A U.S. company has created a tiny supercomputer that fits in a pocket but runs massive AI models. The device can handle complex tasks like problem solving and strategic planning without needing an internet connection.
Tiiny AI calls its invention the AI Pocket Lab. It measures just 5.59 by 3.15 inches, roughly the size of a small notebook. Despite its size, it can run AI models with 120 billion parameters, which is about three times larger than the system behind the original GPT-3.
Most devices this size cannot handle advanced AI because the work usually requires data center servers. The Pocket Lab solves that by doing everything locally. Users get coding help, document analysis, and multi-step reasoning without sending data to the cloud.
READ ALSO: https://modernmechanics24.com/post/byd-unveils-electric-car-rival-porsche/
The system runs on a 12-core ARM processor, the kind found in phones and tablets. But it packs 80 GB of LPDDR5X RAM, far more than typical laptops which have between 8 and 32 GB. A large portion, 48 GB, is reserved just for the neural processing unit that handles AI work.
The device delivers 190 TOPS of computing power, which stands for trillions of operations per second. This qualifies it as a supercomputer because it runs workloads that normally need multiple graphics cards in a data center.
Current models that run on the Pocket Lab include GPT-OSS 120B, large Phi models, and high-parameter versions of Meta’s Llama family. The company says this represents a push toward edge computing, where AI happens on local devices instead of remote servers.
The device is smaller than Nvidia’s recently announced Project Digits mini PC. While it does not match that system’s raw power, it takes miniaturization a step further by fitting true AI capability into a pocket-sized box.
WATCH ALSO: https://modernmechanics24.com/post/ehang-next-gen-evtol-first-public-flight/
There are limits. The Pocket Lab cannot compete with the world’s biggest supercomputers. But for running advanced AI locally, it shows how fast the technology is shrinking. As chips improve, pocket-sized AI could become common for professionals who need private, offline access to powerful models.
