Tiiny AI introduced Pocket Lab at CES 2026 in Las Vegas, a pocket-size personal computer built to run large language models entirely offline. The device needs no internet connection and no subscription, and the company says it can run models with up to 120 billion parameters locally, a class of model that until recently was confined to servers and data centers.
Guinness World Records certified the Pocket Lab in December 2025 as the world’s smallest mini PC capable of running a 100 billion parameter AI model locally, according to the company and coverage from Mashable. The device ships with 80GB of memory and 1TB of storage, and its size is intended to make local AI far more accessible than a desktop workstation or a rack server. Tiiny AI has published performance figures for supported open-weight models, though these are the company’s own measurements.
Privacy and cost as the pitch
The launch arrives as questions grow about data privacy, dependence on cloud providers and recurring subscription fees for AI services. Tiiny AI’s central argument is that a user’s data should stay under the user’s control, processed on the device rather than sent to an external server. That positioning is aimed at consumers, developers and small businesses looking for an alternative to cloud-based AI, and at anyone who wants to run open models such as those from Meta, Mistral or DeepSeek without a GPU server.
The company said Pocket Lab would launch on Kickstarter in February 2026 with a super early bird price of 1,399 dollars. The target buyer, in Tiiny AI’s framing, is someone who wants a powerful but private AI setup whose cost works out lower over time than paying monthly for hosted services.
Part of a move toward edge AI
Pocket Lab is one of several CES 2026 products pointing to the same shift: AI processing is moving toward edge devices and personal computing rather than remaining tied to data centers and constant connectivity. Whether a device this small can run very large models at usable speeds, and how it compares with alternatives such as Nvidia’s compact AI workstations or high-memory Macs, will become clearer once units reach independent reviewers. For now, the product is a concrete example of compute, privacy and control being packaged into a single small box.
