How do power delays affect data-center debt repayment, and under what conditions could project-level losses become a broader financial problem?

Advocate A’s strongest point is correct: enforceable payment obligations and adequate reserves can protect lenders during delays. Its weakest point is treating project liquidity as analytically prior to interconnected exposure. A guarantee may transfer rather than eliminate risk; whether protection survives depends on its activation conditions and the guarantor’s other obligations. A acknowledges this, but the contractual evidence makes it central rather than secondary.

First, TeraWulf’s May 8, 2026 filing provides a concrete counterexample to treating technology-company backing as construction-period protection. Its Google backstop becomes effective only when the corresponding Fluidstack lease commences; none had commenced as of March 31, 2026. Before effectiveness, specified arrangements pledge warrants over TeraWulf shares—not unconditional Google cash payments. My inference is that this collateral depends on the developer’s equity value precisely when unsuccessful completion could weaken that value. This is a contractual comparator, not evidence of an actual power delay. (investors.terawulf.com)

Second, A’s Applied Digital example quantifies the liquidity burden without establishing its containment. The disclosed $2.35 billion principal at 9.25% implies approximately $217.4 million in annual coupon payments before amortization. A hypothetical twelve-month postponement therefore requires financing roughly that amount of additional pre-revenue interest, excluding other carrying costs and assuming unchanged principal. This is a sensitivity calculation, not a forecast or an estimate of reserve exhaustion. The filing also explicitly limits parent support to the completion guarantee rather than general credit support. An interest reserve—cash earmarked for interest—buys time; it does not establish sufficient resources for an indefinite delay. (ir.applieddigital.com)

Third, there is evidence for investigating transmission beyond individual projects. The Financial Stability Board’s May 2026 report identifies deepening connections among private-credit funds, banks and insurers, alongside concentration and monitoring gaps; it does not establish their specific data-center holdings. S&P’s equipment-securitization analysis separately identifies construction and installation delays as risks to lease commencement. Together, these support tracing ultimate exposures across financing vehicles rather than assuming different lenders mean independent risks. (fsb.org)

The strongest evidence against an imminent systemic interpretation must also be retained: the IMF’s April 2026 assessment describes hyperscaler balance sheets as strong and financial-stability risks as contained, while identifying longer-run financing and obsolescence vulnerabilities. My position is conditional contagion, not demonstrated systemic insolvency. (imf.org)

My proposed investigation would match original and revised grid-connection dates to spending, revenue commencement, reserves, cancellation clauses and maturities. Stress tests should combine six-, twelve- and twenty-four-month delays with explicitly hypothetical capacity-price and equipment-value declines. Apply price reductions only where repricing, cancellation or replacement leasing permits them, and equipment haircuts only to relevant collateral. Then consolidate borrowers and guarantors across banks, funds, insurers and securitizations. Broader stress becomes plausible if simultaneous funding gaps exceed independent sponsor resources and creditor loss-absorbing capacity, inducing asset sales or reduced lending elsewhere; the available evidence does not yet measure those thresholds.

A’s specific weakness is its proposed sequencing: treat delays primarily as project-finance problems until aggregated shortfalls demonstrate transmission. Yet A concedes that pre-commencement guarantees may supply no cash, sponsor support depends on ability to pay, and guarantor resources must be counted only once. Those concessions imply that project protection cannot be assessed independently of shared counterparties. My position is conditional financial transmission, not demonstrated systemic insolvency; requiring evidence of the latter does not rebut the former.

A’s escrow example is genuine protection but addresses a narrower risk. Applied Digital’s March 2026 financing allowed proceeds to leave escrow upon execution of an electric-service agreement, with a June 30, 2026 mandatory-redemption deadline if the condition remained unmet. Execution—not actual energization—was the disclosed release condition. My inference is that this safeguard reduces deployment risk before agreement signing but does not establish protection against subsequent connection delays. I am describing the historical terms, not asserting whether redemption occurred. (ir.applieddigital.com)

Additional contractual evidence strengthens the distinction between headline support and usable liquidity. TeraWulf’s May 7, 2026 revolving facility had $250 million of commitments, but its agreement capped pre-commencement availability at $75 million and required at least $100 million of defined quarterly liquidity before first commencement. Thus a stress test cannot treat the full commitment as construction-period cash or assume distress begins only when cash reaches zero; covenant headroom matters. These are contractual constraints, not evidence that TeraWulf breached them. (investors.terawulf.com)

Moreover, TeraWulf’s Google backstop, once effective, permits Google following specified tenant defaults to pay a termination fee instead of assuming the lease and paying outstanding rent. Therefore even post-commencement backing should be modeled against the termination payment and debt outstanding, rather than automatically equated with all remaining contracted rent. This does not establish inadequate recovery; it identifies the missing comparison. (investors.terawulf.com)

A correctly emphasizes the IMF’s April 2026 finding that strong hyperscaler balance sheets kept financial-stability risks contained. But that same assessment warns that equipment obsolescence could increase future financing requirements. The FSB separately identifies bank–private-credit–insurer interconnections, concentration and monitoring gaps. Neither document measures overlapping data-center losses, and I do not claim otherwise. (imf.org)

My analytical recommendation is to test project cash flows and creditor networks together: impose hypothetical 6-, 12- and 24-month energization delays, contractually permissible capacity-price declines and collateral-specific depreciation; then identify covenant breaches, reserve exhaustion and refinancing gaps. Consolidate obligations across banks, private-credit funds, insurers and securitized vehicles, counting shared guarantees once. The conditional transmission mechanism is simultaneous funding demands exceeding independent support and creditor liquidity, prompting sales or reduced lending. A’s rebuttal establishes that protections exist; it does not establish that their timing, size and independence match that combined shock.

A’s strongest position is that delayed electricity need not interrupt debt repayment when reserves and enforceable, collectible funding obligations bridge the delay; carrying costs are not automatically creditor losses. Its provisional containment diagnosis also has empirical support: the IMF’s April 2026 assessment found hyperscaler balance sheets strong and associated financial-stability risks contained. (imf.org)

A’s acceptance of simultaneous project and network testing substantially narrows our disagreement. My position remains stronger as an investigative framework—not as a claim of an existing systemic crisis—because contractual protection must be evaluated against simultaneous demands on its providers, rather than treated as evidence of containment before those demands are measured.

TeraWulf supplies a specific contractual example. Its May 8, 2026 filing states that Google’s backstop activates at the corresponding Fluidstack lease commencement and, after specified tenant defaults, permits a termination payment instead of lease assumption. My inference is that neither the backstop’s headline existence nor total contracted rent establishes cash available during construction or eventual debt recovery. This is not evidence of an observed power-delay default. (investors.terawulf.com)

The same filing discloses $250 million of revolving commitments but a $75 million pre-commencement availability cap and a $100 million quarterly minimum-liquidity covenant before first commencement. Those terms require testing drawable funds and covenant headroom, not simply subtracting expenses from headline commitments. (investors.terawulf.com)

The transmission concern is specific, although unquantified. The FSB’s May 6, 2026 report identifies electricity shortages as a possible trigger for data-center delays, cancellations and credit losses, and overcapacity as another source of weaker returns. Separately, it documents banks lending to companies also financed by private-credit funds, alongside insurer investment and securitization connections. These findings justify investigating overlapping exposures; they do not establish their data-center-specific magnitude. (fsb.org)

Importantly, the FSB also reports relatively small measured direct bank exposures to private-credit funds and recognizes that closed-end structures mitigate liquidity mismatches. My conditional mechanism therefore requires concentrated shortfalls and insufficient liquidity or capital—not merely interconnectedness. (fsb.org)

My analytical recommendation is a monthly, consolidated stress test: hypothetical 6-, 12- and 24-month energization delays, lower capacity prices only where contracts allow repricing or replacement leasing, and equipment depreciation only against relevant collateral. Count shared guarantor resources once and test refinancing without assuming new financing remains available.

I would accept A’s containment diagnosis if an independently validated supervisory dataset covering a representative majority of construction-stage debt linked utility-confirmed schedules, spending, revenue dates, contracts, reserves and maturities to ultimate creditors and guarantors, and demonstrated resilience under those joint shocks. Specifically, collectible support must cover funding gaps, residual losses must remain within creditor buffers, and observed delayed-project outcomes must show no material forced sales or contraction in unrelated lending attributable to those exposures. Isolated successful projects or strong aggregate technology-company balance sheets would not establish that result.