An AI Skill for the A.CRE Data Center Development Model
We’ve been working on a project to make our library of Excel models AI-ready. The idea is straightforward: pair every A.CRE Excel model with an AI Skill, a packaged set of instructions and reference files that teaches an AI assistant how to operate that specific model on your behalf. The Data Center Development Model is the latest in that effort, and this post introduces the AI Skill we built to accompany it.
Think of this as a sister post to the Data Center Development Model post, which walks through the model itself, sections, inputs, outputs, and mechanics if you haven’t seen that one yet, start there. This post focuses specifically on the AI Skill: what it does, how it works, and how to use it.
- While we refer to these as Claude Skills (the format is Anthropic’s), the instructions inside the Skill are largely platform-neutral. You can use the Skill with Claude — where it integrates most natively, especially via the Claude in Excel add-in — but also with ChatGPT, Gemini, or any other capable AI assistant. Just upload the SKILL.md file alongside the Excel model and the assistant can follow the same playbook. Some integrations are smoother than others, but the underlying knowledge transfers.
What is an AI Skill?
If you’re new to the concept, an AI Skill is a packaged set of instructions and reference files that an AI assistant loads alongside your file. It teaches the assistant things it wouldn’t otherwise know — in this case, every input cell, every output, the four user roles the Data Center Development Model serves, the mechanics behind the power-based revenue engine, and the most common ways data center underwriting goes wrong before you ever get to a return number.
The result is an AI assistant that can actually operate the model on your behalf, rather than one that talks about data center development in the abstract.
For a primer with a short video tutorial, see our practical guide to Claude Skills.
What the AI Skill Does for You
Data center development underwriting is fundamentally different from traditional real estate modeling — and those differences run deep enough that a general-purpose AI assistant, without specific instruction, will get it wrong. Revenue is not priced per square foot. It’s priced per kilowatt per month, against an IT Load that each tenant ramps gradually as hardware is deployed and systems are commissioned. Operating expenses scale with power consumption, not occupancy. The development spread — the gap between yield-on-cost and the exit cap rate — is the primary value-creation test, not the IRR. And the model runs circular references that require a VBA macro to resolve, which most AI tools simply cannot handle without explicit guidance.
So the Skill handles a few jobs for you that you’d otherwise be doing manually.
Role Triage
Before the Skill touches a single input, it asks which investment decision you’re making:
- Developer / sponsor — evaluating whether to build and how to structure the deal. The Skill leads with development spread and yield-on-cost, then surfaces unlevered and levered IRR and equity multiple.
- Equity / LP investor — assessing whether projected levered returns justify a capital commitment. The Skill leads with levered IRR, equity multiple, and net profit, and anchors the downside to the untrended stabilized value.
- Development-debt lender — sizing the construction loan and underwriting the takeout. The Skill surfaces total project cost, construction debt, achieved LTC versus target, and stabilized collateral value. DSCR and debt yield are not native outputs but the Skill derives them on request.
- Student / Accelerator member — learning how data center development pencils from first principles. The Skill walks through power-based pricing, how tenant kW loads build up to facility MW, the circular construction-interest problem and why the VBA macro solves it, and how to read the development spread.
Populating Inputs Conversationally
You can describe the facility, share a development budget or GC estimate, paste in colocation lease terms, or provide a debt term sheet — the Skill pulls the relevant assumptions and stages them for the model. A few things it always confirms before writing:
Tenant IT load is entered in kilowatts per tenant (column E in the tenant table). The facility’s megawatt total is derived from summing those kW inputs — it is never entered directly. If you describe a 20 MW facility, the Skill translates that to 20,000 kW distributed across tenants, not a number written to the facility MW cell.
Rent is confirmed as dollars per kilowatt per month, not an annual rate. This is the single most common unit confusion on this model — a $275 figure means $275/kW/month, which is very different from $275 per year. The Skill asks for clarification every time a rent number is provided.
Timing inputs are month indices relative to the analysis start date, not calendar dates. The Skill confirms the convention and converts any calendar-based language the user provides.
Catching Common Mistakes
Seven errors show up repeatedly on this model. Confusing $/kW/month with an annual rent figure. Entering facility MW directly instead of setting the kW loads in the tenant table. Writing to derived cells (net rentable SF, grey space, analysis-end month, construction period) that mirror other inputs and will simply recalculate back. Forgetting to click the yellow Recalculate button after input changes — without it, construction-loan interest and the management fee are stale, and every return metric downstream is wrong. Changing the IT load without revisiting the PUE assumption, which drives both utility expense and the total facility power display. Using calendar dates for timing inputs when the model expects month indices. And entering cap rates, interest rates, or LTC as percentages rather than decimals.
The Recalculate Button — A Critical Note
This model carries circular references — construction-loan interest depends on the loan balance, which depends on the interest — resolved by a VBA macro behind the yellow Recalculate button at the top of the Underwriting tab. Normal Excel recalculation does not resolve them, and AI tools operating via code execution cannot run the macro at all. After any input change, the Skill will remind you to open the file in desktop Excel and click Recalculate before reading any return outputs. This step is not optional.
Framing Outputs in Your Role
For a developer, the headline is the development spread — the gap between yield-on-cost and the exit cap rate. That spread is the answer to “does this build create value?” The Skill also distinguishes between trended and untrended stabilized values, because the two tell very different stories about the deal’s dependence on market assumptions. For a lender, the conservative anchor is the untrended stabilized value as collateral — not the trended figure. For an LP, the stress test is the exit cap sensitivity and the lease-up pace, both of which the Skill can flex conversationally.
Operating Contexts (Chat / Cowork and Claude in Excel)
The Skill works in two environments. You can upload the Excel file to a Claude conversation and have Claude populate inputs via code execution — the Skill bundle includes a clean copy of the model so no upload is needed to get started. Or, if you’re using Claude in Excel, operate the model live with Claude reading and writing to the workbook directly. In both environments, the yellow Recalculate button must be clicked by the user after input changes — the circular-reference macro cannot be triggered by the AI. And as noted earlier, the Skill is also portable to other AI assistants, though the integration may be lighter.
A Note on the Underlying Model
The Data Center Development Model is a monthly-period, ground-up data center development tool. It takes a development budget with construction and mezzanine financing, a tenant-by-tenant colocation lease-up priced on $/kW/month against IT loads, a PUE assumption, operating expenses scaled to delivered power, and a stabilized exit to output unlevered and levered IRR, equity multiple, net profit, yield-on-cost, development spread, and trended and untrended stabilized valuations. It covers a single asset from development through disposition — no partnership waterfall, no portfolio roll-up, no acquisition of an operating data center. The model is currently in beta. See the model post for the full walkthrough.
Note: This AI Skill is built for beta v1.6 of the model. If you’re on an older version, confirm key cell positions before running the Skill. The model is macro-enabled (.xlsm) — macros must be enabled in Excel for the Recalculate button to function.
Video Walkthrough: Using the AI Skill
The video below walks through the full AI-assisted workflow: loading the Skill, selecting your role, populating the development budget, tenant lease-up, and financing inputs from a deal summary, and interpreting the development spread and return outputs in the context of the investment decision.
Before You Use This AI Skill with the Data Center Development Model
A couple of notes worth surfacing before you download.
Who this Skill is for. This Skill is built for real estate professionals with a strong grasp of financial modeling — and ideally prior exposure to development underwriting, data center fundamentals, and the capital stack mechanics of ground-up digital infrastructure. It’s best suited to graduates of our A.CRE Accelerator real estate financial modeling program, or analysts comfortable building development models from scratch. AI assistants make mistakes; the Skill assumes an analyst on the other side who can catch them. Treat its output the way you’d treat work from a sharp junior analyst — useful, fast, and always verified before it informs an investment decision.
Macros required. The model is macro-enabled (.xlsm). Macros must be enabled in Excel for the yellow Recalculate button to function. Without it, construction-loan interest and the management fee will not resolve correctly, and return outputs will be wrong.
License. The Skill is distributed under the A.CRE software license, with full terms in the LICENSE.txt file included in the bundle. The short version: use it for personal, organizational, and client-facing analysis; don’t resell or redistribute it. Use by an AI assistant operating on your behalf is expressly permitted — that’s the whole point.
Download the Data Center Development Model + AI Skill
To make this model accessible to everyone, it is offered on a “Pay What You’re Able” basis with no minimum (enter $0 if you’d like) or maximum (your support helps keep the content coming – typical Excel models sell for $100 – $300+ per license). Just enter a price together with an email address to send the download link to, and then click ‘Continue’. If you have any questions about our “Pay What You’re Able” program or why we offer our models on this basis, you can contact our support team here.
Your download includes three files: the Excel model, the AI Skill (.skill file), and a short README explaining how to use them together. The Skill bundle includes a clean copy of the model — no upload needed to get started with Claude.
We regularly update both the model and the AI Skill (see version notes below). Paid contributors receive a new download link via email each time either is updated.
Frequently Asked Questions about the AI Skill for the A.CRE Data Center Development Model
Version Notes – AI Skill
beta v1.6
- Initial release of the AI Skill for the A.CRE Data Center Development Model
- Paired with beta v1.6 of the Excel model
- Supports both Chat / Cowork (Skill bundle includes a clean copy of the model — no upload needed) and Claude in Excel (operate the live workbook directly)
- Includes 4-role triage (developer/sponsor, equity/LP investor, development-debt lender, student/Accelerator member), conversational input population from development budgets, GC estimates, and colocation lease terms, and mistake-catching across $/kW/month unit confusion, kW-vs-MW entry, derived cell writes, Recalculate button state, PUE alignment, timing conventions, and decimal-vs-percent entry
- Portable to other capable AI assistants (ChatGPT, Gemini, etc.) via the SKILL.md file
