Janction

Janction is an AI-focused distributed computing network and GPU marketplace incubated by Jasmy Corporation. It is designed to pool computing resources and coordinate AI models, data, and computing tasks through smart contracts, with verification mechanisms intended to support contribution-based rewards. Janction has been described as a Layer 2 platform, while its product offering has also developed around GPU rental and distributed rendering services.^[1][2][3]

Overview

Janction’s design combines a computing-power network, data services, and AI models. Smart contracts are intended to automate and coordinate machine-learning and AI processes, bringing models, GPU capacity, data feeds, and data labeling into shared workflows. Janction describes the system as decoupling data, computing power, and models so that tasks and resources can operate in separate but connected processes.^[2][5][9]

The project’s data-service design covers data from on-chain sources, web content, and Internet of Things (IoT) devices. Its described processing flows include crawling, storage, indexing, caching, and querying. Janction also identifies data acquisition and preprocessing as part of its Layer 2 design, alongside methods for compensating data providers.^[9][5]

For computing tasks, Janction’s cluster GPU pool is designed to combine and allocate multiple GPUs for parallel workloads. Its stated engineering concerns include task scheduling, load balancing, security and isolation, fault recovery, cost efficiency, and consistency. The project describes a colocation and routing layer for matching GPU clusters to different workloads, as well as containerization to standardize and manage network resources. Its platform overview names Kademlia and Kubernetes as technologies used in node and container management and resource processing.^[2][5][9][1]

Janction identifies proof of workload as a means of checking whether promised computation has been performed and supporting payment in a trust-minimized network. The system is intended to verify contributions and distribute revenue among participants such as computing providers, data providers, data annotators, and model providers. Its documentation describes a workload-correlation function and the Vickrey–Clarke–Groves (VCG) mechanism as part of pricing computing contributions and coordinating incentives.^[5][1]

The design also addresses privacy and data segregation in training and inference. Janction describes its data and computing services as supporting collaboration without creating data silos, while its Jasmy-related infrastructure is presented as intended to support privacy-preserving data handling. Its technical discussion of distributed deep-learning training covers data, tensor, and pipeline parallelism, which require communication among components as work is divided across computing resources.^[5][1]

History and Development

Janction is an incubation project of Jasmy Inc. JasmyLab Inc., a Jasmy-related company, announced the launch of JANCTION GPU POOL on March 25, 2025, as a decentralized GPU cloud service. At launch, the service offered hourly rentals of NVIDIA H100 SXM GPUs with 80 GB of memory for ¥330 per hour, including tax, and also offered monthly contracts. It supported workloads using Blender, Unreal Engine, PyTorch, and TensorFlow, and was presented for uses including AI training and inference, rendering, game development, and data analytics.^[4]​

Janction’s documentation described a broader plan to develop an AI-collaboration Layer 2, with an initial emphasis on data acquisition and preprocessing, a GPU-computing marketplace, and model inference. On February 14, 2025, JasmyLab announced the launch of Janction’s Layer 2 testnet and said the network used Optimism’s OP Stack.^[11] A Gate.com account separately placed the testnet launch in March 2025, differing from JasmyLab’s February 14 announcement. Gate described mainnet deployment as planned for the first quarter of 2026; that target has passed, and the cited report did not confirm mainnet deployment.^[5]​^[6]​

On May 27, 2025, Janction published an article mapping the emerging AI decentralized physical infrastructure network (DePIN) stack. It described the hardware-aggregation layer as pooling underused GPUs and other computing resources from data centers, miners, and consumer devices, and identified io.net, Gensyn, and Akash as projects operating in that area.^[7]​

In September 2025, Janction’s global account described its Layer 2 concept as rebranded Jasmy Layer 2, according to Phemex. Janction documentation and other project descriptions continued to present the system in terms of Layer 2 infrastructure and decentralized AI services.^[8]​^[2]​

Binance Alpha began JCT spot trading on November 10, 2025, and Binance Futures launched a JCTUSDT perpetual contract the same day, with leverage of up to 40 times.^[1]​

In late August 2026, Jasmy and Panasonic Advanced Technology announced work to bring Smart Render by JANCTION to studios. The companies reported an internal benchmark in which a 4K, 30-frame Blender animation was rendered across 12 nodes in 15 minutes, compared with about four hours and 10 minutes on one GPU. They characterized the result as 16.7 times faster and 94% less rendering time. The announcement said the service would use prepaid credits priced in yen and initially rely on vetted GPU providers in Japanese data centers.^[8]​

GPU Marketplace and Services

The Janction GPU Marketplace is designed to aggregate capacity from multiple sources, including data centers and individual hardware owners, so AI teams and developers can access pooled computing power. Janction describes both on-demand rentals and resource-sharing arrangements. A November 2025 account framed the planned service components as AI-as-a-Service, GPU-market aggregation, and DeFi-linked incentives for network participants.^[2]​^[1]​^[10]​

Janction’s service tiers distinguish individual, business, and institutional use. Its Standard offering is based on GPU capacity equivalent to an NVIDIA RTX 4090. The Business offering adds secure storage, encryption, tamper detection, traceability, Japan-region service, and GPU-resource collaboration. Its Enterprise offering is aimed at organizations, schools, and research institutions and can include multiple licenses, H200 or B200 devices, on-premises environments, and equipment leasing.^[9]​

A newer product, Smart Render by JANCTION, distributes Blender animation rendering frame by frame across multiple GPUs and assembles the completed frames. The announced yen-credit payment model means studio customers may be able to use the rendering service without a cryptocurrency wallet or direct use of JCT. The announced initial deployment was to use vetted GPU providers in Japanese data centers before expanding.^[8]​

Janction has also described potential workloads for its GPU network including text-to-image generation, speech conversion, video enhancement, object detection, and image analysis. Examples given include 4K and 8K upscaling, noise reduction, film restoration, industrial defect detection, and private hosting of large language models for organizations in regulated sectors.^[1]​^[6]​

Participation and Adoption

Janction’s proposed network participants include providers of GPUs, data, annotations, and AI models. GPU owners contribute computing capacity, while developers and organizations use pooled resources for AI and rendering tasks. The project has described aggregation of idle computing resources from personal computers and data centers; a Gate.com account also discussed consumer hardware, including PlayStation 5 consoles, as a possible source and reported discussions about a potential PlayStation-related arrangement. The account did not describe the arrangement as an established partnership.^[5]​^[1]​^[6]​

​Gate.com described two modes of node participation: self-hosted nodes for technically capable GPU owners and managed nodes deployed and maintained in data centers. It reported that the managed-node model involved a one-time fee for an NFT-based license and that sales in Japan exceeded ¥600 million in the model’s first four months. The account also reported that participant earnings varied with task availability and GPU utilization.^[6]​

JCT Token and Governance

JCT is Janction’s native token. Its stated functions include payments for computing tasks, rewards for GPU and data providers and node operators, and participation in protocol governance. Janction’s token-utility materials also describe GPU-provider staking for veJCT, a non-transferable credential granting bidding rights and priority for jobs; compute users may pay in JCT or stablecoins, and veJCT holders may vote on marketplace settings.^[1]​^[8]​

JCT has a stated total supply of 50 billion tokens. At launch, about 6.35 billion JCT, or 12.7% of the supply, was described as initially circulating and unlocked for liquidity, community, and airdrop purposes; full circulation was expected by 2028. A Gate.com account described team, foundation, and investor tokens as subject to vesting, including an 18-month cliff followed by three years of linear unlocking for team tokens and a one-year cliff followed by two years of linear unlocking for foundation tokens. It identified 2027–2028 as a period of potential token-unlock pressure.^[1]​^[6]​

​Phemex reported that JCT was designed to exist on Ethereum and BNB Smart Chain, with a planned allocation of 45 billion tokens to Ethereum and 5 billion to BNB Smart Chain. Under the described cross-chain design, tokens transferred between chains are burned on one chain and minted on the other.^[8]​