Five-year credit default swap (CDS)
What it is. A credit default swap is insurance against a company failing to pay its debts. The buyer pays a yearly premium; if the company defaults, goes bankrupt or misses a payment, the seller pays out the loss. The five-year contract is the most traded, so it is the one quoted in the news and the one the Watch tracks.
The premium is quoted in basis points a year on the amount insured. Meta’s five-year CDS at 124bp means protecting $10m of Meta debt for five years costs $124,000 a year.
What a move means. A rising CDS means protection costs more: the market thinks default is more likely, or more people want to hedge their exposure, or both. A rough way to read the level is that the yearly premium divided by the share of the debt you’d expect to lose in a default (often taken as 60%) gives a market-implied yearly default probability. 124bp ÷ 0.6 is about 2% a year. That figure overstates the real-world risk, because the premium also pays the seller for bearing uncertainty, but it moves in the right direction and it is comparable across companies.
As a rough guide to levels:
| Five-year CDS | What the market is saying |
|---|---|
| Below 100bp | Solid investment-grade borrower |
| 100–300bp | Weaker investment grade, or the stronger end of high yield |
| 300–1,000bp | High yield: real risk of default over the period |
| Above 1,000bp | Distressed: lenders are pricing a meaningful chance of losses |
Two cautions. First, single-company CDS markets can be thin, so a few large trades can move the quote further than the underlying risk has moved. Second, rising CDS can reflect hedging rather than fear: banks that have lent to a data-centre project may buy protection on the company behind it, which pushes the price up without anyone expecting a default. The Watch treats a CDS move as stronger evidence when bond prices for the same company are moving the same way.
Where the Watch uses it. The AI-specific credit pillar tracks five-year CDS for Oracle, Meta, CoreWeave and SoftBank. A hyperscaler’s CDS rising while the broad investment-grade spread stays flat is discrimination: lenders are pricing AI risk specifically. That is stage 2 evidence on the transmission chain, not stage 4.
Basis point (bp)
What it is. One hundredth of a percentage point. A spread moving from 3.31% to 3.38% has moved 7 basis points (7bp).
Why it’s used. Credit moves are small in percentage terms and large in money terms. Saying “7bp” avoids the ambiguity of “up 0.07%” versus “up 0.07 percentage points”.
Credit spread (option-adjusted spread, OAS)
What it is. A company borrows from investors by issuing a bond: it sells a promise to repay on a fixed date, and pays interest until then. The yield is the yearly return the buyer earns on the price they actually paid. Those interest payments stay fixed, so if the bond’s price falls, the yield rises. The same payments are a larger return on a cheaper purchase.
The credit spread is that yield minus the yield on a US government bond of the same length. The government is treated as the safest borrower, so the difference is the premium the company pays for the risk that it might not repay. A mortgage works the same way in principle. A homeowner’s rate sits above the rate the government pays to borrow, and the gap reflects the risk that a household is a less safe borrower than the government.
This is a different price from a credit default swap. A CDS is insurance on the debt: the premium someone pays to be protected if the company defaults. The spread is the lending itself, the extra interest on the bond. The two usually move together, because both respond to the same worry, but one is the cost of the loan and the other is the cost of insuring it.
Option-adjusted means one adjustment. Some bonds let the company repay early, and that right changes the yield. The option-adjusted spread takes it out, so a bond with the right can be compared with one without it.
The company pays the premium when it issues the bond. Investors will only lend if the yield is high enough above the government’s. After that, the bond is bought and sold. If holders can only sell it for less than they paid, the yield rises and the spread widens, even though the company has not borrowed anything more. The next bond it issues has to be offered at roughly the spread the market is already showing, or it will not sell.
What a move means. A wider spread means lenders are charging companies more to borrow. When the spread on an AI borrower widens and the broad index doesn’t, lenders are being selective. When the broad index widens too, and stays wider, stress is spreading.
Where the Watch uses it. The ICE BofA US corporate indices, published daily on FRED, are the backbone of the broad-credit pillar and set the stage 4 tripwires: high yield above 450bp, or investment grade above 130bp, for five consecutive trading days. 450bp means those bonds yield 4.50 percentage points more than US government debt of the same length.
Investment grade and high yield
What they are. Ratings agencies grade borrowers. BBB− (or Moody’s Baa3) and above is investment grade: large, established companies that are expected to repay. Below that is high yield, sometimes called junk: smaller, more indebted or less proven borrowers who pay more to borrow.
Why the split matters. Many pension funds and insurers can only hold investment-grade debt. A company downgraded below BBB− can face forced selling of its bonds, which is why a move across that line is treated as an event rather than a gradual change.
CCC and lower
What it is. The lowest rating tier short of default. These borrowers are the first to lose access to credit when lenders turn cautious.
Where the Watch uses it. The CCC spread is reported as an early warning on the broad-credit pillar. It doesn’t set the pillar’s status by itself, because it is volatile and dominated by a few sectors, but it often moves first.
US 10-year Treasury yield
What it is. The interest rate the US government pays to borrow for ten years. It is the base rate that most long-term borrowing, including data-centre financing, is priced from.
What a move means. A higher 10-year yield raises the cost of every long-lived investment and lowers the present value of profits expected far in the future. AI infrastructure is both: expensive now, paid back over many years. That is why rates sit on the scorecard even though they are not an AI measure.
Capex (capital expenditure)
What it is. Money spent on long-lived physical assets: data centres, chips, power and cooling. For the hyperscalers it is now the largest line in the AI story.
What a move means. Rising capex guidance means the build-out continues. A cut can mean less demand, or it can mean supply problems such as late chip deliveries. The explanation matters as much as the number, which is why the Watch reports both.
Free cash flow
What it is. Cash from operations minus capex: what a company has left after paying for its investment. Negative free cash flow means the investment is being funded from savings or borrowing rather than from the business.
What a move means. One negative quarter during a heavy build is common. Consecutive negative quarters at a hyperscaler mean the AI build-out has outrun the cash the business generates, and the gap has to come from somewhere. That is the central question on the AI-economics axis.
Hyperscaler
What it is. The largest cloud operators, which build and run data centres at global scale: Amazon, Microsoft, Alphabet and Meta, with Oracle increasingly counted among them. Their capex guidance is the clearest single signal of how much AI infrastructure is being built.
Neocloud
What it is. A newer cloud provider built specifically to rent out AI computing power, such as CoreWeave, Nebius or IREN. Neoclouds typically borrow heavily to buy chips, with the loans secured on the chips and on contracts with a small number of large customers.
Why it matters. Neoclouds sit where AI demand meets debt. If customer contracts weaken or chip values fall faster than expected, their lenders are among the first to feel it, which is why CoreWeave’s CDS and bond prices are tracked closely.
Prime brokerage and deleveraging
What it is. Prime brokers are the banks that lend to hedge funds and hold their positions. Their data on how much funds are borrowing (gross leverage) and how they are performing is the best public view of whether funds are being forced to sell.
What a move means. Deleveraging is funds cutting borrowing by selling positions. When it is synchronised across many funds, or driven by margin calls (lenders demanding more collateral), selling begets more selling regardless of what anyone thinks the assets are worth. Documented forced deleveraging is the entry evidence for stage 3.
Private credit and BDCs
What it is. Lending by funds rather than banks, mostly to mid-sized companies. Business development companies (BDCs) are a common vehicle: some listed on the stock market, some sold privately to wealthy investors with limits on how much can be withdrawn each quarter.
What a move means. Private loans don’t trade, so losses show up slowly: in falling valuations of loans against what was paid for them, in listed BDCs trading below the value of their assets, and in investors asking for more money back than the fund allows. The Watch tracks all three on the private-credit pillar.
Special-purpose vehicles and sale-and-leaseback
What it is. Ways of financing data centres without the debt sitting on the main company’s balance sheet. A separate company (the vehicle) borrows to build or buy the asset and leases it back to the operator.
Why it matters. These structures are legitimate and common, but they make it harder to see how much the AI build-out is being funded with debt. When the Watch reports financing flow, it counts these deals alongside ordinary bond issuance.