AI Compute Project Financing: Two Questions to Filter Out the Real Players

Amid the AI compute investment boom, a large number of projects stall in their early stages. According to a practitioner in AI compute trading, distinguishing projects that can actually get built from those that are all talk requires asking just two questions.

Two Key Screening Questions

Question 1: Who is your anchor tenant?

This is the core of what makes a deal financeable. The practitioner emphasizes that the tenant is the single most important factor—without a clear, investment-grade anchor tenant or offtaker, project financing simply isn't possible.

In practice, an anchor tenant means the project has a stable revenue stream before construction even begins. This follows the same logic as traditional commercial real estate development: no developer breaks ground without a signed anchor store. AI compute projects follow the same principle.

Anchor Tenant — This concept originates from commercial real estate, referring to the primary tenant signed before project development begins, typically a large chain brand or a company with a high credit rating. Its core value lies in locking in the project's baseline cash flow through a long-term lease, enabling the developer to secure bank loans or bring in institutional investors on that basis. In the AI compute sector, an anchor tenant usually refers to an AI company, cloud provider, or large enterprise customer that has signed a multi-year Capacity Offtake Agreement. Their credit rating and contract term directly determine whether the project can obtain Project Finance.

Offtake Agreement — This is the legal instrument underpinning the anchor tenant relationship, in which the buyer commits to purchasing a certain volume of products or services at an agreed price over a future period. In compute projects, the offtaker typically commits to purchasing a fixed number of GPU hours or reserving compute capacity. This provides predictable revenue for the project and is a key basis for financiers to assess risk. Without an offtake agreement or letter of intent, a data center project can almost never pass a bank's project finance review.

Question 2: Have you done a similar project elsewhere?

This question quickly separates speculators from industry participants. Experienced teams have reusable knowledge in site selection, power interconnection, cooling solutions, and supply chain management, while teams that are simply chasing the hype often underestimate the engineering complexity.

The practitioner points out that the key isn't the chips themselves, nor whether you can procure chips—it's whether you're speculating on future resale value without creating immediate value, or whether you have a serious, investment-grade anchor tenant. That's how project financing actually works.

Project Finance — This is a financing method that uses the project's own cash flows and assets as the source of repayment, rather than relying on the sponsor's parent company credit. Lenders focus primarily on the project's future revenue capacity and risk allocation structure. In AI compute projects, the feasibility of project finance depends on two core factors: first, whether the tenant's credit and contract term can cover the loan period; and second, whether the project's technical and operational risks are manageable. Teams with a track record of delivering similar projects can significantly reduce lenders' concerns about engineering risk, because inexperienced teams often encounter delays or even failures in power interconnection, cooling solutions, and regulatory approvals—directly threatening project cash flows.

Why These Two Questions Are So Effective

The first question tests the authenticity of demand. In the AI compute sector, many projects claim they will build large-scale clusters, but when asked who will use them, the answer is often a vague "market demand is huge." Real projects point to specific AI companies, cloud providers, or enterprise customers, backed by letters of intent or formal contracts.

The second question tests the capability on the supply side. Building data centers and compute clusters involves complex engineering across power supply, cooling systems, network architecture, and regulatory approvals. Teams with delivery experience have clear understanding and solutions for these areas, while novice teams tend to focus only on GPU procurement.

Together, these two questions form a highly effective filter: real demand plus delivery capability—only then does a project have a foundation to move forward.

Implications for AI Compute Investment

For investors and practitioners focused on AI infrastructure, this screening logic has several direct applications.

When evaluating compute projects, priority should be given to tenant commitments rather than hardware specifications. Even the most advanced chips can't generate returns without clear users. At the same time, a team's historical delivery record is more informative than its technology roadmap—past execution ability is the best predictor of future delivery.

The practitioner's company, Zillion Labs, is engaged in AI compute trading, and these views come from actual deal experience. The underlying logic of project finance isn't complicated: with real demand and execution capability, capital is naturally willing to come in. Without either one, a project will struggle to move from paper to reality.

One-line Summary

Two key questions for screening AI compute projects: who is the anchor tenant, and does the team have experience delivering similar projects?

Paragraph Summary

Amid the AI compute investment boom, many projects stall early. A practitioner points out that distinguishing viable projects from empty talk requires just two questions: First, who is your anchor tenant? This tests the authenticity of demand—without an investment-grade tenant or offtaker, project financing is impossible. Second, have you done a similar project elsewhere? This tests supply-side capability—experienced teams have reusable knowledge in site selection, power, cooling, and supply chain, while speculators tend to underestimate engineering complexity. These two questions form an efficient filter: real demand plus delivery capability gives a project the foundation to proceed. For investors, tenant commitments should be examined before hardware specs, and a team's historical delivery record is more informative than its technology roadmap.