The AI data-center money machine is stalling — short the indebted AI infrastructure names

Verdict: The short thesis reads well, but the tradable evidence isn't there yet — wait

The idea's credit-cost argument has a real anchor: per Bloomberg (2026-10-20), tech credit-insurance costs hit fresh highs just as a borrowing wave looms, and EQIX's free cash flow has now been negative for four straight quarters through Q1 2026, landing at the 1.4th percentile of 833 IT-sector peers. But the strongest point against is that the most recent filings cut the other way — EQIX's Q2 2026 revenue rose 7.4% to $2.6B with operating margin up to 25.3%, and DLR's Q1 2026 net income jumped 74% with net margin expanding to 11.0% — neither company is visibly cracking like NextDC did per Reuters (2026-10-21). On the tradable side, the completed backtest of this exact rule set shows only 6 trades in 60 months, a 16.7% win rate, a 6.6% total return and a 10.6% maximum drawdown, and the trailing 12 months produced no triggers at all. Nothing is live today: DLR's trend-strength reading of 21.4 still needs to clear 22.0, and EQIX sits at 18.3, 3.7 points away. Notably, the published rules fire long while the idea argues short, so the bearish case rests on the fundamentals alone. The verdict would flip if the next quarterly filings (periods ending December 2026, due early 2027) show the margin expansions reversing — or if EQIX's $16.1M net insider selling from the provisional September 30, 2026 ownership filings escalates into sustained open-market selling.

Trade now

Nothing is triggered yet — this setup is waiting. DLR closed at $175.31, already below its 125-day average of $183.21 (condition met), but trend strength reads 21.4 against the required 22.0 (a small gap) and the MACD cross needs to confirm. EQIX closed at $987.44, comfortably below its 125-day average of $1,020.27 (condition met), but its trend-strength reading of 18.3 is still about 3.7 points away from the 22.0 trigger, so it is the laggard. "Wait" concretely means: do not enter until trend strength clears 22.0 and a fresh MACD-signal cross prints for the symbol you intend to trade. Risk framing once triggered: the strategy's position sizing risks 2% per position on a 38.2% retracement-based stop, with a hard stop at a 2% unrealized loss and a take-profit at a 4% gain, capped at 25% of the book per position. On the entry side, the nearest reference levels are about $175 and $177 for DLR and about $979 for EQIX. Reward-to-risk is defined mechanically here — roughly 2% risked against 4% targeted, about 2:1 on the fixed exits — before any signal-based exit. The completed backtest on this exact rule set covers the last 60 months: 6 trades, a 16.7% win rate, a 6.6% total return and a 10.6% maximum drawdown over that window. Note what that win rate means — most individual trades lose the fixed stop, and the return comes from the exits letting the occasional winner run. The 24-month window shows 3 trades and a 33% win rate, and the 12-month window shows no triggers at all; entries are genuinely sparse. That sparsity is a feature of the setup, not a reason to chase price before conditions confirm. Two additional checks before acting. First, parameter-sensitivity evaluation ran out of its time budget, so no robust alternative parameter setup was established — you are trading the published rules as-is. Second, the idea's bearish thesis (per its own argument, citing the NextDC revenue-delay selloff and rising debt-insurance costs for big tech) is directionally opposed to these long-entry rules; the rules fire long. If you believe the short thesis, treat a triggered entry as a signal that downside momentum conditions have aligned, and size accordingly — or stand aside.

The bear case has the macro tailwind — and the tape shows where a short on this pair pays

The thesis argues that AI data-center equities are running out of forgiveness, and the cited news supports that framing. Per Reuters, Australian operator NextDC shed 44% on a single revenue-timing delay (October 21, 2026); per Bloomberg, AI 'neocloud' names are stalling on IPOs and sinking (October 20, 2026); and Bloomberg also reports tech credit-insurance costs at fresh highs on an expected post-earnings borrowing wave (October 20, 2026). Two short candidates, DLR and EQIX, sit squarely in the crosshairs: both are capex-heavy REITs whose free cash flow is already negative. EQIX posted negative $292M of free cash flow in Q4 2025 — sitting at the 1.4th percentile among 833 IT-sector peers — and swung to negative $539M in Q1 2026. That is exactly the cash-quality profile the credit-cost argument targets. The company-level numbers do show deterioration where it matters for a credit-cost shock. DLR's Q4 2025 operating margin was just 6.8%, far below its historical norm of roughly 20-30% from the 2009-2016 era, leaving little cushion if borrowing costs rise on $116.5M of quarterly interest expense. (Its Q1 2026 operating margin recovered to 16.3%.) EQIX's interest costs come on top of free cash flow that has now been negative for four straight quarters through Q1 2026. On realized evidence: this backtested setup over the 60-month window produced a +6.60% total return on 6 trades, with a 10.61% worst-pair maximum drawdown over the 5-year evaluation. Over the trailing 24 months, the same setup generated +1.87% across 3 trades with a 5.61% maximum drawdown. The equity curve's strongest stretch — peaking near +10.9% in mid-2026 — coincides with the late-2025/2026 period in which these names repeatedly de-rated on earnings and guidance,…

Scores

• Conviction score breakdown: 43
• Thesis support: 55
• Trade readiness: 35
• Risk quality: 50
• Backtest evidence: 40
• Fundamentals trend: 35

Watch items

• DLR — ADX (14)
• DLR — MACD (12,26,9) signal cross
• EQIX — ADX (14)
• EQIX — MACD (12,26,9) signal cross
• EQIX — Insider net open-market activity
• DLR — Net margin
• EQIX — Dividend ex-date