Polkadot Super AI Center Proposal

Proposer: Hashforest LLC (US)
Funding Request: $1,500,000 USD (375,000 DOT)
Proposal Date: 2024.10.02

1. Abstract

This proposal requests funding to establish a Polkadot Super AI Center operated by Hashforest LLC. The aim is to build a decentralized AI infrastructure utilizing high-performance GPUs owned by the Polkadot community, forming the foundation of a minimal AI training center that serves Web3 and Web2 AI developers alike. This center will accelerate the integration of AI into the Polkadot ecosystem and make AI training resources accessible to the broader Web3 community.

Read full report here: https://hackmd.io/@Marvin/SyEaXwsCA

2. Project Scope

The Polkadot Super AI Center will provide AI infrastructure by deploying 16 NVIDIA H100 GPUs. These will be used for training large-scale AI models, initially managed by Hashforest LLC, with full ownership by the Polkadot community. It will serve as a decentralized solution to the growing demand for AI compute resources.

Key Objectives:

• Build and deploy a minimal AI training center with 16 NVIDIA H100 GPUs.
• Provide access to AI compute for Polkadot ecosystem participants and AI developers.
• Create research partnerships with Web3 and AI companies to push the boundaries of decentralized AI.

3. Rationale

Why Decentralized AI?

Artificial intelligence is becoming ubiquitous, and decentralized AI is emerging as the next frontier in Web3. Polkadot's existing ecosystem (e.g., Phala Network, OriginTrail, and Bittensor) is well-positioned to address the inefficiencies in Web2 AI solutions. The Polkadot Super AI Center aims to solve key bottlenecks faced by AI builders, such as limited access to affordable GPU resources for training and inferencing.

Problem to Solve:

• High entry barriers: AI companies need affordable, scalable compute resources.
• Web3 AI alignment: A decentralized approach will align with Web3’s trustless and open infrastructure while addressing AI's needs for scalability and performance.

4. Strategy

To make AI development accessible and incentivize its growth within Polkadot, the Polkadot Super AI Center will offer GPU compute credits to approved participants in the Polkadot ecosystem.

Customer Plans and Pricing:

Customer	Plan	Key Performance Indicator (KPI)
Parity / Web3 Foundation	Free ($50k credits/month)	N/A
Polkadot Ecosystem Teams (30)	Free ($30k credits/month)	Social media/blog contributions
Web3 Partners (20)	50% market price ($5k credits)	Research papers, PR announcements
AI Partners (20)	50% market price ($5k credits)	Research papers, PR announcements
Public (20)	80% market price	20+ paid companies onboarded

5. Milestones

Milestone 1 - Initial Setup (6 months)

• Establish an AI training center with 16 NVIDIA H100 GPUs.
• Deploy and integrate AI models like Llama3.2 into the infrastructure.

Milestone 2 - Inference Services (3 months)

• Set up model inference services with gated access.
• Create an AI grant system for Web3 & Web2 AI developers.

Milestone 3 - Onboarding (3 months)

• Onboard at least 20 AI developers/companies.
• Publish five collaborative research papers with top-tier Web3 and AI companies.

Milestone 4 - Market Report (3 months)

• Submit a comprehensive report evaluating AI product market fit within the Polkadot ecosystem.

6. Benefits to the Polkadot Ecosystem

• Increased DOT utility: Revenue generated from the AI center will be used to buy and burn DOT.
• AI Ecosystem Growth: Accelerating the adoption of AI within Web3 through Polkadot’s infrastructure will drive value back to the network.
• Strategic Partnerships: Building relationships with AI companies and Web3 projects to further Polkadot's position in the AI and decentralized computing space.

7. Monitoring and Governance

To ensure transparency, a Polkadot AI Committee will be created. This body, along with Hashforest LLC, will submit monthly reports on the operations and usage of the AI center. All income generated will be public, and the center's progress will be reviewed at each milestone.

8. Cost Breakdown

Total cost of setup and operations for the first year:

• GPU Hardware: $778,000 to $990,000 for 2 NVIDIA DGX H100 systems.
• Additional Servers & Networking: $130,000.
• Power, Cooling, and Facility Costs: $160,000.
• Staffing & Maintenance: $230,000 annually.
• Estimated Setup Costs: $1,252,000 to $1,464,000.
• Annual Operating Costs: $308,000.

9. Why Hashforest LLC?

Hashforest LLC has been actively contributing to Polkadot since 2019 through Phala Network. With deep expertise in decentralized computing, AI model integration, and trusted execution environments (TEE), Hashforest LLC is uniquely qualified to spearhead the Polkadot Super AI Center.