Tiiny AI unveils pocket-sized supercomputer that runs 120B models without cloud dependency

Tiiny AI Pocket Lab achieves Guinness World Record certification

US deep-tech startup Tiiny AI has unveiled the Tiiny AI Pocket Lab, officially verified by Guinness World Records as the world's smallest personal AI supercomputer 1

1

. The pocket-sized device, measuring just 14.2 × 8 × 2.53 cm and weighing only 300 grams, resembles a power bank but delivers capabilities typically associated with server racks and high-end GPUs 3

3

. The certification recognizes the device under the category "The Smallest MiniPC (100B LLM Locally)," marking a significant milestone in portable AI computing 5

5

.

The AI supercomputer can run large language models locally with up to 120 billion parameters entirely on-device without relying on cloud connectivity, external servers, or discrete accelerators 1

1

. This capability addresses a fundamental shift in how AI processing occurs, moving computation from centralized data centers to individual devices. Tiiny AI launched the Pocket Lab on December 10 and plans to showcase it at CES 2026, though pricing and specific release details remain undisclosed 3

3

.

Addressing cloud dependency and privacy concerns through local execution

Tiiny AI positions the device as a direct response to growing concerns about cloud infrastructure dependency, sustainability challenges, and privacy risks in current AI systems 1

1

. Samar Bhoj, GTM Director of Tiiny AI, stated: "Cloud AI has brought remarkable progress, but it also created dependency, vulnerability, and sustainability challenges. With Tiiny AI Pocket Lab, we believe intelligence shouldn't belong to data centers, but to people" 2

2

. This philosophy drives the company's mission to reduce cloud dependency while making data-center-level power accessible to common users 1

1

.

The emphasis on local data storage and on-device AI processing removes network latency and limits external data exposure 2

2

. The device provides true long-term personal memory by storing user data, preferences, and documents locally with bank-level encryption, offering privacy and persistence that cloud-based AI systems cannot match 5

5

. This approach enables secure processing of sensitive information even without internet access, appealing to professionals, researchers, and developers who handle confidential data.

Technical specifications enable server-grade performance in compact form

The Tiiny AI Pocket Lab operates within a 65W power envelope, delivering energy-efficient personal intelligence at a fraction of the energy and carbon footprint of traditional GPU-based systems 5

5

. At its core sits a 12-core ARM v9.2 CPU paired with a discrete Neural Processing Unit (NPU) capable of delivering approximately 190 TOPS of compute 2

2

. The system includes 80GB of LPDDR5X memory alongside a 1TB SSD, with total power draw staying within the 65W system envelope 2

2

.

Tiiny AI claims consistent performance for Large Language Models (LLMs) in the 10B-100B parameter range, with support extending to 120B 2

2

. This upper limit approaches the capability of leading cloud systems, enabling advanced reasoning and extended context to run locally. The 80 gigabytes of LPDDR5X memory facilitates aggressive quantization techniques, compressing model data to lower precision levels while preserving essential computational accuracy 4

4

.

PowerInfer and TurboSparse drive efficiency gains

The device's performance relies on two proprietary software-driven optimizations rather than scaling raw silicon performance 2

2

. TurboSparse employs neuron-level sparse activation, selectively deactivating less critical neural pathways during inference to increase efficiency by reducing unnecessary computations 4

4

. This method maintains full model intelligence without quality degradation, improving inference efficiency without reducing model capabilities 3

3

.

PowerInfer functions as a heterogeneous inference engine that distributes AI workloads dynamically between the CPU and NPU, coordinating components to approach server-grade throughput at lower power 2

2

. This division optimizes resource use, delivering performance comparable to server environments at reduced power consumption levels suitable for portable devices 4

4

. The system includes no discrete GPU, with Tiiny AI arguing that careful scheduling removes the need for expensive accelerators.

Market positioning and accessibility considerations

Tiiny AI positions the Pocket Lab as a more accessible alternative to existing compact AI devices. Other small supercomputers such as NVIDIA's Project Digits, priced around $3,000, and the DGX Spark at $4,000, sit at price points that put them out of reach for most everyday users 3

3

. While pricing for the Pocket Lab remains undisclosed, the company targets widespread adoption among developers, researchers, creators, professionals, and students 5

5

.

The device supports popular open-source models including GPT-OSS, Llama, Qwen, DeepSeek, Mistral, and Phi 3

3

. Tiiny AI emphasizes an open-source ecosystem that supports one-click installation of major models and agent frameworks, with continuous updates planned 2

2

. The device enables multi-step reasoning, deep context understanding, agent workflows, content generation, and secure processing across diverse applications in research, robotics, and advanced reasoning tasks 5

5

.

Performance verification and future outlook

While Guinness World Records has certified the hardware for local 100B-class model execution, independent real-world performance data is not yet available 2

2

. Claims regarding capability, server-grade performance, and seamless scaling on such constrained hardware remain difficult to independently verify until the device reaches real users. The industry will be watching closely at CES 2026 to assess how this pocket-sized device performs against established benchmarks and whether it can deliver on its ambitious promises to enhance privacy while reducing energy costs and cloud infrastructure dependency.