Why Continuous Token Auctions Beat Batch Grants for AI Funding

Batch grants create multi-month funding gaps where promising projects starve. Committees meet quarterly, causing ~90-day decision cycles while innovation moves at internet speed. This misalignment kills momentum.

• Opportunity Cost: Valuable researcher time wasted on grant writing, not building.
• Market Disconnect: Funding decisions are based on outdated proposals, not real-time traction.

Centralized committees are black boxes prone to cronyism and herd mentality. Merit is secondary to networking, creating a 'grant capture' ecosystem. The process lacks the price discovery of a free market.

• Lack of Accountability: No skin-in-the-game for committee members using treasury funds.
• Inefficient Discovery: The best technical work is often the worst marketed, and gets overlooked.

Modeled after mechanisms like Olympus Pro bonds or Liquidity Bootstrapping Pools (LBPs), continuous auctions create a permissionless, real-time funding market. Projects list future token streams; the market prices them instantly.

• Real-Time Price Discovery: Community conviction is quantified via capital commitment, not committee votes.
• Aligned Incentives: Contributors become immediate stakeholders, tied to project success.

Instead of funding vague proposals, auctions can finance specific, milestone-based workstreams with on-chain verification. Think Gitcoin Grants 2.0 meets Optimism's RetroPGF, but continuous. Each deliverable (e.g., model checkpoint, dataset) is a tradable asset.

• Modular Funding: Back specific outputs, not entire organizations.
• Provable Impact: Completion triggers payment automatically via oracle or proof.

Continuous auctions bake in the exit strategy. Early funders (e.g., VCs, grants) can offload their token allocations to the community gradually, preventing toxic dumps. This mirrors the gradual decentralization of projects like Curve and its veToken model.

• Liquidity from Day 1: Contributors have a clear path to realize value.
• Sustainable Treasury: Project treasuries earn fees from secondary market activity.

Hyperliquid's perpetual grant auctions are a live prototype. Researchers auction a stream of their future token supply; the protocol uses the proceeds to market-make a perp for that token. This creates immediate liquidity and price discovery, solving the 'illiquid grant token' problem.

• Capital Efficiency: Funding and liquidity are solved simultaneously.
• Novel Mechanism: Demonstrates how DeFi primitives can be repurposed for R&D funding.

Traditional funding is a discrete, high-friction event misaligned with AI's continuous training cycles. Grants are awarded based on proposals, not performance, creating months of lag and misallocated capital.

• Vote-Buying & Politics: DAO governance is slow and susceptible to influence campaigns.
• No Live Feedback: Teams can't adjust funding based on real-time model performance or data.
• Capital Inefficiency: Funds are locked upfront with no mechanism for reallocation to better performers.

Projects sell a stream of future tokens or revenue in exchange for continuous, real-time funding. Think perpetual bonding curve for AI development.

• Real-Time Price Discovery: The market continuously prices the project's future value, not a committee.
• Performance-Linked Funding: As model metrics improve, demand for the token stream increases, raising more capital.
• Automatic Reallocation: Underperforming projects see funding dry up as investors exit the stream, freeing capital.

A live case study. Miners (AI models) earn TAO tokens in real-time based on the value of their intelligence as determined by peer validation.

• Continuous Evaluation: Models are scored and paid every ~12 seconds, not quarterly.
• Meritocratic Sink: Poor models are slashed and replaced, creating a Darwinian market for AI.
• Subnet Specialization: Over 70 subnets compete for emissions in niches from data scraping to image generation.

Builds a sovereign compute layer where AI models are funded and accessed via crypto-economic incentives. Infernet nodes earn for serving inferences.

• Pay-Per-Query Funding: Developers fund model usage via continuous micropayments, creating a direct revenue stream.
• Model Staking: Model publishers can stake to signal quality and earn fees, aligning long-term incentives.
• Composable Treasury: DAOs can auto-allocate treasury funds to model streams based on usage and performance data.

Continuous funding turns capital from a static resource into a dynamic, high-velocity asset. This is the core financial innovation.

• Reduced Dead Weight: Capital isn't parked in multisigs; it's constantly working or being re-deployed.
• Compound Innovation: Faster funding cycles enable more rapid experimentation, accelerating the entire field's R&D flywheel.
• VC Disintermediation: The best projects can bootstrap liquidity from a global, 24/7 market, not a handful of partners.

This model isn't a panacea. It introduces novel attack vectors that protocols must solve.

• Oracle Manipulation: Model performance scores must be trust-minimized (e.g., zk-proofs, decentralized validation).
• Flash Loan Attacks: Rapid capital flows can be exploited to temporarily distort valuation metrics.
• Regulatory Gray Area: Continuous token sales may face scrutiny as unregistered securities offerings.

Traditional grant programs like those from the Ethereum Foundation or Optimism Collective operate on slow, opaque voting cycles. This creates misaligned incentives and stifles high-velocity experimentation.

• Decision Lag: ~3-6 month cycles vs. AI's weekly iteration pace.
• Opaque Valuation: Subjective committee decisions lack price discovery.
• Misaligned Incentives: Grantees optimize for proposal approval, not market traction.

Model funding after bonding curves or Harberger taxes, creating a real-time market for project equity or revenue shares. Inspired by mechanisms in Radicle or Factory DAOs.

• Continuous Liquidity: Projects raise capital 24/7 based on verifiable milestones.
• Clear Price Signals: Token price reflects collective belief in project value.
• Efficient Exit: Early backers can sell their "share", recycling capital to new projects.

Teams post work commitments (e.g., train a model to X accuracy) with a bonded stake. The market funds the work in exchange for future tokens or revenue, creating a verifiable, trust-minimized pipeline.

• Skin in the Game: Teams must bond capital, filtering out low-effort proposals.
• Automated Payouts: Funds release upon on-chain verification (e.g., via Oracle like Chainlink).
• Composable Funding: Successful projects can instantly spin up new auctions for next milestone.

Continuous auctions create a competitive, meritocratic arena for AI talent, far more efficient than Y Combinator or a16z batch processes. It's the Uniswap of talent discovery.

• Global Pool: Any developer worldwide can participate, not just those in Silicon Valley.
• Rapid Iteration: Failed experiments quickly lose funding; successful ones attract more.
• Capital Efficiency: Money constantly flows to the hottest, most verifiable opportunities.