Tiiny AI Inc., a US deep-tech startup, has officially unveiled the Tiiny AI Pocket Lab, now verified by Guinness World Records as the “World’s Smallest MiniPC (100B LLM Locally).” The device, roughly the size of a small paperback, packs enough firepower to run a 120-billion-parameter large language model entirely on-device, no cloud, servers, or high-end GPUs required.
The Pocket Lab represents a paradigm shift in AI computing. While cloud-based AI dominates today, it carries growing sustainability concerns, rising energy costs, global outages, and privacy risks. Tiiny AI’s new approach puts personal intelligence back in the hands of users, emphasizing offline operation, privacy, and energy efficiency.
“Cloud AI has brought remarkable progress, but it also created dependency, vulnerability, and sustainability challenges,” said Samar Bhoj, GTM Director of Tiiny AI. “With Tiiny AI Pocket Lab, intelligence shouldn’t belong to data centers—it belongs to people.”
Pocket-Sized Powerhouse
Despite its compact form factor—14.2 × 8 × 2.53 cm and ~300g—the device delivers ≈190 TOPS of AI compute through a custom heterogeneous module combining SoC and dNPU. It features 80GB LPDDR5X memory and 1TB SSD storage, supporting nearly all major personal AI use cases: multi-step reasoning, deep context understanding, content generation, secure document processing, and agent workflows.
The Pocket Lab operates in the “golden zone” of personal AI—10B–100B parameters, covering over 80% of real-world tasks—but also scales up to 120B parameters, offering intelligence comparable to GPT-4o, including PhD-level reasoning and complex multi-step analysis. All computation happens offline, with user data stored locally and encrypted at bank-level standards.
TurboSparse and PowerInfer: Key Breakthroughs
Two proprietary innovations make the Pocket Lab possible:
- TurboSparse: A neuron-level sparse activation technique that improves inference efficiency without sacrificing model fidelity.
- PowerInfer: An open-source heterogeneous inference engine that dynamically distributes workloads across CPU and NPU, providing GPU-level performance at a fraction of the power consumption.
The device runs on 65W typical system power, far below traditional GPU setups, combining performance with energy-conscious design.
Ready-to-Use Open-Source Ecosystem
The Pocket Lab supports one-click installation of leading open-source models such as GPT-OSS, Llama, Qwen, DeepSeek, Mistral, and Phi, along with AI agents including OpenManus, ComfyUI, Flowise, Presenton, Libra, Bella, and SillyTavern. The ecosystem includes continuous updates and over-the-air hardware upgrades, making it ready for developers, researchers, and creators alike.
Market Implications
The launch comes as the global LLM market, valued at $7.4B in 2025, is projected to surge to $35.4B by 2030 (CAGR 36.9%), driven by enterprise adoption, domain-specific model development, and advances in self-supervised learning. Tiiny AI’s Pocket Lab offers a compelling alternative to cloud dependence, promising private, low-latency, energy-efficient AI for personal and professional users.
About Tiiny AI
Founded in 2024, Tiiny AI brings together engineers from MIT, Stanford, HKUST, SJTU, Intel, and Meta. Their work spans systems engineering, AI inference optimization, and hardware–software co-design, with research published in SOSP, OSDI, ASPLOS, and EuroSys. In 2025, the startup raised a multi-million-dollar seed round from top global investors. The Tiiny AI Pocket Lab will debut publicly at CES 2026.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
