Frequently asked questions
Twenty questions covering conventions, readings, data sources, and usage limits. Every answer stands on its own first sentence; all numbers are sourced from on-site pages, with the API as the single source of truth.
What is NCSX-IS?
An issuance-spread benchmark for AI-infrastructure (neocloud) corporate bonds, published by SpreadBench: per-bond spread = (coupon − same-maturity Treasury on the bond's own pricing date) × 100, face-value weighted. Current level 432.3bp (as of 2026-11-04, methodology v0.2.1-real). Source →
At what spread is AI-infrastructure debt pricing?
Face-value weighted 432.3bp. By tier: senior secured 391.2bp / unsecured 540.1bp — the unsecured tier is a single-issuer snapshot (all 4 bonds are CoreWeave, $7.75B); cite it as "CoreWeave unsecured 540.1bp". Source →
How many bonds does the index cover?
13 public fixed-rate USD notes from 6 issuers, $28.06B face value outstanding — the full population of the sector's public fixed-rate USD notes, not a sample. Source →
Which issuers are in it?
CoreWeave, Galaxy Digital, Cipher Mining, Core Scientific, Applied Digital, and TeraWulf. Each has a per-issuer spread page with per-bond links to the source SEC filings. Source →
How does it compare with US HY OAS?
On the same observation date US HY OAS was 332.0bp, so NCSX-IS sits 100.3bp above it; BB 199.0 / B 351.0 / CCC 1297.0bp. Different conventions (issuance-time vs secondary-implied) — magnitude reference only, not a relative-value input. Source →
What is CoreWeave's bond spread?
Its 4 unsecured notes have a face-value weighted issuance spread of 540.1bp ($7.75B, single-issuer snapshot); individual bonds priced at 506.4–580.8bp. Per N-PORT marks, CoreWeave's notes now sit +231.4bp above issuance (face-value weighted). Source →
How wide is the secured-unsecured gap?
A structural premium of 148.9bp (391.2 vs 540.1). But the unsecured tier contains CoreWeave only — cite it with that qualifier, or readers will mistake it for a diversified tier. Source →
Why are there three readings?
They answer three different questions: 432.3bp — what these bonds priced at (issuance); 361.3bp — the pure-rate view (curve-adjusted; the −71.0bp delta is entirely Treasury curve shift, zero credit information); 616.4bp — what fund NAVs imply (N-PORT marks on 7 bonds). Conflating them is the most common citation error. Source →
Is there a daily series or history?
No — and none is simulated. The convention is issuance-time spread, event-driven: the index re-versions only on inclusion, maturity, or erratum. Intraday/daily per-bond prices require FINRA TRACE licensing, which this site does not hold. Source →
Where does the data come from?
Primary sources only: SEC filings (8-K / 424B / 10-Q for coupons, sizes, and security terms) plus the US Treasury par yield curve — every row verifiable; the only intermediate step is a fixed linear interpolation. Source →
Can I recompute it myself?
Yes. Replication package v1.1 (DOI 10.5281/zenodo.24142915, CC BY 4.0) ships per-bond data, code, and a data dictionary; recompute.py asserts the index level and all three readings. Source →
When does AI-infrastructure debt mature?
$17.86B in 2031 (63.6%, 8 bonds); another $8.95B in 2030 (31.9%, 4 bonds), of which $6.95B is packed into the 17 days from 2030-11-15 to 12-01 — a maturity wall built in a one-year window. Source →
What do data centers pay for power?
DCPI reads 64.0 $/MWh (pre-registered 2026-11-11: unweighted median of all full-level verifiable tariff sheets at 80% load factor; unchanged after a sixth sheet was added on 2026-11-12). Of the 11-row tariff ledger, 5 rows carry no verifiable price — the terms layer is fully public, the price layer almost never is; that gap is why the index exists. Source →
Is there a GPU residual-value reading?
Not yet — and it is not forced. GRVI currently publishes a listing-price median panel (eBay convention); an exhaustive SEC full-text search (823 hits since 2019, each classified) found zero buyer-side A100 per-card prices, so the denominator remains unanchored and the reading stays null. Source →
Is this investment advice or a credit rating?
Neither. The site aggregates public data and publishes recomputable factual calculations; nothing here is a credit rating, investment advice, procurement advice, or a power-price forecast. The red lines are stated in the methodology. Source →
Why is Hut 8 excluded?
Its two notes ($7.5B combined) are amortizing SPV project-finance debt (post-completion semi-annual amortization + DSCR + single-tenant lease), structurally incomparable with the notes in the index. Mechanically including them yields 370.4bp — disclosed as a counterfactual, never included. Source →
How do I use the API? Is it free?
Free, read-only, no registration. Master entry: api/v1/real/index.json; OpenAPI 3.1 spec at openapi.json; rate limits and the endpoint list are on the API page. Every payload carries data_status so machines can tell citable from retired data. Source →
How do AI agents connect?
Three paths: retrieval via llms.txt; tool calls via the MCP server (npm i -g @ncsx/mcp, streamable-http at /mcp, 8 read-only tools); common guessed paths (e.g. /api/measures) now 301-redirect to the real endpoints. Source →
Is there a working paper?
Yes: "Measuring an Asset Class That Has No Benchmark" (English abstract, replication code and data dictionary included) — see the paper page; academic citations should use version DOI 10.5281/zenodo.24142915. Source →
How should I cite it?
Recommended form: "NCSX-IS (SpreadBench) issuance spread of 432.3bp as of 2026-11-04 (methodology v0.2.1-real)". 16 citable facts — each with a one-click citation line and its calculation — are on the citable-facts page. Source →