How it works
There are two families. Maintenance covenants are tests the borrower must pass on a schedule, usually quarterly: net debt no more than some multiple of EBITDA, earnings at least some multiple of interest expense, a minimum cash balance. Incurrence covenants only bite when the borrower tries to do something — raise more debt, pay a dividend to its owners, sell a subsidiary. You can fail a maintenance covenant by sitting still and having a bad quarter. You can only fail an incurrence covenant by acting.
A worked example, with invented round numbers. A company borrows $500m against $100m of EBITDA, and the loan says net leverage must stay under 6x. That is a comfortable 5x on day one. Earnings drop to $70m and leverage is suddenly 7.1x. Nobody has missed a payment. The company is nonetheless in default, and the lenders can in theory demand the whole $500m back tomorrow.
They almost never do, because a lender who accelerates a loan the borrower cannot repay has just bought a company. What happens instead is a negotiation. The lenders grant a waiver or reset the test, and they charge for it: an amendment fee, a higher margin, mandatory amortisation so the loan starts paying down instead of sitting there until maturity, and a tighter covenant package for the next round. Sometimes the owner is allowed to "cure" the breach by injecting equity, which counts as EBITDA for testing purposes by pure legal fiction.
This produces no default, no missed payment, no ratings action, and no in any credit index. The risk gets repriced entirely inside the documents.
Cov-lite loans strip out the maintenance tests altogether, leaving only incurrence covenants. The borrower can bleed for years without tripping anything. Lenders find out something is wrong when the cash runs out, which is later and worse.
The other usage is real estate. A building let to a strong tenant on a long lease is not really a property investment, it is a bond issued by the tenant with concrete attached. So the lease's value is described as its covenant strength, and "the covenant" becomes shorthand for the tenant itself. When someone says a data centre has an investment-grade covenant, they mean a has promised to pay rent for fifteen years, and that promise is what the debt is secured on.
Why it matters to this crash
Both meanings are doing heavy lifting right now, in different corners of the same problem.
In , covenants are where losses go when nobody wants to admit there are losses. 's Sophos went looking for private credit backing for more than $2bn of loans, failed to find it, and went back to its existing leveraged-loan lenders offering a higher coupon, amortisation payments and a tighter covenant package. We covered this on 19 August and again the same day. The told lenders it would not inject fresh capital. So the lenders took payment in terms instead, which is lenders doing their job, and also the reason a deteriorating credit can be repriced twice without ever appearing as a default statistic. Set that against at the twenty largest listed reaching a median 2.8% of cost in Q2 2026, the worst since 2017, and the covenant channel is where the rest of the deterioration is hiding.
There is a second-order version. We noted on 18 August that Blackstone and Blue Owl vehicles sold high-grade bonds into a record August of investment-grade issuance. Those funds now have bondholders, and covenants, and leverage ratios calculated off the value of loans that nobody daily. A leverage test measured against a number the tested party estimates is a test with an interesting design flaw.
In , the tenant's covenant is the entire credit. Project Odyssey, the roughly $3.9bn deal backing a Georgia data centre leased by Microsoft, is expected to be rated investment grade while pricing at a low-to-mid 7% yield, where single-B junk trades (our note, 18 August). The agencies are rating Microsoft's covenant over a defined term. The market is pricing what the building is worth in year six. And at Meta and BlackRock's $14bn Sopaipilla campus, the venture holds up to $427m of property cover during construction and $450m once running, and is not insured against total loss (19 August). The covenant is carrying the debt; the concrete is barely insured.
What would make this dangerous
Watch how often refinancings close with amortisation and tighter covenants attached rather than cleanly. One is a negotiation. A run of them across a single sponsor's portfolio, as with Medallia in June 2026, Proofpoint in July and Sophos in August, is a repricing of an entire collateral pool that no default rate will show.
Watch whether sponsors keep declining to cure. Thoma Bravo's refusal to add equity to Sophos is the tell: an equity cure means the owner still believes; a refusal means the loss has been handed to creditors and the covenant is now a bargaining chip rather than a backstop.
Watch the private credit funds' own bond covenants. The pitch for the asset class is that it is unlevered and unrunnable. Public bonds bring leverage tests, and those tests are measured against valuations the manager supplies, so the point at which one of them gets close is not visible to anybody outside the firm.
And on the lease side, the single observable that matters is whether a data-centre lease abates rent on casualty. If Meta keeps paying on a destroyed building, bondholders are fine. If not, the credit disappears with the concrete. That clause is not public, and we have not seen it.