Big Tech's AI Build-Out Has a Transparency Problem

AI Data Centers Are Rewiring Construction-But the Money Trail Is Blurry

AI is driving an unprecedented build cycle for data centers, but the accounting obscures where the money actually goes. Most tech giants lump shells, power, cooling, and short-lived AI chips into one construction-in-progress (CIP) bucket. That bucket isn't depreciated until assets go live-masking what's long-term infrastructure versus gear that may be swapped in a couple of years.

If you're in real estate or construction, that lack of detail hits everything from contracts to cash flow. You need clearer splits to protect margins, reduce change orders, and keep projects moving when chip plans shift midstream.

The accounting blind spot: construction-in-progress

Companies report big CIP numbers without breaking out buildings, MEP, networking, and GPUs. Buildings might last 20-40 years. Many AI chips could be obsolete in under three.

Investors are pushing for more granularity because risk profiles differ. You should too-because payment terms, warranties, and refresh cycles are not the same for concrete and steel versus accelerators and racks.

• What gets buried in CIP: site acquisition, shell, power/cooling, networking, racks, GPUs, and owner-furnished equipment (OFE).
• Why it matters to builders: mismatched useful lives can delay acceptance, create scope creep, and hold up retention if not separated up front.

The scale-and the squeeze

Big Tech's CIP tied to AI infrastructure is enormous. One company disclosed roughly $50.6B of assets not yet in service; another booked about $46.4B; a third showed $26.8B. In several cases, CIP represents a large slice of net PPE. One major player reported $32.1B in commitments for data center build and improvements-note: commitments don't always equal near-term spend.

Meanwhile, power is the bottleneck. Many projects face delays from interconnect queues, substation lead times, and local politics. Expect starts and stops-and investors watching for any hint of slippage.

Check public filings (EDGAR) for shifts in capex timing and disclosures around assets not yet in service.

Useful life changes = financial optics

Several public companies extended the useful lives of servers and network equipment in 2024-the highest pace in years. Longer lives reduce depreciation expense and boost reported profit. The debate: Will AI chips really remain productive long enough to justify those estimates?

For construction, this translates into more frequent equipment refresh allowances, tighter OFE sequencing, and re-commissioning cycles. You can't let a three-year hardware horizon derail a 30-year building.

What real estate and construction teams should do now

• Split the scope and budget by useful life. Separate shell/core, power, and cooling from racks, networking, and GPUs. Tie payment milestones to building systems; use distinct milestones for OFE delivery/installation.
• Make OFE explicit. Include a serialized equipment schedule, delivery windows, storage conditions, and acceptance criteria. Add a change-order protocol for GPU substitutions and density shifts.
• Design for churn. Build flexible MEP pathways, scalable power blocks, and hot/cold aisle containment that can convert to liquid cooling. Plan floor loading for denser racks and future CDUs.
• Power strategy first. Lock interconnect timelines early, pursue dual feeds where feasible, and evaluate on-site generation or temporary power. Coordinate substation design with utility long-lead items on day one.
• Phase delivery. Use modular plants and phased white space to bring capacity online while GPUs trail. Enable concurrent fit-out without rework of base systems.
• Warranties and turnover. Separate warranty clocks for building systems versus electronics; deliver a commissioning package that anticipates future refresh cycles and partial re-commissioning.
• Financial hygiene. Provide owners with a project-level CIP breakout by component, lead time, and expected life. Align draw schedules with the long-lived assets to minimize disputes if chip deliveries slip.
• Entitlements and community. De-risk approvals early with clear narratives on water use, heat reuse, and tax impact. Power plus politics is where many projects stall.
• Cost segregation and leases. For owners and developers, consider cost segregation studies, sale-leasebacks, and clauses that shift obsolescence risk for short-lived gear away from the base building.

Signals investors will watch (so you should too)

• Changes in "assets not yet in service" and CIP as a share of total PPE.
• Useful life updates for servers/networking versus building systems.
• Power interconnect dates, substation progress, and regional capacity constraints.
• GPU lead times, delivery risk, and supplier concentration.
• Commitments vs. cash spend and any commentary on delays or rephasing.

FASB resources on PPE disclosures can help your finance team frame more decision-useful reporting for stakeholders.

What this means for your pipeline

Expect more shells started, with fit-out paced by power and GPU availability. Labor planning needs flexibility; so do procurement and storage for OFE. Liquid cooling, higher densities, and rework for future generations are not edge cases-they're standard assumptions now.

Bottom line

The cash is flowing, but the categorization is murky. If you separate long-lived infrastructure from fast-moving tech in contracts, design, and reporting, you accelerate pay, cut disputes, and keep projects moving-even when chip plans change mid-build.

If you're upskilling teams to speak AI fluently across development, PM, and owner's rep roles, explore practical courses by job at Complete AI Training.