Short NQ, NKY and SMH; Long TLT and Gold: The Neocloud Capex Trap

Coverage: the capex bill overwhelms earnings

Four neoclouds plan $218.1bn of capex in 2027–28. Annual spending rises from $76bn in 2026 to $104.3bn in 2027—up 37%—and $113.8bn in 2028. Operating earnings do not fund that trajectory. (Bloomberg consensus/BQL, 14 October 2026.)

CoreWeave’s 2027 EBITDA/interest is 3.15×, but GAAP EBIT/interest is 0.77×. After the full capex bill, coverage falls to −5.96×. Applied Digital is weaker: 1.70×, 0.64× and −12.61×, respectively. EBITDA coverage conceals the scale of the investment funding requirement.

An earnings-only capex budget that leaves 1× interest cover is EBITDA minus interest. Across all four companies, that supports at most $66.8bn of the $218.1bn programme. $151.3bn—69% of spending—needs cash beyond that earnings benchmark.

The financing wall peaks in 2027

Forecast negative FCF totals $135.2bn in 2027–28. Add CoreWeave’s disclosed principal repayments of $6.2bn in 2027 and $4.4bn in 2028, and known cash requirements reach $145.8bn: $82.6bn in 2027 and $63.2bn in 2028. That is before principal on other companies’ debt or future borrowing.

A quarter of the funding failing to arrive leaves a $36.5bn shortfall. Management must replace that capital, sell assets or reduce the buildout. New borrowing also adds interest and amortization to subsequent funding requirements.

The later repayment wall is already visible: Nebius has $5.6bn of convertible redemptions in 2030; Applied Digital has $4.3bn of scheduled principal in FY31. Financing the next expansion adds to the obligations beyond these disclosed schedules.

Download the coverage and financing model. Sources: CoreWeave, Nebius, IREN and Applied Digital.

Why capex gets cut

Global AI-related debt issuance dropped from $113bn in June to $23bn in September—down 80%. Earlier fundraising explains part of the pause. The next tranche must still arrive before construction payments fall due. (FT/Morgan Stanley.)

Customer and project funding protect specific deployments. Nebius’s 50–60% prepayment coverage applies to associated Q2 deals. IREN’s separate FY27 $25–30bn capex plan faces a $3–8bn remaining capital requirement even after its $14bn funding package and another $8bn of targeted financing. (Nebius; Bloomberg/Todaro.)

Lenders’ capital requirements determine the construction budget. Applied Digital models 80% loan-to-cost for investment-grade tenants and 70% for its CoreWeave campus. On $10bn of project cost, the required other capital rises from $2bn to $3bn—a 50% increase. (Applied Digital.)

Our bear case is a 20–25% capex undershoot: $44–55bn of spending removed or deferred in 2027–28. Unfunded expansion phases slip first. Supplier orders and earnings follow, unless other buyers absorb the displaced demand.

CoreWeave’s capex trap

Cutting CoreWeave’s buildout weakens the revenue growth needed to support its debt. Sustaining it requires continuous capital markets access. In 2027, $16.7bn of EBITDA minus $5.3bn interest and $6.2bn scheduled principal leaves only $5.2bn against $48.3bn capex.

At our 1,050bp CDS stress reference, refinancing dominates the equity case. The escape routes are:

1. Massive equity recapitalisation: replace expensive debt and fund construction, with heavy dilution.
2. A hyperscaler acquisition or asset sale: Microsoft or Google buys capacity and contracts; creditor claims are paid before equity receives value.
3. NVIDIA or customer support: more capacity purchases, financing or a new asset transaction. NVIDIA’s current $6.3bn agreement purchases unused compute capacity; a GPU-fleet acquisition requires a separate deal.
4. Credit recovery: realised demand and cash generation compress spreads and support the 2029 earnings inflection in our recovery case.

We put the odds of a CoreWeave equity wipeout at 50%. A rescue can preserve the GPU fleet and contracts while cancelling the existing shares. (NVIDIA agreement.)

The trades

Short NQ, NKY and the SMH complex. Semiconductors take the direct hit from delayed GPU, memory and infrastructure orders. NQ captures broader AI earnings repricing; NKY extends the trade to Japanese chip-production and testing suppliers.

Long TLT and gold. Peak capex means less new corporate issuance and less competition for duration capital. We expect a weaker investment impulse to ease yield pressure, giving healthier corporate bonds breathing room and supporting Treasuries. Lower real yields and defensive demand support gold. (FOMC minutes; World Gold Council.)

TLT’s 14.7-year duration gives roughly 7.4% price sensitivity to a 50bp yield decline, before carry and convexity. The basket expresses the spending reset across AI equities, duration and gold. (iShares, 13 October.)