---

Bottom Line

The PRZC Research bearish datacenter call (Oct 2025, 40–60% correction in 18–24 months) has not played out in equity prices over the first six months. Data center stocks have broadly appreciated. However, every structural element of the thesis has deepened rather than resolved. The call is not wrong — it is early. The most probable correction window has shifted right to 2027–2028.

Six-Month Scorecard

Thesis Element	Status	Key Fact
CapEx acceleration as bubble signal	CONFIRMED	Big-5 capex now $600–690B in 2026, +36% YoY
Revenue-capex monetization gap	DEEPENING	~$25B AI revenue vs. $450B+ AI infra spend
Leverage financing of buildout	CONFIRMED	$108B in hyperscaler debt raised in 2025; Amazon FCF going negative
Physical constraints materializing	EMERGING	50% of planned data centers at risk of delay; PJM 6GW short in 2027
40–60% equity correction	NOT YET	EQIX +10%+, DLR +48% TTM, VRT +64% YTD March 2026
Project cancellations at top	PARTIAL	Microsoft paused/canceled 2GW; Stargate site canceled

Disconfirming Evidence (Leading with CBOM Principle)

1. CapEx re-accelerated — $335B in 2025 → $600–690B in 2026. Not the plateau the thesis expected as near-term signal.
2. Equity prices up materially — A short in EQIX, VRT, or DLR from Oct 2025 is deeply underwater in March 2026.
3. Demand signals intact — Occupancy rates projected 95%+ late 2026; IRM digital segment +30% YoY.
4. DeepSeek did not change hyperscaler plans — Meta and Microsoft explicitly reaffirmed capex commitments after the Jan 2026 DeepSeek shock.
5. Analyst consensus remains constructive — S&P Global: 80% of respondents expect valuations to remain high 2 more years.

Confirming Evidence

1. Monetization gap widening — Vanguard model requires $3.1T cumulative AI revenue 2025–2027 to justify valuations; current trajectory is far short.
2. Leverage becoming structural — $1.5T in projected hyperscaler debt issuance; capex-to-revenue ratios of 45–57%, unprecedented in tech.
3. Power constraint is real and binding — Fortune (18 Mar 2026): "A bend in the trajectory." Only 1/3 of pipeline projects under active development; PJM 6GW reliability shortfall by 2027; grid connection lead times 24–72 months.
4. Microsoft 2GW lease cancellations — The single most important confirmatory signal. The largest enterprise cloud operator returning previously contracted capacity = first visible supply-demand mismatch in physical infrastructure.
5. Project-level failures accumulating — $64B in projects blocked by local opposition; OpenAI Stargate site canceled; only 5GW of planned 16GW 2026 pipeline under active construction.
6. Goldman Sachs now citing 2027 oversupply risk — Sell-side is beginning to price in the correction scenario identified by PRZC in Oct 2025, on a 12-month lag.
7. Technology obsolescence risk — JLL (2026): 5–7 year facility obsolescence cycle driven by GPU generational churn.
8. Bipartisan political opposition — Sanders + DeSantis both publicly opposing data center expansion (CNBC, 01 Jan 2026); regulatory risk vector hardening.

Hyperscaler CapEx vs. AI Revenue (2026E)

Hyperscaler	2026E CapEx	YoY Change	FCF Risk
Amazon (AWS)	$200B	~+33%	Negative FCF projected
Alphabet / Google	$175–185B	~+43%	Strained
Microsoft (Azure)	$71B	+15%	Absorbing
Meta	$115–135B	~+60%+	Leverage building
Oracle	$50B+	n/a	Debt-funded (Stargate)
Big-5 Total	$611–641B	+36% avg.	Structurally mismatched

AI-specific infra: ~$450B (2026). AI services revenue: ~$25B (2025 est.). Gap: 18:1 ratio.

Updated Probability Distribution

Scenario	Oct 2025	Mar 2026	Description
Bear Case	45%	40%	AI revenue misses; debt cycle turns; 40–60% correction in data center equities 2027–2028
Base Case	35%	40%	Demand plateaus 2027–2028 as power binds; 20–35% valuation compression; sector re-rates to utility multiples
Bull Case	20%	20%	Agentic AI inference absorbs all capacity; FCF validates debt; no correction

Combined Bear + Base: 80%. Primary change: 5pp shift from Bear to Base (right-shift in timing), reflecting near-term demand strength. Bull case probability unchanged.

Revised Position Statement

• Original call: 40–60% correction within 18–24 months of Oct 2025 (by Q2–Q4 2027)
• Revised call (March 2026): Timeline's outer bound of Oct 2027 remains operative. Trigger has shifted from near-term demand collapse to 2027 monetization reckoning — when multi-year capex cycles are reported against underwhelming AI revenue.
• PRZC Rating: UNDERWEIGHT — maintained. Sector requires bull scenario with high confidence to justify current multiples. Short-side positioning premature given near-term demand; patience for 2027 inflection is operative posture.

Report Status

• Category: Infrastructure / AI
• Outlook Badge: THESIS UNDER PRESSURE — TIMELINE INTACT
• Report Code: T32 | Follow-up Brief | Format C