A US deep tech startup has unveiled what it claims is a major shift in how powerful artificial intelligence can be deployed. Tiiny AI has introduced the Tiiny AI Pocket Lab, a pocket sized device officially verified by Guinness World Records as the world’s smallest personal AI supercomputer.
Roughly the size of a power bank, the Pocket Lab is capable of running large language models with up to 120 billion parameters entirely on device. It does not require cloud access, external servers, or high end GPUs, marking a sharp departure from how advanced AI systems are typically deployed today.
According to Tiiny AI, the goal is to bring data center level AI performance directly to individuals. The company argues that reliance on cloud based AI has created challenges around sustainability, rising energy costs, data privacy, and centralized control. By moving intelligence off servers and into personal devices, the Pocket Lab aims to make advanced AI more accessible and secure.
“Cloud AI has brought remarkable progress, but it also created dependency, vulnerability, and sustainability challenges,” said Samar Bhoj, GTM Director of Tiiny AI, in the company’s official announcement. He added that the Pocket Lab is designed to make AI personal, private, and fully owned by the user.
The device is positioned as a general purpose personal AI system. It is intended for creators, developers, researchers, students, and professionals who need advanced reasoning, content generation, agent workflows, and deep context understanding without relying on an internet connection. All user data, documents, and preferences are stored locally with bank level encryption, offering stronger privacy protections than cloud based platforms.
Tiiny AI says the Pocket Lab is optimized to run models in the 10 billion to 100 billion parameter range, which it claims covers more than 80 percent of real world AI tasks. For more demanding workloads, the system can scale up to 120 billion parameter models, delivering performance comparable to GPT 4 class systems while remaining fully offline.
Under the hood, the Pocket Lab is powered by an ARMv9.2 12 core CPU with a 65 watt power capacity. Instead of relying on energy intensive GPUs, it uses two key technologies to achieve high performance. TurboSparse improves efficiency by activating only the neurons required for a given task, while Powerinfer, an open source inference engine, distributes workloads across the CPU and NPU to reduce power consumption without sacrificing capability.
Tiiny AI also emphasizes its commitment to an open ecosystem. The Pocket Lab supports one click installation of popular open source models such as Llama, Qwen, DeepSeek, Mistral, Phi, and GPT OSS, along with AI agent frameworks like ComfyUI, Flowise, OpenManus, and SillyTavern. Regular updates and over the air upgrades are planned, with broader availability and demonstrations expected at CES in January 2026.
If the device performs as advertised, the Tiiny AI Pocket Lab could signal a meaningful shift away from cloud dependent AI toward fully local, user controlled intelligence.
