Capital always pursues future opportunities. The real money of European and American venture capital is often an important indicator of the prospects of a track.

On the one hand, Nvidia's stock is rising steadily; on the other hand, global institutions are rushing to buy Bitcoin ETFs.

This undoubtedly shows that AI and Web3 are the hottest fields in recent years, and will also be the core force that will change the world pattern in the future, with far-reaching influence.

However, in a world increasingly affected by AI, the dominance of innovation and subversion has long been in the hands of a few people.

The computing resources and infrastructure required to develop artificial intelligence have become the key to opening this door, but the acquisition of these resources is often highly concentrated and limited to those who People with strong capital or institutional support.

In addition, high usage costs, lack of credible verification of calculation results, and privacy security issues further limit the popularity and fairness of AI. .

The future of AI should not only serve the commercial interests of a few people, but should become a public domain that everyone can participate in and benefit from, just like Web3. wealth. This is a journey that belongs to everyone, not the exclusive domain of a few.

01 Introduction and Functions

Hyperbolic is an open source AI computing and reasoning service provider , was born from the vision of challenging the status quo, and is committed to allowing innovators around the world to have equal access to AI technology regardless of resources or geographical location.

Hyperliqui's three core functions include:

1.1 GPU market: on-demand computing power, cost-effective

Hyperbolic's GPU market breaks the traditional computing power rental model and provides developers with on-demand computing power services by aggregating idle GPU resources around the world, helping them save up to 75% of costs. Relying on the Hyper-dOS decentralized operating system, developers only need toThe required computing power can be obtained in one minute, greatly lowering the threshold for innovation.

1.2 Inference service: low cost, high efficiency

Hyperbolic's inference service processes more than one billion Tokens every day, provides the latest open source models at extremely low costs, and supports the BF16 format, ensuring excellent performance in efficiency and accuracy.

1.3 Proof of Sampling (PoSP): the gold standard for verification

Hyperbolic's original sampling proof protocol ensures that the output results are both reliable and cost-effective through strict data privacy protection, making it the only Web3 real-time inference product that can provide verifiable AI results.

02 Goal

Hyperbolic has three goals: 1. Provide decentralized heterogeneity Compute 2. Ensure the security and verifiability of decentralized artificial intelligence 3. Protect privacy in decentralized AI.

2.1 Provide decentralized heterogeneous computing

Hyperbolic is committed to building a scalable system that integrates global GPU computing power to optimize the performance of various types of GPUs. This vision aims to break through the bottleneck of computing resource allocation and provide high-performance support to AI researchers and developers around the world.

Hyperbolic first built an AI service layer, allowing developers to deploy and utilize global computing resources to run different AI services.

It can compile various advanced machine learning frameworks (such as PyTorch, TensorFlow, JAX) into underlying languages ​​that adapt to different hardware platforms (such as NVIDIA's CUDA, AMD's ROCm, Apple's Metal).

In addition, Hyperbolic also cooperates with AMD to improve the performance of AMD chips.able. Under Hyperbolic optimization, the input throughput of the Llama3-8B model on the AMD MI250 platform increased by 120.4%, and the output throughput increased by 144.8%.

Hyperbolic’s solution is not only favored by Web3 AI projects, but also attracts a large number of Web2 Adoption by AI developers.

Although Web2 developers often worry that decentralized solutions may affect performance and reliability, Hyperbolic has demonstrated outstanding performance in the fields of large language models and image generation.

Even though the team size is much smaller than that of mainstream competitors, Hyperbolic still achieves performance comparable to or even exceeding theirs, fully proving the superiority of its technical architecture.

This breakthrough eliminates doubts about decentralized solutions and opens the possibility of cooperation for more developers.

Hyperbolic’s decentralized computing advantage stems from its unique system architecture - based on Hyper-dOS is designed with inspiration from the solar system. The architecture uses a hierarchical cluster model to combine efficiency and stability.

Sun Cluster is a central governance node, similar to the core position of the sun in the planetary system, providing basic services and support for the entire system to ensure stability reliability and efficient operation.

Surrounding it are multiple planet-level clusters, including: Mercury Cluster (single node), Mars Cluster (multi-node) and Jupiter Cluster (multi-satellite) node). Each cluster has different sizes and governance characteristics and can be flexibly adapted to different needs.

Three key features of the system

Auto-scaling: the cluster can automatically expand or expand according to computing needs Shrink scale and flexibly respond to load changes.

Self-healing: The system can automatically detect problems and recover from failures to ensure stable operation.

This layered architecture not only ensures the high availability and scalability of the system, but also achieves a balance between autonomy and overall coordination. Users only need to own a machine or a cluster, and after installing Hyper-dOS, they can easily access the Hyperbolic network, obtain global computing resources, and achieve seamless collaboration.

2.2 Ensure decentralization Security and verifiability of artificial intelligence

A key challenge faces in decentralized networks: how to ensure that the results generated by random nodes are correct. Security and verifiability have always been unresolved issues in deployed AI systems.

The current popular AI verification mechanisms include consensus/voting, optimistic mechanisms and zero-knowledge proofs.

The consensus/voting mechanism requires multiple nodes to run the same request at the same time and pass a majority Vote to determine the answer. However, the cost of this approach is very high. If 10 nodes handle the same request, the overhead will increase 10 times.

Optimistic mechanism (OPML) verifies the results by allowing a single node to generate results and setting a challenge window (usually 7 days) for other nodes to raise objections.

But this method is not practical in real-time scenarios. For example, if a user asks "What fun places are there in Singapore?" it is meaningless if it takes 7 days to confirm whether the answer is correct.

Zero-knowledge proof has excellent performance in terms of privacy and verification, but the computational cost is too high, making it difficult to be practical in the short term.

In order to solve these problems, Hyperbolic teamed up with experts from the University of California, Berkeley and Columbia University to propose a new verification mechanism based on Nash equilibrium, called "sampling" Proof” (PoSP). This mechanism focuses on sampling verification rather than comprehensive inspection of all results.check.

Normally, only one node generates the result, but the network will randomly ask another node to regenerate with a certain probability. If the results of the two nodes are inconsistent, an arbitration process will be initiated. Dishonest nodes will be subject to high economic penalties.

The threshold formula of pledge and reward derived through mathematical models shows that as long as the probability of checking is higher than the threshold, the system can reach pure Nash in game theory The equilibrium state ensures that all nodes choose to be 100% honest for their own benefit.

This sampling proof mechanism is not only effective for AI reasoning, but can also be applied to AI training, fine-tuning and other fields, and can even be extended to services outside the AI ​​field, such as L2 Rollup and data availability, etc.

Hyperbolic is working with re-staking protocols such as EigenLayer and Karak to build a common verifiable service layer (AVS) that other AVS service providers can also take advantage of. This verification mechanism ensures the security and reliability of its services.

2.3 Protecting privacy in decentralized AI

In a decentralized AI network, how to ensure data privacy and model integrity at the same time is a big problem that needs to be solved. When your data is distributed across nodes around the world, security challenges arise.

Existing technologies such as fully homomorphic encryption (FHE), zero-knowledge proof (ZKP) and multi-party computation (MPC), although these problems can theoretically be solved, However, in practical applications, it will greatly reduce the calculation speed and cannot meet the needs of real-time reasoning.

Hyperbolic uses the Trusted Execution Environment (TEE) technology on NVIDIA's latest Hopper and Blackwell GPUs to provide an efficient privacy protection solution.

Through TEE technology, it is equivalent to creating a "privacy safe" on the GPU: although the outside world cannot spy on the data content, the GPU can still complete the data normally deal with.

Moreover, this privacy-preserving mechanism only loses about 1% of computing performance during the inference process.

Hyperbolic will introduce a confidential computing layer throughout the decentralized network. This will ensure that data and AI models are always safe during use, providing users with reliable privacy and security guarantees.

03Hyperbolic application scenarios

AI Agent is currently the hottest track. AI Agent can realize a number of innovative functions through Hyperbolic:

3.1 Support crypto payment

AI Agent can be paid via cryptocurrency to achieve self-sustainability and independent operation.

3.2 Hosting customized models

Each AI Agent can have exclusive characteristics and skills to form personalized services.

3.3 Self-evolution ability

Through continuous fine-tuning and learning, AI Agent can continuously improve its capabilities according to user needs or environmental changes, making it more efficient and intelligent.

3.4 Verifiable Reasoning

The reasoning process of AI Agents is transparent and verifiable, ensuring their independence from external control or malicious interference, and enhancing user trust.

3.5 has memory function

With the help of retrieval-augmented generation (RAG) technology, AI Agent can record and store information interacted with users to form long-term memory. This allows them to provide more thoughtful services, such as remembering user preferences.

3.6 Cross-Agent communication

AI Agents can communicate with each other Communicate and collaborate to form a network for solving complex tasks. For example, different Agents can collaborate to complete a multi-step project.

3.7 Flexibly call APIs and tools

AI Agent can integrate and use various external APIs and tools, greatly expanding its functional scope. For example, calling the weather API to plan trips for users, or using financial tools to provide investment advice.

3.8 Autonomous computing capabilities

They can have their own computing devices and run tasks independently. This means that AI Agents can get rid of their dependence on centralized servers and become more decentralized and independent.

3.9 Become a blockchain verification node

AI Agent can even participate in the blockchain network and serve as a verification node. Not only does this enhance network security, but it also enables you to earn rewards for validating transactions, furthering your self-sufficiency.

Recently, Hyperbolic has cooperated with Virtuals Protocol, the most popular Base chain AI launch platform, to provide powerful technical support for AI agents and comprehensively improve their performance. and the ability to develop oneself.

By directly connecting Virtuals Protocol's agents to Hyperbolic's infrastructure, each agent can obtain the highly scalable computing resources and stable reasoning provided by Hyperbolic API capabilities and a seamless dynamic interaction experience that maintains efficient and consistent performance regardless of the number of agents or task complexity.

This cooperation not only enhances the computing power of the AI ​​agent, but also improves its performance in diverse applications.Adaptability and intelligence in use scenarios.

For example, Hyperbolic's infrastructure provides the game's intelligent NPCs (non-player characters) with persistent memory and personality development capabilities.

In the game "Legendary Quest", Virtuals Protocol's advanced AI agents are integrated. These NPCs can maintain consistent personalities based on player interaction and adjust based on past experiences. behavior patterns, and continue to develop your own plot even when the player is offline.

All this is due to Hyperbolic's scalable computing network, which allows these NPCs to make complex decisions and personality evolution without affecting game performance. .

This collaboration enables developers to transform AI concepts into practical solutions and promote innovative development in gaming, virtual assistants, education, content creation and other fields.

04 compared with competing products

4.1 Partnerships

Hyperbolic has earned the trust of leading artificial intelligence companies such as Hugging Face, Quora, Black Forest Labs and Nous Research, as well as Stanford University, Supported by top universities such as New York University and University of California, Berkeley.

Developers can seamlessly create and share AI applications on Hugging Face Spaces through Hyperbolic's inference API, greatly simplifying the deployment and distribution process.

In addition, doctoral students and postdoctoral researchers at Stanford University, Cornell University and New York University can also enjoy up to 75% discount on GPU rental, significantly reducing computing costs. cost.

Hyperbolic's AI model, including the basic model, is now available on Quora's Poe platform. Developers can easily create and deploy chatbots and implement them directly through the platform Business realization.

4.2 Optimized performance

Hyperbolic's proprietary compiler ensures GPU Efficient operation, its performance can match or even exceed the centralized system

4.3 Excellent model quality

All models use BF16 precision, providing superior accuracy and performance ahead of competitors still using FP8.

4.4 Data Privacy and Security

Hyperbolic solves the security problem in AI verification through the Proof of Sampling Protocol (PoSP), while achieving minimal The computational cost of zkML, opML and consensus-based alternatives have advantages. In addition, Hyperbolic does not store user data at all, further protecting privacy. -tpl="t"> 4.5 mature real-time product

Unlike many Web3 AI projects that are still under development or have restricted access, Hyperbolic has launched two There are currently more than 100 products available in real time. 40,000 Web2 developers are using its services

4.6 Unified Computing and Inference

Hyperbolic is the only company that can provide both GPU computing and inference services on the same platform, successfully realizing a unified computing solution.

In conclusion, compared with Web2 AI companies whose teams are 10 to 30 times larger, Hyperbolic has achieved comparable or even superior performance with only a streamlined team, while providing more cost-effective services through Web3 mechanism design.< /p>

In the field of Web3 AI, Hyperbolic is far ahead with its leading technology and has won the trust of Web2 developers. Hyperbolic has built a high-speed and convenient bridge between the AI ​​fields of Web2 and Web3, becoming a driving force in the industry. An important cornerstone of development

05

Financing situation

On December 10, Hyperbolic announced the completion of a US$12 million strategic financing round led by Variant and Polychain Capital, bringing the company’s total financing to $20 million.

This round also attracted Chapter One, Lightspeed Faction, Bankless Ventures, IOSG, Vertex, GSR, Wintermute. Ventures, Blockchain Builders Fund, Alumni Ventures and Ambush, among other well-known investors

Previously, Hyperbolic had completed a $7 million seed round from Polychain Capital. and Lightspeed Faction led the investment; earlier, it also received US$725,000 in pre-seed funding from investors including Chapter One and Samsung Next.

In addition, the lineup of angel investors in this round of financing is also very strong, including Sreeram Kannan (EigenLayer), Devin Walsh (Uniswap Foundation), Ethan Sun (MyShell), Daniel Shorr (Modulus), Bidhan Roy (Bagel), Ying Sheng and Lianmin Zheng (LMSYS), Dillon Rolnick (Nous Research), Alex Atallah (OpenRouter), Chainyoda, Comfy Capital, NicolaGreco (Protocol Labs), Alex Atallah (OpenRouter) and Thomas Scott (ex-Worldcoin).

Variant partner Jesse Walden expressed high recognition for Hyperbolic: "Hyperbolic is the first solution we have ever seen that truly solves the 'trust cost' in decentralized GPU networks. 'Problem company while maintaining high levels of performance, quality and user experience"

Hyperbolic in Web3 AI. It is in a leading position in financing in the field, which fully proves that its technical strength and product feasibility have won the favor and trust of "smart money" in the industry.

06 Team Background

Co-founder Jasper Zhang graduated from the Department of Mathematics of Peking University and obtained a PhD in Mathematics from the University of California, Berkeley, within two years at an astonishing speed.

Prior to founding Hyperbolic, he worked as a quantitative researcher at Citadel Securities and as a senior blockchain researcher at Avalanche.

Co-founder and part-time CTO Yuzhen Jin holds a PhD in computer science from the University of Washington and served as a senior engineering manager at OctoAI before founding Hyperbolic.

Hyperbolic's team members all have backgrounds in top schools, the founder has a solid technical foundation, and many team members have worked with Avalanche before.

The company's consulting team is also composed of top professionals in the industry.

Dr. Reynold Xin is the co-founder and chief architect of Databricks, a major contributor to Apache Spark, and the author of the most highly cited paper in SIGMOD.

Prof. Raluca Ada Popa is an associate professor at the University of California, Berkeley, co-director of RISELab and SkyLab, and co-founder of Opaque Systems.

Prof . Ciamac C. Moallemi is a professor at Columbia University Business School, a research consultant at Paradigm, and the director of the Briger Family Digital Finance Lab

Prof. Yi Ma. He is the chairman of the Department of Computer Science at the University of Hong Kong and a chair professor in the field of AI. He is also a professor of computer science at the University of California, Berkeley, and an academician of IEEE, ACM and SIAM

07. How to participate

7.1 Company

Targeting enterprises over expensive APIs For expenses on calls and high-cost machine rentals, Hyperbolic provides competitive optimization solutions

Hyperbolic's technical support on the premise of ensuring stable service quality. It can help enterprises reduce costs by up to 75%.

At the same time, Hyperbolic addresses the inefficient use of resources caused by long-term GPU rental agreements. A resource reallocation mechanism has been launched to allow customers to sublease idle equipment to the platform. This model not only improves asset utilization, but also finds the optimal balance between flexibility and cost control.

7.2 Researchers

For developers due to problems in project testing To solve the problem of limited GPU resources and inability to advance, Hyperbolic provides a wealth of GPU options, and the price is only a fraction of traditional cloud service providers such as AWS. By providing cost-effective resources, Hyperbolic provides developers with the most competitive solutions in the market, helping them quickly turn innovative ideas into reality.

7.3 Data Center

Hyperbolic For those whose return on investment of existing resources has not met expectations, or who hope to break through the traditional book value limit The data center provides a platform to achieve higher profits.

7.4 people

High performance The potential of GPUs should not be limited to gaming. Through Hyperbolic, individuals rent out GPUs and turn them into high-quality assets that continue to generate income. Currently in the whitelist stage, you can register first.

In addition, Hyperbolic provides a variety of large models for personal use. Users can perform text, picture generation, voice reading and other activities.

In the future, Hyperbolic will also build AI agents on Base for users to use. You can stay tuned.

Hyperbolic web page:

app.hyperbolic.xyz?utm_source=x&utm_campaign=seriesA&utm_content=biteye

08 summary

Hyperbolic provides GPU Marketplaces, inference services, and proof-of-sampling, the gold standard verification protocol, set a new standard for reliable, high-performance AI for Web3 by maximizing GPU performance, higher-accuracy models, and secure and cost-effective solutions.

The emergence of Hyperbolic has brought decentralized AI from concept to practice. Hyperbolic occupies a unique position in the ecosystem with its multi-source computing strategy, competitive pricing, and deep understanding of Web2 and Web3 customer needs.

Hyperbolic's efforts in promoting the democratization and efficient utilization of computing resources will promote the development of the AI ​​​​track and bring continuous innovation and growth to the industry.