The AI industry faces significant challenges, including the high cost and limited availability of computational resources. Training and deploying AI models require substantial GPU power, often expensive and inaccessible for many users. Additionally, there is a lack of transparency and verification in AI processes, leading to trust issues and inefficiencies.

Hyperbolic addresses these problems by leveraging decentralized technology. By creating a decentralized GPU marketplace, Hyperbolic makes computational resources more affordable and accessible. This marketplace allows users to rent idle GPU capacity from various suppliers, reducing costs and increasing availability. Additionally, Hyperbolic introduces Proof of Sampling (PoSP) for reliable AI computations and employs spML and the Hyperbolic Decentralized Operating System (Hyper-dOS) to optimize AI model training and deployment.

What is Hyperbolic?

Source: Hyperbolic website

Hyperbolic is a decentralized platform designed to revolutionize access to AI and computational resources. At its core, Hyperbolic creates a marketplace where users can rent idle GPU capacity from various suppliers. This approach ensures efficient use of resources, significantly reducing the costs associated with high-performance computing. By decentralizing the availability of GPU power, Hyperbolic makes it possible for a broader range of users to engage in AI development and deployment, breaking down barriers that have traditionally limited access to these technologies. Hyperbolic was founded by Dr Jasper Zhang and Dr Yuchen Jin.

Mission and Vision of Hyperbolic

Hyperbolic’s mission is to democratize access to AI and computational power, making these resources available to everyone, regardless of their financial or technical capabilities. The vision is to create a transparent, efficient, and decentralized ecosystem that fosters innovation and collaboration in the AI industry. By leveraging blockchain technology and decentralized principles, Hyperbolic aims to build a future where AI development is more inclusive and equitable.

Users of Hyperbolic

• Companies: Businesses can utilize Hyperbolic for cost-effective AI model training and deployment, reducing operational expenses and accelerating AI initiatives.
• Researchers: Academic and industry researchers gain access to affordable computational resources, enabling them to conduct advanced AI research without the financial burden of traditional GPU costs.
• Data Centers: Data centers can monetize their idle GPU capacity by renting it out on the Hyperbolic marketplace, turning unused resources into a revenue stream.
• Individuals: Hobbyists, students, and independent developers benefit from accessible AI tools and resources, allowing them to experiment, learn, and innovate without significant financial investment.

Technology Behind Hyperbolic

Proof of Sampling (PoSP)

Proof of Sampling (PoSP) is a verification protocol designed to ensure the integrity and reliability of AI computations in decentralized systems. PoSP uses a sampling method to verify transactions and data interactions, significantly reducing the computational load compared to traditional verification methods. By leveraging game-theoretic principles, PoSP encourages participants to act honestly, enhancing the overall trustworthiness of the network.

How Proof of Sampling Works

PoSP operates by selecting random samples from a dataset or computational task and verifying these samples instead of the entire dataset. Using a game-theoretic model known as pure strategy Nash Equilibrium, PoSP encourages all participants to act honestly, enhancing the network’s trustworthiness and reliability. This method reduces the amount of data that needs to be processed, making the verification process more efficient. Participants in the network are incentivized to provide accurate samples because dishonest behavior can be detected through statistical analysis. If discrepancies are found, penalties are imposed, ensuring that most participants act honestly.

spML

spML is a verification mechanism designed to address the pitfalls of previous systems (zkML and opML) by streamlining the process of AI inference verification in decentralized networks. The spML protocol utilizes a straightforward yet effective method to ensure both rapid processing and high security without the computational overhead and complexity associated with zkML or the vulnerability to fraud seen in opML.

How spML works

The protocol begins when a user sends an input with their digital signature to a randomly selected server, known as Server A. Server A processes the input and returns the output along with its hash, also signed to verify its authenticity. To ensure the reliability of the inference, the protocol may randomly involve an additional server, Server B, to independently verify the output. This occurs with a predetermined probability; if Server B is not selected, Server A receives a reward, and the transaction concludes successfully.

If Server B is involved, it processes the same input and sends back its output and hash to the user. The user then compares both hashes. If they match, indicating consistent results, both servers are rewarded. If the hashes differ, indicating a potential discrepancy or fraud, the user broadcasts this information to the entire network. The network, consisting of multiple nodes, votes to adjudicate the claim based on the majority rule. Sanctions are imposed on any dishonest party to maintain the system’s integrity and trust.

Hyperbolic Decentralized Operating System (Hyper-dOS)

Hyper-dOS is the decentralized operating system that manages and orchestrates resources within the Hyperbolic ecosystem. It ensures that computational tasks are distributed and executed efficiently across the network. Hyper-dOS plays a crucial role in maintaining the performance and scalability of Hyperbolic’s decentralized infrastructure, enabling seamless integration and operation of various AI services. By coordinating the allocation of resources, Hyper-dOS maximizes the utilization of available computational power, ensuring that tasks are completed promptly and efficiently.

Hyperbolic Ecosystem Architecture

Source: Hyperbolic’s blog

Hyperbolic Decentralized Orchestration Layer

The decentralized orchestration layer is the backbone of Hyperbolic’s infrastructure. Powered by the Hyperbolic Decentralized Operating System (Hyper-dOS), this layer manages and optimizes the global GPU infrastructure. It integrates computing power from diverse sources, including data centers, mining farms, personal machines, and on-premises systems.

Key features include:

• Auto-scaling: The system can automatically adjust the size of GPU clusters based on real-time demand, ensuring efficient resource utilization.
• Self-healing: The orchestration layer can detect and recover from failures autonomously, maintaining continuous operation without manual intervention.
• Customizability: Users can tailor clusters to meet specific requirements, providing flexibility and adaptability for various applications.

AI Services Layer

This layer hosts a comprehensive suite of AI services and engines, providing essential functionalities for AI applications. It supports a wide range of tasks, from simple automation to complex optimization and enhancement processes.

Key components include:

• Inference Services: These services enable real-time predictions and decision-making by AI models, ensuring high performance and accuracy.
• Model Training and Fine-tuning: Tools for training and fine-tuning AI models allow developers to adapt models to specific tasks and datasets, improving their effectiveness.
• AI Model Evaluation: This includes tools and benchmarks for assessing the performance and accuracy of AI models, helping developers refine and improve their models continuously.

Verification and Privacy Layer

The verification and privacy layer ensures the integrity and confidentiality of AI computations. It incorporates Proof of Sampling (PoSP) to verify the accuracy of computations, protecting against fraudulent activities. Additionally, this layer includes privacy-preserving techniques to safeguard sensitive data during processing, ensuring that user data remains secure and confidential.

Hyperbolic Blockchain Layer

The blockchain layer is the foundation of Hyperbolic’s security and transparency. It provides a secure and immutable ledger for all transactions and interactions within the ecosystem. This layer enhances trust and accountability by ensuring that all activities are recorded transparently. It also supports smart contracts, enabling automated and secure agreements between parties, which streamline operations and reduce the need for intermediaries.

Application Layer

The application layer is the interface through which end-users interact with the Hyperbolic ecosystem. It includes various applications and user interfaces designed to be intuitive and accessible. This layer ensures that both technical and non-technical users can easily access and utilize Hyperbolic’s services. Applications in this layer range from simple user dashboards to complex development environments, catering to a wide array of user needs.

Hyperbolic’s AI Inference

AI inference is the process where trained AI models interpret new data and make decisions based on their training. Unlike the training phase, which involves learning patterns from vast datasets, inference applies this learned knowledge to new, unseen data to generate predictions or results. Hyperbolic’s AI inference capabilities are designed to be efficient and scalable, leveraging a decentralized network of GPU resources to provide fast and accurate results. This decentralized approach ensures that inference tasks can be distributed across multiple nodes, enhancing performance and reliability.

Benefits of Hyperbolic’s AI inference

• Scalability: The decentralized network can handle large-scale inference tasks, ensuring consistent performance even during peak usage.
• Cost Efficiency: By utilizing idle GPU resources, Hyperbolic reduces the cost of AI inference, making it more accessible to a broader range of users.
• Energy Efficiency: Efficient inference reduces the computational cost and energy consumption, contributing to more sustainable AI applications.

Hyperbolic Open Source AI Models

Source: Hyperbolic’s website

Hyperbolic provides access to a variety of high-performance open-source AI models, enabling developers to leverage cutting-edge technology without the high costs associated with traditional providers. Some examples of available models include:

• Vision Language Models (VLMs), combining visual and textual understanding such as Qwen2-VL-7B-Instruct, Pixtral-12B and Qwen2-VL-7B
• Base Models allowing access to the raw power of foundational AI such as Llama 3.1–405B-BASE (BF16) and Llama 3.1–405B-BASE (FP8)
• Text-to-text models for natural language processing tasks such as Qwen2.5–Code-32B, Llama 3.2–3B, DeepSeek-V2.5, Llama 3.1–7B, Hermes–3-70B, Llama 3.1–405B, Llama 3.1–3B and Llama 3.1–88
• Text-to-image models to unleash your creativity with AI-generated visuals such as Flux 1 [dev], SDXL-1.0, Segmind SD–1B, Stable Diffusion–1.5 and SDXL-1.0-Turbo
• Text-to-speech models for voice synthesis applications such as Melo TTS.

Guide to Accessing and Using Open Source Models

To access and use Hyperbolic’s open-source AI models, follow these steps:

1. Create an Account: Sign up on the Hyperbolic platform to gain access to the available models.
2. Select a Model: Browse the list of available models and choose one that fits your needs.
3. Deploy the Model: Use Hyperbolic’s decentralized infrastructure to deploy the model, ensuring it runs on the best available hardware.
4. Integrate with Applications: Utilize the model’s API to integrate it with your applications, enabling seamless AI capabilities.

Hyperbolic GPU Marketplace

The Hyperbolic GPU Marketplace is a decentralized platform that allows users to rent idle GPU capacity from various suppliers. This marketplace connects those needing computational power for AI tasks with those with excess GPU resources, creating a cost-effective and efficient solution for both parties. By leveraging this marketplace, users can access high-performance GPUs at a fraction of the cost compared to traditional cloud providers.

Tech Behind Hyperbolic GPU Marketplace

The technology driving the Hyperbolic GPU Marketplace is built on the Hyperbolic Decentralized Operating System (Hyper-dOS). This system manages and optimizes the global GPU infrastructure, ensuring that computational tasks are distributed efficiently across the network. Hyper-dOS integrates various sources of GPU power, including data centers, mining farms, personal machines, and on-premises systems, to create a robust and scalable infrastructure.

Key Differentiators of Hyperbolic GPU Marketplace

• Suppliers: Suppliers in the Hyperbolic GPU Marketplace can monetize their idle GPU resources by renting them out to users. This provides an additional revenue stream for data centers, mining farms, and individuals with high-performance GPUs. Suppliers benefit from the platform’s decentralized nature, which ensures fair compensation and efficient resource utilization.
• Renters: Renters can access various GPU options to meet their computational needs. The marketplace offers a seamless experience, allowing users to rent GPUs with just a few clicks. This accessibility and competitive pricing make it an attractive option for researchers, developers, and companies looking to reduce their AI infrastructure costs.

Hyperbolic Pricing

GPU Marketplace Pricing

The Hyperbolic GPU Marketplace offers a flexible and competitive pricing structure for renting GPU resources. Suppliers can set their own prices within the guidelines provided by Hyperbolic, ensuring fair market rates. Here’s a detailed breakdown of the pricing:

80GB VRAM:

• H100 SXM: $3.20 per hour
• H100 PCIe: $3.00 per hour
• A100 SXM: $1.80 per hour
• A100 PCIe: $1.60 per hour

48GB VRAM:

• L40: $1.00 per hour
• L40S: $1.00 per hour
• RTX 6000 Ada: $0.90 per hour
• RTX A6000: $0.75 per hour
• A40: $0.50 per hour

24GB VRAM and Under:

• RTX 4090: $0.50 per hour
• RTX 3090 Ti: $0.30 per hour
• RTX 3090: $0.30 per hour
• RTX A5000: $0.30 per hour
• RTX A4000 Ada: $0.30 per hour
• RTX A4500: $0.30 per hour
• RTX A4000: $0.30 per hour
• RTX 3080: $0.20 per hour
• RTX 3070: $0.20 per hour
• A30: $0.20 per hour
• Tesla T4: $0.20 per hour

Hyperbolic takes a 10% platform fee from the rental income. For example, if a supplier sets the price for an H100 SXM at $2.50 per hour, they will receive $2.25 per hour after the platform fee is deducted. This fee structure ensures that suppliers are fairly compensated while maintaining competitive prices for renters.

AI Inference Pricing

Hyperbolic offers a tiered pricing model for AI inference services, catering to different user needs and budgets. Here’s a detailed look at the pricing tiers:

Basic Tier:

• Free Users: Up to 60 requests per minute.
• Paid Users: Up to 600 requests per minute for users who deposit a minimum of $10 into their accounts.

Services Includes access to text-to-text, text-to-speech, text-to-image, and text-to-video models, as well as fine-tuning services.

Enterprise Tier:

• Unlimited Requests: Suitable for large-scale operations.
• Full Suite of AI Models: Access to all available models.
• Dedicated Support: Custom SLAs and dedicated instances for enterprise users.

Getting Started with Hyperbolic

How to Create an Account with Hyperbolic

• Visit the Hyperbolic Website.
• Sign Up: Choose to log in with your Google or GitHub accounts, or select “Create An Account” to set up a unique password.
• Complete Registration: Fill in the required details and confirm your email address.
• Access the Dashboard: Once registered, you’ll have immediate access to Hyperbolic’s AI Dashboard, where you can explore various AI models and GPU resources.

Getting Started with Hyperbolic AI Inference

To start using Hyperbolic’s AI inference services:

1. Obtain an API Key: After creating your account, navigate to the Settings page on the Hyperbolic AI Dashboard to get your API key.

2. Select a Model: Choose from a variety of AI models available on the platform.

3. Run Inference: Use the provided API endpoints to run inference tasks. For example, you can generate text, images, or audio by sending requests to the appropriate endpoints.

How to Rent GPUs on Hyperbolic

1. Navigate to the Rent GPUs Tab: On the Hyperbolic platform, go to the “Rent GPUs” section.
2. Choose a GPU Instance: Select the GPU instance that best fits your needs from the available options.
3. Rent the GPU: Click ‘Rent’ and wait for the instance to show “Ready to Connect”.
4. Connect to the GPU: Use the provided SSH command to connect to the GPU instance using your preferred SSH client.
5. Utilize the GPU: Once connected, you can start using the GPU for your computational tasks.

Hosting and Monetizing AI Models

1. Prepare Your Model: Ensure your AI model is ready for deployment.
2. Upload Your Model: Use the Hyperbolic platform to upload your model.
3. Set Up Hosting: Configure the hosting settings, including API endpoints and resource allocation.
4. Monetize Your Model: Set pricing for access to your model. Hyperbolic provides tools to manage payments and track usage.
5. Monitor Performance: Use the dashboard to monitor your model’s performance and make adjustments as needed.

Hyperbolic Fundraising Journey

Source: Hyperbolic’s website

Hyperbolic has successfully raised $20 million through multiple rounds of financing, demonstrating strong investor confidence in its vision and technology. Hyperbolic raised $725,000 in its pre-seed funding round in November 2022. This early investment helped the company develop its core technology and build a foundational team. In July 2024 the company secured $7 million in a seed funding round led by Faction and Polychain Capital, with participation from Longhash Ventures, Bankless Ventures, and Nomad. This round enabled Hyperbolic to expand its infrastructure and enhance its decentralized GPU marketplace. Hyperbolic then proceeded to raise $12 million in its Series A round, led by Polychain Capital and Variant, with additional investments from Republic Capital, IOSG Ventures, and Wintermute. This funding was crucial for scaling operations, improving AI services, and expanding the user base.

Conclusion

By addressing critical challenges such as high costs, limited access to computational power, and the need for transparent verification, Hyperbolic is democratizing AI technology. Its decentralized GPU marketplace, innovative technologies like Proof of Sampling (PoSP) and spML, and comprehensive ecosystem architecture provide robust, efficient, and secure solutions for many users. As Hyperbolic moves forward, it remains committed to its vision of creating a transparent, efficient, and decentralized ecosystem that fosters innovation and collaboration. Whether you are a company, researcher, data center, or individual, Hyperbolic offers the tools and resources needed to leverage the power of AI and computational technology effectively.