AI Capex & Earnings Bubble Simulator

Stress-test what happens to hardware suppliers, hyperscaler cash flows, and Wall Street deal fees when the AI infrastructure build-out peaks and enters a digestion cycle.

Quote thesis: “There's a good chance we're in an earnings bubble because this build-out is going to end at some point...” — Stanley Druckenmiller

Scenarios:

Cycle Levers

DYNAMIC ENGINE

Baseline Annual Capex $210B Aggregate big-tech AI data center, power, silicon & optics spend (Year 1). Build-out Crest Year Year 2 The year capex growth hits ceiling before hardware digestion kicks in. Post-Peak Capex Contraction -35% Drop in infrastructure spend once primary cluster build-out saturates. End-User Software Revenue Growth +35% / yr Real enterprise & consumer software revenue converting from compute usage. Cluster Useful Lifespan 3 Years Amortization window for GPUs, custom ASICs, and optical switches. Circular / Vendor Funding Share 22% Portion of cloud compute paid by venture/equity-backed labs back to hyperscalers. Wall St. Advisory & Underwriting Pool $28B / yr Annual investment bank fees from AI IPOs, convertibles, debt, and M&A.

Peak-to-Trough Capex $262B → $170B Year 2 Peak • -$92B Digestion

Supplier Margin Shock -48.2% Operating profit swing for chip/fab supply

Cumulative Monetization Gap -$284B Cumulative Capex + D&A vs Software Revenue

Bank Deal Fee Contraction -42.0% Post-crest equity & debt underwriting

5-Year Multi-Horizon Trajectory Model Year 3 Focus

Capex ($B) Cluster D&A Expense ($B) Downstream Revenue ($B) Circular Vulnerability ($B)

Macro Domino Cascade Analysis (At Cycle Trough) Phase: Hardware Digestion

1. Hyperscalers D&A Lag

$87B/yr

Amortization headwind lingers 2+ years after capex cuts.

2. Semi / Supply Operating Deleveraging

-38% EPS

Fixed fab and wafer commitments compress gross margins.

3. Circular Loop Cloud Credit Squeeze

$37B at risk

Startup equity trades back into cloud credits slow down.

4. Wall St. Fees Financing Pause

-$11.8B

M&A and debt underwriting dried up as multiple compresses.

Metric ($B)	Y1 (Base)	Y2	Y3	Y4	Y5

Simulation active: Severe earnings vulnerability detected in Year 3-4.

The Anatomy of an AI Build-Out Digestion

Stanley Druckenmiller's thesis rests on a classical capital goods cycle: infrastructure build-outs (railroads in the 1870s, fiber-optic telecoms in 1999, data centers in 2024–2026) are front-loaded. Companies build ahead of demand to avoid being constrained by chip scarcity or power interconnect delays.

When hyperscaler data center capacity reaches parity with customer workloads, capex doesn't just stop growing—it contracts. Because semiconductor foundries and equipment makers trade at peak multiples on peak margins, even a modest 25% reduction in capex triggers massive operating deleveraging.

The Circularity & Banking Exposure

Two amplifiers distinguish the current cycle from standard semiconductor downcycles:

Circular Financing Loops: Billions in corporate venture investments from hyperscalers into foundation model developers are contractually spent right back on cloud computing credits, artificially inflating software demand.
Wall Street Underwriting Dependency: Commercial & investment banks generate record investment banking revenue underwriting debt, convertible notes, and IPOs for pre-revenue AI infra plays.
Depreciation Tail: While capex cuts instantly improve free cash flow, the 3-year depreciation of previous capex crushes reported GAAP operating income for years afterward.