[CRFC] Origination Parameters for New sUSDai Financings

Summary

This proposal publishes the lending standards by which new GPU financings can be funded through sUSDai. It sets the origination criteria, approved collateral, maximum sUSDai utilization and the Compute IQ underwriting framework that every new financing must meet. Terms for selling facilities through GLX will be set in a separate governance proposal.

Motivation

Formalizing programmatic parameters for sUSDai curation. Publishing them lets CHIP holders, originators and integrations see the standards every new financing is held to, and means any future change goes through governance.

Specification

Origination criteria

Parameter Value
Maximum blended portfolio Loan To Value (LTV) 80%
Minimum quarterly portfolio Debt Service Coverage Ratio (DSCR) 1.15x
Residual Value Warranty Required for any non-investment-grade offtaker or on-demand rental
Minimum Compute IQ score needed for a GPU financing to be eligible for sUSDai funding 60 out of 100
Maximum sUSDai utilization to deploy new financings 90%

Approved collateral

Type Approved
NVIDIA enterprise-grade GPUs B200, B300, GB200, GB300, VR200
AMD enterprise-grade GPUs MI455X, MI430X
Non-core inference chips (e.g. Cerebras, Etched) Only alongside contracted offtake, and subject to passing the Compute IQ score

Existing facilities keep their contractual terms.

Underwriting (Compute IQ)

Compute IQ is the underwriting system used for every sUSDai financing. It scores each deal on a 100-point scale across three pillars, and the score sets facility size, pricing tier and structure. Deals scoring below 60 are not financeable.

Pillar Key metrics
Operator Track record; operator experience; GPU deployments to date; financial strength; cash runway; EBITDA margin; net debt / EBITDA; equity backing
Customer Cash runway; absolute EBITDA; funded liquidity; net debt / EBITDA; equity backing quality; contract coverage from revenue; operating history
Contract Contract tenor vs debt term; contract coverage; termination protection; payment unconditionality (take-or-pay); termination-for-cause standard; termination for convenience and make-whole; SLA burden; service-credit exposure; performance scope; commitment vs achievability

Governance sets the three pillars and the minimum score. The underwriting team may adjust individual metrics and point allocations within each pillar as market conditions change.

Voting

For: These parameters apply to every new financing funded through sUSDai. Any change requires a new governance proposal.

Against: The parameters are not adopted. Revised parameters can be proposed in a new CRFC.

References

Licence

CC0

1 Like

Whilst these parameters seem modest & safe, there is no external point of reference for “industry standards” and thus the selection for these parameters in the proposal. Even the reliance ‘Compute IQ score’ is something that simply reads as “trust me” more so than a manner to justify the parameters selected.

Scalability (and recycling rate & profit margins derived from sale of loans via GLX) per $100M in sUSDAI is not shared and is important information.

I understand that this is a 0-1 implementation of credit funding mechanisms for compute however, some form of “proof” of these parameters working either via disclosure of existing loans’ parameters or a comparison with industry standards is needed - along with projections RE scaling, recycling + margins, default scenario risks.

Regarding industry standards, you’re right that the industry itself is still quite young w/o a broadly accepted standard (which is in many ways why the opportunity exists in the first place), but a quick google search will tell you that these proposed parameters fall within the range of normal for the industry - 60-80% LTVs, 1.15-1.5x DSCRs, 3-5 year term, low to mid teens cost of capital for non-investment grade offtakers.

ComputeIQ is USD.AI’s proprietary underwriting scorecard, refined based on the ~$25BN in loans reviewed to-date. The key metrics being scored are all listed in the governance post (without the specific scores for each defined to maintain some edge over other lenders looking at the space). Essentially, ComputeIQ is stacking up a number of qualitative and quantitative metrics to derive an overall underwriting score that can both binarily screen yes/no decisions and then also capture some nuance across different deals (ie, while a 61 and a 85 are both financable, an 85 should be prioritized as a better credit, either by getting first look at available capital for lending, or with more borrower friendly overall terms).

You can see existing loan parameters on app.usd.ai along with the audit/attestation from TNF here: Proof of Loans | USD.AI. To-date, there have been no defaults across the loan book. As the loan book continues to expand across borrowers and time, there will be more data to backtest against.

GLX loan sale concept will be part of a separate governance post.

This discourse has moved to Snapshot voting and is closed. To review the proposal and/or vote, please go to Snapshot.