By Tony Grayson | Independent Strategic Advisor, AI Infrastructure & Defense | Top 10 Data Center Influencer | Former SVP Oracle, AWS & Meta | U.S. Navy Nuclear Submarine Commander, USS Providence (SSN-719) | Stockdale Award Recipient | Veterans Chair, Infrastructure Masons

Published: November 30, 2025 | Updated: February 20, 2026 | Verified: February 20, 2026

TL;DR

The golden era of the 15-year hyperscale lease is over. AI hardware refresh cycles now move faster than the building economics designed to support them. A 3-year tenant in a 30-year building creates stranded capital when the tenant leaves — retooling costs $40-60M on a $150M asset.

The fix: Stop building bespoke data halls. Start building Universal Docks — long-life shells with standardized utility interfaces that accept any vendor's skid-based power and cooling. The OCP is already standardizing the interfaces (Project Mount Diablo for power, Project Deschutes for cooling). The playbook is here. The market will reward it with tighter cap rates.

"Physical agility is the only true hedge against volatility." — Tony Grayson

A Note on Perspective: The prevailing consensus among data center developers is that Build-to-Suit remains the gold standard for large-scale AI infrastructure. Hyperscalers with decade-long visibility do justify bespoke construction. This post doesn't dispute that. What it challenges is applying the same model to a market where net new demand is shifting toward AI Training (3-year cycles), Inference (churny, distributed), and Enterprise Repatriation (flexibility-first). These segments have fundamentally different temporal requirements. Applying a 30-year financial model to a 3-year hardware cycle creates a structural mismatch — not a financing quirk.

Key Concepts

Universal Dock: Tony Grayson's term for a long-life data center shell with standardized utility interfaces that can accept any vendor's skid-based power and cooling. The Exoskeleton (30-year shell) provides physical security and a Utility Bus. The Cartridges inside (3-year AI hardware) are swappable without structural modification. Like a shipping dock that accepts any standardized container.

Exoskeleton: The shell component of the Universal Dock. A durable structure providing only physical security and a Utility Bus — main power and water connections. Nothing tenant-specific is embedded. Everything inside is a Cartridge.

Cartridge: The swappable computing module inside the Exoskeleton. Built on volumetric skids with standardized OCP interfaces. Contains power and cooling for a specific AI hardware generation. When the hardware generation changes, the Cartridge swaps — not the building.

Lease Mismatch: The temporal misalignment between AI hardware refresh cycles (18-36 months) and data center real estate underwriting (15-30 years). When a tenant's hardware generation ends and they leave, bespoke infrastructure requires $40-60M to retool — destroying IRR.

Industrialized Construction: Moving construction complexity from on-site stick-built to factory-based, modular, standardized manufacturing. Achieves precision and speed that site construction cannot. Not the container fallacy — volumetric skids creating massive open white space at manufacturing quality.

Project Mount Diablo: OCP initiative led by Meta standardizing 400V disaggregated power delivery. The critical enabler for Universal Dock power interfaces.

Project Deschutes: OCP initiative led by Google standardizing liquid cooling interconnects. Tony Grayson identifies this as the hardest problem — cooling requirements vary most dramatically between AI hardware generations.

The golden era of the 15-year hyperscale lease is rapidly evolving into something far more volatile. We have moved from 10–15-year, single-tenant data halls to 3–5-year, hardware-driven deployments.

"The lease used to outlive the gear. Now the AI hardware roadmap dictates the lease."

— Tony Grayson, Former SVP Oracle, AWS & Meta | Top 10 Data Center Influencer

For hyperscalers with decade-long visibility, "Build-to-Suit" still has a place. However, that population is shrinking relative to the market's total volume. Net new demand is shifting toward AI Training (short-term, massive density), Inference (distributed, churny), and Repatriated Enterprise workloads. Companies are realizing public cloud isn’t the only answer for steady-state AI but still demand flexibility.

The problem?

"We are building static monuments for a fluid market."

— Tony Grayson, Founder & Exited Top-10 Modular Data Center Company

With the arrival of the NVIDIA GB200 NVL72 and the massive density spike that comes with it, we face a hard truth: If you build a bespoke hall for today’s AI requirement, you are risking stranded capital tomorrow. As I wrote earlier, we are seeing signs of a Vendor Financing Trap where infrastructure spend is outpacing sustainable demand. In this environment, you need an Industrialized Data Center Strategy that prioritizes agility over monuments.

We need to stop building data halls and start building "Universal Docks."

The "Universal Dock": Rethinking Data Center Industrialization

By "Universal Dock," I mean a long-life shell with standardized power and liquid interfaces that can accept any vendor’s skid-based power and cooling. This concept is central to a robust Industrialized Data Center Strategy.

The "Lease Mismatch" and IRR Destruction

The risk profile has fundamentally shifted. If you hard-code a facility for a specific high-density tenant who leaves after a single hardware generation, the economics collapse.

Consider this scenario: You build a $150M Build-to-Suit (BTS) hall with a 7-year underwriting horizon. The tenant signs a 3-year lease for a specific AI cluster. When they exit, retooling the embedded plumbing and rigid power distribution can cost $40–60M.

This leads to Capital Inefficiency and IRR destruction.

The Real Estate Case: Valuation & Liquidity

A forward-thinking Industrialized Data Center Strategy focused on the Universal Dock concept changes the asset class:

• Tighter Cap Rates: Assets that can pivot rapidly between AI Training and Standard Cloud without costly demolition have lower re-tenanting risk and shorter downtime (weeks vs. months). The market will reward this liquidity with tighter cap rates.

• Redeployable Assets: By moving Tenant Improvements into skid-based modules, we shift from "write-off when tenant leaves" to "redeployable inventory." Lenders and rating agencies will ultimately price this differently than sunk concrete costs. This approach also minimizes the risk of Business Continuity Plan failures during tenant transitions.

The Tenant’s Perspective: Why Should You Care?

If you are an AI tenant, you might ask: "Why do I care about your IRR?" You should care because fungibility equals velocity:

• Speed to Deploy: You don't wait 18 months for a bespoke build; you plug into a dock that’s already permitted and energized.

• No Exit Penalties: You aren't amortizing a concrete monument over a 3-year lease.

• Roadmap Insurance: If NVIDIA or AMD shifts the cooling spec in 2027, you aren't stuck in a facility hard-wired for 2025. You just swap the skids.

The Exoskeleton Approach: The Future of Industrialized Data Centers

The solution is to separate the shell from the cell. We must treat the shell as an Exoskeleton: a durable, long-life structure providing only physical security and a "Utility Bus" (main power/water). Everything inside, like the "data hall" itself, must be a swappable cartridge.

This is not the "container fallacy" of poor ergonomics. This is Industrialized Construction, creating massive, open white space using volumetric skids that plug and play.

In the nuclear navy, we learned that Contextual Intelligence is everything. You cannot apply a static framework to a dynamic battlefield. The same applies here: You cannot apply stick-built construction methods to a dynamic AI hardware market. Moving complexity to the factory floor (Industrialized Construction) is the only way to achieve the precision and speed required.

The Standardization Imperative

For this to work at scale, we need standardization. We are seeing real traction:

• Power: Project Mount Diablo (led by Meta within OCP) is driving standards for 400V disaggregated power delivery.

• Cooling (Hardest Problem): Project Deschutes (led by Google within OCP) is attempting to standardize liquid cooling interconnects. This is the highest-risk area. If we don't standardize the plumbing interface, we risk a proprietary lock-in war that benefits no one.

• Grid: OCP’s roadmap now includes grid interactions and BESS, this is critical because the substation connection must outlive the first tenant.

The Playbook: Navigating the Industrialized Shift

We are likely 2–3 years away from this being table stakes. Here is how we get there:

• Developers: Stop pouring complexity into the slab. Design shells as Universal Docks with "USB-like" utility headers.

• Tenants: Demand fungible shells. Refuse to pay for the demolition of the previous tenant's mistakes.

• OEMs: Build against OCP-style interfaces so your skids are plug-compatible across different developers' sites.

The days of the static data hall are numbered. We have to stop building "perfect" data centers for today's specs. As I often say, you will die with one of two things: a collection of scars or a collection of "what-ifs." Read more on The Math on Regret. Sticking to the "safe" Build-to-Suit model because it worked for the last decade is a recipe for a "what-if" when the market leaves you behind.

"Physical agility is the only true hedge against volatility."

— Tony Grayson, Former Commanding Officer, USS Providence (SSN-719)

Frequently Asked Questions: Industrialized Data Centers & the Universal Dock Strategy

Who is Tony Grayson and what is his background in data center infrastructure?

Tony Grayson is a former U.S. Navy Nuclear Submarine Commander (USS Providence SSN-719), Stockdale Award recipient, and veteran of SVP-level leadership at Oracle ($1.3B budget and 1,000+ person team), AWS, and Meta. He founded and exited a top-10 modular data center company — one of the first to operationalize industrialized construction at scale — before transitioning to independent advisory work. He is a Top 10 Data Center Influencer. His unique perspective on data center strategy comes from having built hyperscale infrastructure at Fortune 500 scale and from nuclear submarine command, where modularity and redundancy are not design preferences — they are survival requirements. More at tonygrayson.ai

What is an Industrialized Data Center Strategy?

An Industrialized Data Center Strategy involves moving significant portions of data center construction from traditional stick-built methods on-site to factory-based, modular, and standardized manufacturing. The goal is to achieve greater speed, precision, quality, and cost-effectiveness in building highly adaptable infrastructure. It treats data center components as manufactured products rather than custom construction projects—moving complexity to the factory floor where precision and repeatability are achievable.

What is the Universal Dock concept?

The Universal Dock refers to designing a data center shell as a long-life, power-agnostic structure (the "Exoskeleton") with standardized utility interfaces that can accept interchangeable, skid-based, high-density computing modules (the "Cartridges"). This enables rapid swapping of AI hardware generations and tenant types—like a shipping dock that can accept any standardized container. The shell provides "USB-like" utility headers for power and cooling.

Why is Build-to-Suit becoming problematic for AI infrastructure?

Build-to-Suit is problematic because it creates static, highly customized facilities (e.g., specific liquid cooling configurations) for a market driven by rapidly evolving AI hardware with 18-36 month refresh cycles. This leads to a "lease mismatch" where expensive, specialized infrastructure becomes obsolete and costly to retool ($40-60M) long before the building's 30-year lifespan ends. As discussed in the Vendor Financing Trap, infrastructure spend is outpacing sustainable demand.

What is the 3-Year Tenant in a 30-Year Building problem?

This phrase highlights the temporal misalignment between short-term AI hardware lifecycles (requiring new facilities every 3-5 years as GPU generations change like the NVIDIA GB200 NVL72) and traditional data center real estate investments (underwritten over 15-30 years). When a tenant leaves after one hardware generation, retooling embedded plumbing and rigid power distribution can cost $40-60M, destroying IRR and creating stranded capital.

How does OCP standardization support the Universal Dock strategy?

The Open Compute Project (OCP), through initiatives like Project Mount Diablo (400V disaggregated power delivery led by Meta) and Project Deschutes (liquid cooling interconnects led by Google), is working to formalize open standards for data center components. This standardization is crucial for achieving true plug-and-play fungibility across different vendors and developers, preventing proprietary lock-in.

What is the Exoskeleton approach to data center design?

The Exoskeleton approach separates the shell from the cell. The shell is a durable, long-life structure providing only physical security and a "Utility Bus" (main power/water connections). Everything inside—the "data hall" itself—becomes a swappable cartridge. This is industrialized construction using volumetric skids that plug and play, not the "container fallacy" of poor ergonomics.

What is Project Mount Diablo?

Project Mount Diablo is an Open Compute Project initiative led by Meta driving standards for 400V disaggregated power delivery in data centers. At the 2025 OCP Summit, the industry showed convergence toward 800V DC power distribution, reducing copper use and energy losses at megawatt rack scales. It's part of the standardization imperative needed for Universal Dock architecture to work at scale.

What is Project Deschutes?

Project Deschutes is an OCP initiative led by Google attempting to standardize liquid cooling interconnects—the highest-risk area for the Universal Dock concept. Without standardized plumbing interfaces, the industry risks a proprietary lock-in war that benefits no one. The 2025 OCP Summit showed convergence on standardized trays, manifolds, and sidecar coolers that shorten integration cycles and simplify retrofits.

How does lease mismatch destroy IRR?

Consider this scenario: You build a $150M Build-to-Suit hall with a 7-year underwriting horizon. The tenant signs a 3-year lease for a specific AI cluster. When they exit after one hardware generation, retooling the embedded plumbing and rigid power distribution costs $40-60M. This capital inefficiency destroys the expected Internal Rate of Return and creates stranded assets.

Why should AI tenants care about the Universal Dock concept?

Fungibility equals velocity. Benefits include: Speed to Deploy (plug into a dock already permitted and energized vs. 18-month bespoke build), No Exit Penalties (not amortizing a concrete monument over a 3-year lease), and Roadmap Insurance (if NVIDIA or AMD shifts cooling specs in 2027, you swap skids rather than being stuck in a facility hard-wired for 2025). This aligns with Contextual Intelligence—applying dynamic solutions to dynamic problems.

What is the difference between hyperscale Build-to-Suit and the new AI market?

For hyperscalers with decade-long visibility (AWS, Google, Microsoft), Build-to-Suit still has a place. However, that population is shrinking relative to total market volume. Net new demand is shifting toward AI Training (short-term, massive density), Inference (distributed, churny), and Repatriated Enterprise workloads where companies realize public cloud isn't the only answer for steady-state AI but still demand flexibility over monuments.

How does the Universal Dock improve data center valuation and liquidity?

Assets that can pivot rapidly between AI Training and Standard Cloud without costly demolition have lower re-tenanting risk and shorter downtime (weeks vs. months). The market rewards this liquidity with tighter cap rates. By moving Tenant Improvements into skid-based modules, costs shift from "write-off when tenant leaves" to "redeployable inventory"—lenders and rating agencies price this differently than sunk concrete. This also minimizes risk of Business Continuity Plan failures during tenant transitions.

Related Articles from Tony Grayson:

• Nvidia Vendor Financing Trap: The Infrastructure Risk Nobody's Pricing — Why vendor financing is accelerating the lease mismatch problem

• Is AI Infrastructure Overbuilt? An Operator's View on the $100B Gamble — The capital efficiency question behind hyperscale overbuilding

• AI Bubble 2025: Why Herd Mentality is the Real Investment Trap — The psychology driving synchronized AI infrastructure investment

• From Parameters to Physics: Why Watts per Token is the Only Metric for Industrial AI — Where AI infrastructure demand is actually heading

___________________________

Tony Grayson is a recognized Top 10 Data Center Influencer, independent strategic advisor on AI infrastructure, nuclear energy, and defense technology, and a former U.S. Navy Nuclear Submarine Commander.

A recipient of the prestigious VADM Stockdale Award for Inspirational Leadership, Tony commanded USS Providence (SSN-719) before leading global infrastructure strategy as Senior Vice President at Oracle, AWS, and Meta. He founded and exited a top-10 modular data center company — one of the first to operationalize the industrialized construction approach described here.

He serves as Veterans Chair for Infrastructure Masons, helping veterans transition into technology careers.

Read more at tonygrayson.ai