Introducing Berkeley Compute: Revolutionizing AI with Decentralized GPU Power

We are thrilled to announce the official launch of Berkeley Compute, a revolutionary new decentralized GPU platform that stands at the intersection of artificial intelligence (AI) and blockchain technology. We have been operating in stealth mode for the last six months, launching amidst the growing demand for AI compute resources. Today, we want to share our vision with the world and our $5M seed round, led by Chia Network. More details about the funding here.

Our mission: To empower millions to own a piece of the largest distributed GPU network.

Access and ownership to AI compute resources should not be limited to a select few. We're enabling individuals and organizations of all sizes to participate in the AI revolution. Our platform allows GPU owners to monetize their hardware, data center operators to manage resources efficiently, and AI developers to access affordable, scalable compute power.

At Berkeley Compute, we want to build the most powerful, cost-effective compute network to accelerate the AI revolution. We believe that by providing great tools to operators and GPU cluster owners, we can accelerate faster than any one operator.  

Our vision: to decentralize AI compute infrastructure

Berkeley Compute enables data center operators to have powerful blockchain capabilities to crowdsource capital to sell GPUs, automate operations and payments, and manage large GPU clusters. At scale, anyone can join and provide resources to the network. We also view the operation of GPU clusters as separate from ownership.

Market participants have permissionless access to purchase tokenized assets that generate real revenue from the growing AI compute demand.

AI startups have instant access to scalable, reliable, and inexpensive compute to grow their businesses.

Blockchain technology is at the heart of our platform, transforming physical AI infrastructure into digital twins, providing the transparency, permissionless markets, and blockchain ecosystem. We believe that:

• Transparency ensures all stakeholders have clear visibility into operations and transactions.
• Permissionless Markets allow for seamless trading and instant payment rails
• Blockchain ecosystems can fund development of the best in class operator toolkits, marketplaces, and AI platforms.

Our thesis for AI compute networks

Berkeley Compute’s mission and vision are unpinned by a multifaceted thesis.

“We need to act wisely but with conviction. The dawn of the Intelligence Age is a momentous development with very complex and extremely high-stakes challenges” @sama

1. An AI revolution: We believe AI is the final revolution. AI is rapidly changing how we work, live, and interact. The Intelligence Age will require unprecedented compute.
2. Open-Source Models Fuel Innovation: The emergence of open-source AI models like Meta's LLaMA is driving growth in decentralized AI development. These models enable a surge of innovation by allowing developers to fine-tune and build upon existing architectures. The proliferation of these smaller, specialized models increases the need for compute resources, particularly for fine-tuning and inference, and drives a market hungry for the lowest-cost, high-quality providers.
3. Decentralization Unlocks Efficiency: Historically, decentralized networks such as Bitcoin and other proof-of-work blockchains have demonstrated that permissionless systems can achieve remarkable efficiencies. When many participants can contribute capacity, innovative providers emerge, capitalizing on unique cost advantages like access to low-cost energy, affordable hardware, or alternative funding sources. We believe that in the realm of AI compute, small to mid-sized data center operators can leverage these advantages to outperform hyperscalers like Amazon Web Services and Google Cloud Services, offering more competitive pricing and specialized services.
4. Blockchain advantages: Massively distributed networks of operators and GPU cluster owners creates complexity, especially in payments. Fortunately, blockchains are the ultimate settlement machine, highly efficient and low cost. With payments and GPU utilization reported on chain, we create transparency and auditability that enable GPU owners to trust, and verify.
5. DePIN tokenomics: As a decentralized physical infrastructure (DePIN) project, Berkeley Compute’s protocol uses crypto incentives via tokenomics to help solve the typical cold start problem in network growth. Early providers of compute on the network benefit by being early and providing compute, while we build up demand.  
6. Tokenizing GPUs accelerates growth: GPUs on Berkeley Compute are represented on chain as a digital twin of their physical GPU. This enables operators to raise capital by selling these digital twins to crowdfund capital, acquire clusters, and host them on behalf of owners. GPU owners can enjoy passive income without requiring the technical sophistication required to operate a cluster. By reducing barriers to ownership, anyone can own a cashflow generating asset, and through blockchain protocols, sell or transfer the asset to anyone else, making a liquid market for compute.
7. Data center operators are key: The US market alone has over 5500 data centers. We believe that these players are key to growing the network, as they can provide uptime, quality of service, security and the technical sophistication required to provide compute for AI.

Why Chia? 🌱

We chose to start our journey on Chia blockchain for a variety of reasons.

Firstly, we know that Chia has fantastic technology that can enable Berkeley Compute. Its NFT standard enables a highly secure, self-custodied ownership and provenance for real world assets. Chia’s pioneering Offers primitive creates One Market for trading GPUs and compute tokens.

Secondly, Berkeley’s co-founders are all known entities in the Chia ecosystem: Paul Hainsworth is a former Chief Product Officer; JM Hands is a former VP of business development at Chia and a current advisor; Dylan Rose built Evergreen, the largest business in terms of revenue by a third party Chia ecosystem player. As we seek to grow Berkeley Compute alongside the Chia ecosystem, we are well placed to build on our reputations.

Finally, we believe that Chia’s ecosystem is a massive accelerator to our growth. There are tens of thousands of farmers, including thousands running data center operations. Most of these data center farmers are already running highly sophisticated operations including with GPUs. These farmers are well placed to contribute idle resources to the Berkeley Compute ecosystem, and to grow their business as a significant operator on our network.

Current momentum

We are driving toward a testnet release in Q1 2025, with mainnet in  Q2 2025. Berkeley Compute grows on the supply side by adding operators, and GPUs to the network. Today, we announce two partnerships on the operator side. Evergreen, Dylan’s company, has over 800 GPUs running in a data center. The majority of these GPUs will be tokenized and put to work on Berkeley Compute. FarmGPU, JM’s company, will have about 200 GPUs by the end of year, online in its own data center. Similar to Evergreen, most of these will be tokenized and available for rent on Berkeley Compute.

In Q4, we plan to start onboarding new data center operators. You can join our waitlist here. We will partner with Evergreen, FarmGPU and our newest operators to enable early participation as a GPU owner. For those interested in signing up to purchase GPU NFTs, you can get on our waitlist here.