chip that operates independently of cloud servers or internet connections.
Named the AI Pro, this innovative chip is designed by Professor Hussam Amrouch and is modeled after the human brain, allowing it to perform calculations locally and securely.
Unlike traditional AI chips that require constant data transfers to and from the cloud, the AI Pro completes all its processing on-site, ensuring maximum privacy and efficiency.
This revolutionary approach not only enhances cybersecurity but also significantly reduces energy consumption.
The first prototypes of the AI Pro have already been produced by the semiconductor manufacturer Global Foundries in Dresden.
What makes this chip truly unique is its neuromorphic architecture. Unlike conventional chips where computing and memory are separate, the AI Pro integrates both into one structure.
This design is inspired by how the human brain processes information—by recognizing patterns and similarities instead of relying on massive amounts of data.
For example, while traditional AI models require thousands of car images to learn what a car looks like, the AI Pro simply understands key features, such as four wheels and the ability to drive on roads.
This method, known as “hyperdimensional computing,” mimics human intuition, allowing the chip to make intelligent decisions with far less data.
One of the standout features of the AI Pro is its energy efficiency. In tests, the chip consumed just 24 microjoules to complete a sample task, which is up to 100 times less energy than similar chips.
This efficiency is a significant achievement in the world of AI technology, where energy consumption is a major concern.
“This mix of modern processor architecture, algorithm specialization, and innovative data processing makes the AI chip something special,” says Professor Amrouch. His team is confident that the AI Pro’s localized processing power and low energy usage set it apart from industry giants like NVIDIA, whose chips rely heavily on cloud data for processing.
Amrouch also highlights the importance of customized solutions. While NVIDIA’s platform is designed to tackle a wide range of tasks through the cloud, the AI Pro is tailored for specific applications.
This makes it incredibly efficient for tasks like processing health data from smartwatches or navigation data from drones.
Because the chip processes information directly on the device, there is no need to send sensitive personal data over the internet, eliminating risks associated with cybersecurity and unstable network connections.
This local processing not only improves privacy but also reduces dependence on internet infrastructure, making it ideal for remote or secure environments.
The AI Pro is currently under review for the European Solid-State Electronics Research Conference (ESSERC 2025), where it is expected to draw considerable attention.
Despite its relatively small size of one square millimeter and its cost of 30,000 euros, the chip represents a new direction for AI technology—one where processing happens locally, securely, and with minimal energy use.
According to Professor Amrouch, this could be the beginning of a shift toward hardware-based AI solutions, where data remains in the hands of the user, not in distant cloud servers. “The future belongs to the people who own the hardware,” he states confidently.
This innovation not only promises enhanced privacy and energy savings but also marks a significant step forward in AI technology, opening the door to more sustainable and secure applications across multiple industries.
