manifesto

← Back to Homepage

JoinHive🐝, the collective AI network where a global swarm of AI workers collectively host large AI models and MCP servers to power your AI apps, Agents and daily workflows.

JoinHive allows developers to tremendously save on AI costs compared to OpenAI or other traditional providers by utilising existing hardware and idle compute around the world. Turn your PC into passive income by joining the swarm, run small parts of AI models or other AI tools and help us create a become part of the collective AI network that makes AI more affordable and accessible for thousands of developers around the world.

Our Vision

To build the world's largest, most accesible, and cost-effective collective, global AI swarm. We are creating the distributed foundations for the next era of AI, with a relentless focus on user experience to escape the crypto bubble and capture the attention of millions of non-crypto natives.

Our Team

We're a seasoned team of Web3 infrastructure veterans from industry-leading projects like Livepeer, Gelato, and The Graph, committed to delivering real utility, genuine innovation, and meaningful adoption in the rapidly evolving intersection of crypto and AI.

Read the Technical Paper

Dive into the technical underpinnings of the Hive network—including our architecture, incentive mechanisms, and roadmap.

The AI compute bottleneck

Artificial intelligence is advancing at breakneck speed, but a critical bottleneck threatens to stall progress: compute scarcity.

"AI compute needs are easily a hundred times more than anticipated a year ago"

— Jensen Huang, Nvidia CEO
March 2025, GTC Keynote

The amount of AI users has grown significantly as more people get over their initial skepticism. But more importantly, we are using AI magnitudes more effectively than just a year ago. Reasoning capabilities are creating magnificent results, but also consume vastly more resources.

This exploding demand crashes against some stark realities today:

Astronomical costs: Renting dedicated GPU power at traditional cloud providers is extremely expensive, making cutting-edge AI prohibitively expensive for most.
Inflexible access: Restrictive APIs, rate limits, and opaque "black boxes" prevent developers from truly leveraging and fine-tuning models.
Supply chain constraints: Even for hyperscalers, acquiring enough specialized AI hardware is a massive, ongoing challenge.
Centralisation & control: Innovation is stifled by the walled gardens of a few tech giants, limiting access, customization, and permissionless development.

JoinHive: Unleash the swarm 🐝🐝🐝

JoinHive, the collaborative AI network where massive AI models operate across a global swarm of community-run nodes.

JoinHive taps into the vast, underutilized power of consumer GPUs globally. Our unique peer-to-peer model-sharding architecture intelligently splits massive AI models across many nodes and consumer-grade GPUs, allowing anyone to contribute compute and earn meaningful rewards.

For GPU owners ("workers")

Turn your idle PC into an income source, just like the early days of crypto mining
Earn HIVE tokens proportional to your work completed
Become an owner and stakeholder in the foundational infrastructure of distributed AI

For AI Builders

Inference and fine-tuning at radically lower costs (up to 90% cheaper)
No upfront hardware investments — integrate seamlessly via easy-to-use APIs
Unlock true customization with PEFT fine-tuning; own your weights

JoinHive isn't just another GPU marketplace; it's a collective compute organism, purpose-built to efficiently orchestrate distributed resources for the unique demands of AI.

Built for humans, not just experts

JoinHive is engineered with a relentless focus on simplicity and user experience. Unlike many DePIN projects demanding deep technical skill, our goal is effortless participation.

One-Click Worker GUI: Download, install, run. Our intuitive GUI (see screenshot 1) guides users easily. No complex configurations, command lines, or manual network setups such as port forwarding.
Accessible Earnings: A clear dashboard shows performance and earnings, with future plans for seamless fiat integration.

This pragmatic, product-led approach is key to unlocking the vast pool of consumer GPUs globally. By minimizing friction, we accelerate adoption, drive network effects, and build a truly accessible compute layer for the AI revolution.

Worker set-up GUI screenshot — Worker set-up GUI

JoinHive dashboard screenshot — Joinhive dashboard to track workers and earnings

Technology: Collective intelligence at scale

Distributed Model Execution: Large models (Llama, Deepseek, Mixtral, …) are split into layers; workers run only fractions of models, collaborating seamlessly.
Proof-of-Tensors: Real-time, efficient verification of inference and fine-tuning using local sensitivity hashing and multi-party compute techniques, execution == verification.
Optimized P2P Routing: Requests flow efficiently through the swarm, optimizing for throughput and minimizing latency.
Consumer GPU Optimized: Specialized quantization techniques reduce communication rounds and maximize performance on readily available hardware.
Accessible Fine-Tuning (PEFT): Enables powerful customization (LoRa etc.) without needing massive individual resources. Users manage small adapter weights; the swarm handles the base model.
Resilient & Fault-Tolerant: The network automatically adapts to nodes joining/leaving, ensuring service continuity.

Our optimizations are extremely effective, JoinHive is 15-20× faster than offloading large models locally.

Read the Technical Paper

Dive deeper into Hive's architecture, incentive mechanisms, and roadmap.

Comparison: Purpose-built for collective AI compute

vs. Centralized Cloud (AWS/GCP): Drastically lower cost (50-90%+ target savings), permissionless, avoids lock-in, leverages globally distributed consumer hardware.
vs. DePIN Marketplaces (Akash, Render): We are NOT a 1-to-1 rental marketplace. JoinHive is a coordinated, multi-party compute network specifically designed for sharding and running single large AI models across many nodes simultaneously. This collective approach is essential for massive models.
vs. Decentralized Dedicated Instances (Aethir, io.net): These focus on renting entire GPUs. JoinHive efficiently utilizes fractions of many GPUs (often consumer-grade) working together—you only pay for actual usage and not a dedicated instance.

Hive is unique in its approach to collectively run models with real-time verification

Tokenomics: Honey for the Bees 🍯

Our token model orchestrates and incentivises the JoinHive ecosystem, aligning incentives and ensuring sustainable growth

Two-fold utility

Worker staking (primary): Workers can stake tokens to improve network alignment. Higher stakes directly increase potential workload allocation and earnings for reliable, high-performance nodes. This secures the network and incentivizes quality service.
Compute Payments (secondary): End-users and applications pay for AI tasks (inference, fine-tuning) in HIVE. This can be abstracted via API (e.g., paying with stablecoins converted seamlessly to HIVE on the backend) for a frictionless user experience

Economic Loop

Demand-side users pay tokens for compute.
Payments flow to active worker nodes as rewards.
Workers are incentivized to stake more tokens to maximize their earnings potential.
Increased demand for compute drives demand for tokens (for payments and staking).

Distribution & Incentives

Initial token distribution will strategically incentivize various types of demand, not only attracting the initial demand side, but distributing tokens fairly based on performed work. After the token distribution; decentralised work orchestration on the network and staking mechanics take over.

We think proof-of-work like mechanics are extremely valuable for token distribution, while proof-of-stake mechanics are more robust and scalable once the initial token distribution has been established.

Seize the opportunity, fill the gap ⏳

There hasn't been a better timing for JoinHive yet.

Unprecedented AI demand: Driven by larger models and complex reasoning, demand is outstripping available (cloud) supply by orders of magnitude (as highlighted by Nvidia's CEO).
Consumer hardware power: Modern GPUs offer incredible performance-per-dollar, beating Nvidia's flagship A100 and H100 on optimized inference, making them ideal for our network.
The rise of open source: Open source models like Llama and Deepseek are competing with proprietary models.
Web3 maturing: Several projects have demonstrated the power of token incentives to bootstrap global hardware networks, solving the coordination problem that stalled altruistic, academically driven attempts.

The AI infrastructure market is projected to be worth hundreds of billions, with the exponential growing demand for compute being the primary cost driver.

By unlocking latent supply and drastically reducing costs, JoinHive isn't just capturing market share; we are expanding the entire market for AI innovation.

AI Startups & SMBs: Currently priced out or severely limited by cloud costs. (Initial Go-to-Market Focus)
AI Agent Developers: Requiring persistent, affordable inference for autonomous systems. (Strong Early Use Case)
Independent Researchers & Developers: Democratising access to cutting-edge models.
Collective Pool of GPU Owners: Providing a compelling, accessible way to monetize hardware.