🔒 Private preview · Capex Watch
A standing fragility read on the AI infrastructure build-out. Six corroborating indicators, a five-layer structural map, and the exhibits the consensus skips — every figure either traced to a primary filing or labelled as attributed, every gap named rather than hidden.
The convergence scorecard · NVIDIA
Higher = more fragile. Gold = elevated · slate = contained.
Four of six indicators are elevated — capex outrunning demand, circular financing, soft end-user demand, and stretched depreciation. They share a common driver (capex ahead of monetization), so we weight their agreement as corroboration, not as four fully independent votes. The two we deliberately leave contained — insider selling and energy — keep the read honest.
NVIDIA's full profile in Markets → · Browse all 68 companies →
The divergence gauge · D(t) = M(t) − G(t)
A market signal (SOXX momentum, overextension, instability) set against a ground-truth signal (AI-layoff share, insider selling, the capex gap). Through early 2025 they track; in 2026Q2 the gap inverts hard — D(t) swings from −1.80 to +4.06 as price detaches from fundamentals. The strongest single-quarter divergence in the series.
Full-window z-scores · descriptive. Source: SOXX daily + ground-truth series — reproducible from the brief.
The recycling loop
Compute committed out of the core labs (OpenAI, Anthropic, xAI) divided by equity put in, across provenance tiers. The same disclosed dollar of equity supports roughly 26× committed compute on a funded-cash basis, easing to ~21× present-valued — far above any arm's-length benchmark.
Provenance tiers (funded cash → filed → +reported → PV-adjusted). Source: financing_edges.csv — reproducible from the brief.
Exhibit 01 · the money carousel
The money carousel
Capital recirculates among a tight loop — Nvidia, the hyperscalers, OpenAI, Oracle, CoreWeave, Anthropic — where vendors finance the customers who buy their product. When a vendor funds the customer that buys its product, reported revenue and demand stop meaning what they appear to mean. The question isn't whether AI is growing; it's whether this is genuine end demand or one of the largest circular capital flows in tech history.
Indicator I4 · circular leverageExhibit 02 · the unmarked cost
Useful-life accounting · from the 10-K filings
Amazon shortened a subset of servers and networking from 6 to 5 years, effective Jan 1 2025, citing AI/ML obsolescence verbatim — a change cutting 2025 operating income by ~$0.7B. Separately, it took ~$920M of accelerated depreciation in Q4 2024 retiring AI-exposed gear early. That reversal is the canary: the first time a hyperscaler names AI to shorten asset lives.
Meta extended useful life to ~5.5 years (Jan 2025) — about −$2.9B of FY2025 depreciation, flattering earnings. Microsoft, Google and Meta extended useful lives and held them; Amazon alone reversed, shortening for AI-exposed gear — the contrarian tell.
Stretch the assumed life and today's earnings look better than the cash can support. Michael Burry (Scion) estimates understated depreciation near $176B across 2026–2028 — large enough to flatter the very earnings base that justifies the capex. We carry it as his allegation: a disclosed short-seller's projection, not an audited figure. The canary, by contrast, is in a filed 10-K.
Assumed server useful life · years
Three of four stretched to six years; Amazon alone cut to five — the contrarian tell. Longer life = lower reported depreciation = flattered earnings.
Amazon FY2024 10-K · Meta FY2025 · $176B — Burry/Scion allegation, Nov 2025The structural map · where the weight sits
Concentration, not fragility for its own sake: the broad market rests on a compute base only a handful of names wide. That base is the sturdiest, highest-margin layer — but it is also the single point of correlation. A shock above it — demand, financing, or depreciation — arrives at the base as one event, with little breadth left to absorb it.
Every indicator and exhibit feeds one measurement: the race between the productivity lag and the financing runway. Can the spending survive long enough for the payoff to arrive?
Read the founding teardown → Download the full fragility brief (PDF) ↓