Goldman's bet on
the software short
Goldman Sachs is building total return swaps on AI-exposed corporate loans. The product isn't ready. The demand is. Here's what the structure does, what it doesn't do, and what happens when it goes live at scale.
June 2026 · Finalized June 23, 2026
Icarus Asia Research is editorially independent. No compensation was received from Goldman Sachs, JPMorgan, or any other institution referenced in this report.
Executive Summary
Goldman Sachs has been pitching hedge funds on total return swaps (TRS) referencing baskets of corporate and leveraged loans, with particular focus on enterprise software borrowers that Goldman's own research identifies as vulnerable to AI disruption. As of late March 2026, no trades have been executed. Goldman told clients the tool "isn't ready yet."
The product is a TRS, not a credit default swap. That distinction matters. A CDS pays out on defined credit events — bankruptcy, restructuring, failure to pay. A TRS transfers the full economic performance of the reference portfolio, including spread widening and mark-to-market moves that occur well before any formal default. Hedge funds going short via TRS don't need a default to get paid. They need prices to fall.
The market context makes the demand legible. Private credit has lent over $500 billion to SaaS and enterprise software firms (BIS Quarterly Review, March 2026). Those borrowers built their valuations on subscription stickiness and pricing power that agentic AI is now testing. JPMorgan Asset Management has flagged a "structural reassessment of software credit risk." Goldman's own Cracks in Private Credit report questions whether leverage multiples in software LBOs can survive the disruption cycle. The TRS is the instrument that closes the gap between that thesis and the ability to trade it.
Both Goldman and JPMorgan are exploring these structures, per Bloomberg's March 31, 2026 reporting. No confirmed trade volumes or counterparty details appear in any public record. The systemic risk implications, while real, remain speculative given the product has not yet launched as of this report's finalization date.
Goldman is not selling insurance against AI loan defaults. It is building a tool that transfers the total return of AI-exposed loan portfolios — spread risk, price risk, and credit risk combined. Press characterizations of this as "default insurance" misread the instrument and understate the exposure Goldman warehouses on the other side.
Section 1
The market Goldman is trying to short
Private credit's bet on software is large and concentrated. BIS data shows outstanding direct-lending loans to SaaS firms grew from roughly $8 billion in 2015 to more than $500 billion by end-2025. Software now accounts for approximately 19% of total direct-lending loans. That concentration built up during a decade of low rates, when the predictability of subscription revenue made software companies look like investment-grade borrowers in leveraged-loan clothing.
AI is testing that thesis from two directions at once. On the demand side, enterprise software companies face customers who can now build point solutions with agentic tools rather than buying off the shelf. On the supply side, the same AI infrastructure build-out is generating its own debt: S&P Global Market Intelligence estimates that lenders committed $121.9 billion in credit for data-center properties in 2025 alone. BIS Bulletin 120 puts private credit lending to AI-related sectors above $200 billion, with projections of $300–600 billion by 2030.
SaaS Exposure
Editor's note — The $8B (2015) and $500B+ (2025) figures are primary-sourced from the BIS Quarterly Review, March 2026 (r_qt2603v). All intermediate data points are Icarus Asia estimates interpolated for visual continuity and should not be treated as primary data. The ~62x growth multiple is an Icarus Asia calculation from the two BIS endpoints. Sources: BIS Quarterly Review Mar 2026; Icarus Asia estimates.
Two credit clusters, two different risk profiles
The AI disruption cycle is creating two distinct credit risk groups, not one. They differ in direction, timing, and instrument.
| Dimension | Cluster A — Legacy Software | Cluster B — AI Infrastructure |
|---|---|---|
| Core risk | AI erodes revenue, pricing power, and renewal rates of existing SaaS borrowers | Overcapacity, compute-cost deflation, or macro shock hits data-center and GPU-provider debt |
| Direction of trade | Short — TRS, equity short, thin CDS | Long (currently) with latent short risk at scale |
| Estimated credit exposure | $500B+ in SaaS direct-lending loans (BIS, Mar 2026) | $200B+ private credit AI loans; $122B data-center credit committed in 2025 (BIS Bul. 120; S&P Global MI) |
| Goldman TRS focus | Yes — enterprise software and AI-vulnerable corporates named as reference targets | Not yet confirmed; logical extension if infrastructure debt matures and stresses |
| Liquidity of underlying | Low — private direct-lending loans; sparse secondary market | Mixed — data-center bonds are public; GPU/AI-infra loans are often private |
| Key uncertainty | Speed of AI adoption vs. borrower refinancing runway | Durability of hyperscaler capex; technology obsolescence timing |
Goldman's TRS, as described in current public reporting, targets Cluster A. But as AI-infrastructure debt scales toward the $300–600 billion range projected by BIS, Cluster B becomes a candidate for the same product architecture.
Goldman Sachs research published in May 2026 confirmed the broader investment community has accelerated this rotation: hedge funds and mutual funds are "doubling down on AI," moving capital toward semiconductor stocks and away from software (CNBC, May 25, 2026, citing Goldman prime brokerage data). The FSB's May 2026 report on private credit vulnerabilities (P060526) explicitly identified loan-linked derivatives as a transmission channel warranting macroprudential attention. As of June 23, 2026 — this report's finalization date — no public evidence has emerged that Goldman's TRS product has launched or that any trades have been executed.
Section 2
How the TRS works, and how it differs from a CDS
A total return swap is a contract where one party pays another the full profit and loss of an asset — including income and price moves — in exchange for a fixed or floating rate. If the asset falls in value, the party that "sold" the total return pays the shortfall. If it rises, the other party pays.
Goldman's product lets hedge funds take the short side of that contract on a basket of software company loans. When loan prices drop, the hedge fund gets paid. Goldman, on the other side, absorbs the loss — and must hedge it somehow. That "somehow" is the systemic risk question.
A credit default swap (CDS) is simpler: you pay a regular premium; if the borrower formally defaults, you receive a lump-sum payout. The TRS is more complex — and pays out on deteriorating prices even without a formal default event.
The Goldman product is a total return swap on a basket of corporate and leveraged loans. Goldman acts as the total return payer, transferring the full economic performance of the reference loan portfolio to the receiver. That includes coupon income, fees, and all mark-to-market moves. The hedge fund pays a funding leg (typically SOFR plus a spread) and receives or pays the net of the portfolio's total return. When prices fall and spreads widen, the short-side hedge fund collects. When prices rise, it pays the difference.
A hedge fund that wants to be short structures the trade so it profits when total return is negative — collecting on spread widening, price deterioration, and defaults. The TRS does not require a formal credit event to trigger payment. That is the core distinction from a CDS, and it matters for how Goldman's risk on the long side should be understood.
| Feature | Total Return Swap (TRS) | Credit Default Swap (CDS) |
|---|---|---|
| Risk transferred | Full economic performance: coupon income, price changes, credit events, mark-to-market moves | Default and recovery risk only; no payment for spread widening absent a credit event |
| Payout trigger | Ongoing total return of reference portfolio on each reset date (path-dependent) | Defined credit event only: bankruptcy, restructuring, or failure to pay |
| Payout amount | Net of total return leg vs. funding leg — accumulates over time | Par minus recovery (single-name) or auction-based settlement |
| Basis risk | High — depends on accuracy and timeliness of loan pricing | Lower — binary credit event, though CDS-bond basis exists |
| Embedded leverage | High — large notional exposure relative to posted margin | Moderate — limited to premium payments and collateral posting |
| Documentation | ISDA with bespoke loan TRS confirmation; loan identifiers, tranche details, reset dates | ISDA with ISDA Credit Definitions; standardized credit event language |
| Goldman product status | In development; no confirmed trades as of June 23, 2026 | Existing market; available for names with active public CDS |
Some press coverage framed Goldman's product as "insurance against defaults in AI loans." That framing is inaccurate. Goldman, when it warehouses the long-side of the TRS against a client's short, bears credit risk, spread risk, and mark-to-market risk simultaneously. A CDS protection seller bears credit event risk alone. The TRS exposure is broader, not narrower, and it accrues continuously rather than at a single trigger point.
What Goldman holds on its books
When a hedge fund buys the short TRS, Goldman holds the offsetting exposure. The bank would likely delta-hedge via CDS where names overlap with public CDS markets, CLO tranches, or secondary loan purchases. For names without liquid CDS markets — which describes most of private credit — Goldman is left with a residual that is hard to hedge cleanly.
Loan pricing is the other problem. Private credit marks are slow to move. They rely on agent bank pricing and manager discretion, not continuous market discovery. A TRS based on stale marks may not recognize losses until they are unavoidable. That lag creates a dispute-resolution risk in stress scenarios, when the client (short) and dealer (long) are likely to hold very different views of fair value.
Debt Surge
Editor's note — The $121.9B data-center credit figure is primary-sourced from S&P Global Market Intelligence, February 2026. The $200B+ private credit AI loans figure is primary-sourced from BIS Bulletin 120. All 2023–2024 data points and 2026E figures are Icarus Asia estimates and should not be treated as primary data. The $300–600B 2030 BIS projection is a wide-range scenario estimate published in BIS Bulletin 120; it is treated as directionally indicative, not as a point forecast. Sources: S&P Global MI, Feb 2026; BIS Bulletin 120; Icarus Asia estimates.
Section 3
Risk analysis: three scenarios
Goldman's product is in development, so risk analysis here is forward-looking and scenario-dependent. The risks fall into three categories: execution risks for Goldman, basis and model risks for hedge fund clients, and systemic spillovers if the product scales.
Bear Scenario — Correlated drawdown with model failure
Trigger: A wave of AI-driven revenue declines hits enterprise software borrowers in 2026–2027. Loan prices fall 10–20 points. Private-credit managers slow their mark-to-market process, leaving TRS valuations dependent on stale inputs.
Goldman's exposure: Short-TRS clients profit. Goldman, which warehoused the long side, faces mounting mark-to-market losses on a portfolio it cannot hedge efficiently because most names lack public CDS. Disputes over fair value delay settlement. Goldman's hedges (CLO tranches, secondary loan purchases) are correlated with the falling reference portfolio, undermining their effectiveness.
Systemic dimension: If multiple banks offer parallel structures and all warehouse correlated long-side exposure, a software credit downturn could produce simultaneous dealer losses across FICC desks. No confirmed multi-bank volumes exist as of this writing; this is a speculative but structurally plausible channel, identified by the FSB as a monitoring priority in its May 2026 report.
Scenario inputs are Icarus Asia estimates based on historical loan drawdown patterns and public BIS/FSB analyses of private credit vulnerability. No confirmed Goldman or counterparty loss data exists. This is a risk illustration, not a forecast.
Base Scenario — Product launches with limited initial volumes
Assumption: Goldman completes infrastructure buildout in H2 2026. Initial trades are small notional, with three to five hedge fund counterparties running bespoke short baskets against 15–25 named software borrowers.
Outcomes: Hedge funds gain a usable expression vehicle for the AI disruption thesis. Goldman earns structuring fees and SOFR-plus-spread funding income. Risk warehousing stays manageable because volumes are small. Pricing disputes are resolved bilaterally. Regulators monitor but do not intervene.
Limitation: At low volumes, the product is a niche service that deepens Goldman's hedge fund franchise without systemic significance. The model and illiquidity risks are real but contained.
Development timelines for complex OTC derivatives structures typically run 6–18 months from informal pitch to first trade (Icarus Asia inference from market structure precedent). Bloomberg reported the tool was "not ready" as of March 31, 2026.
Bull Scenario (for shorts) — AI disruption outruns private-credit marks
Assumption: Agentic AI adoption compresses enterprise software renewal rates faster than consensus expects. EBITDA misses at LBO-backed software companies become visible in 2026 earnings. Secondary loan prices begin falling two to three quarters before private-credit managers adjust their marks.
Outcome: Hedge funds with short TRS positions profit from secondary-market price deterioration while private-credit managers still report par or near-par valuations. The TRS captures the timing gap. If Goldman's reference basket uses secondary-market prices rather than manager-provided marks, basis risk cuts in the client's favor.
Limitation: This scenario depends on Goldman choosing a reference basket with active secondary pricing — a structural choice not confirmed in any public reporting.
Renewal rate and EBITDA assumptions are Icarus Asia inferences based on JPMorgan Asset Management's published commentary on AI software credit risk. No primary financial data for individual borrowers has been used.
The mark-timing gap: how the TRS opportunity works in practice
The bull scenario depends on a specific market dynamic: secondary loan prices falling faster than private-credit managers mark their books. The chart below illustrates this gap. It is hypothetical and uses no actual loan price data — it is included to clarify the mechanism, not to forecast it.
Gap — Illustrated
Editor's note — All values in this chart are Icarus Asia estimates constructed for illustrative purposes only. They do not represent actual loan prices, actual manager marks, or any proprietary pricing data. The chart is intended to explain the structural mechanism described in the bull scenario, not to project actual performance. Sources: Icarus Asia hypothetical model; structural description based on BIS Quarterly Review, Mar 2026 and JPMorgan Asset Management, 2026.
Regulatory posture
Loan TRS fall within existing derivatives reporting regimes under Dodd-Frank (US) and EMIR (EU). Bilateral margin and capital charges may apply depending on counterparty classification. No public record shows specific regulatory intervention targeting Goldman's AI-linked loan TRS. The FSB's May 2026 report on private credit vulnerabilities (P060526) identifies loan-linked derivatives as a potential systemic channel but stops short of specific recommendations. Regulatory scrutiny will rise proportionally to volume growth — that is the consistent historical pattern for new OTC derivatives categories.
Section 4
Key considerations for market participants
-
Product confirmed unready as of March 2026. Bloomberg reported on March 31, 2026 that Goldman told clients the TRS tool "isn't ready yet." No executed trades appear in any public record as of this report's June 23, 2026 finalization date. Any representation that the product is live or that trades have occurred should be verified independently before acting on it.
-
Reference basket composition is unconfirmed. Public reporting describes a focus on enterprise software and AI-exposed corporate borrowers but does not name specific credits. Reference basket construction drives almost everything: mark timing, basis risk, and hedge fund P&L attribution. Confirm basket specifications before entering any trade.
-
Pricing methodology for illiquid loans is a live open question. TRS economics depend entirely on how the reference portfolio is marked. Whether Goldman uses secondary-market prices, agent-bank marks, or model-based estimates will determine how quickly losses are recognized and how disputes are resolved. This is unresolved in public reporting.
-
Regulatory classification may evolve. As TRS on private loans grow in notional size, regulators may require central clearing, enhanced margin, or reporting beyond existing bilateral OTC frameworks. The FSB flagged private credit derivatives as a monitoring priority in its May 2026 report on private credit vulnerabilities.
-
Demand for the product is real and structural. Multiple Reuters and FT sources confirm hedge fund demand for short-loan instruments targeting AI-disrupted software credits. JPMorgan is reportedly exploring parallel structures (Bloomberg, Mar 31, 2026). Goldman prime brokerage data from May 2026 shows hedge funds rotating away from software toward semiconductors — corroborating the directional trade thesis.
-
TRS vs. CDS distinction is material, not semantic. The product transfers total return risk, not credit event risk alone. Hedge funds should not treat it as equivalent to buying CDS protection. The payout profile, trigger mechanics, and basis risk are fundamentally different, and so is the exposure Goldman warehouses on the other side.
Methodology
Research approach and data standards
Research cutoff: June 23, 2026. All claims are based on publicly available information as of that date. No proprietary data, non-public information, or confidential sources were used.
Primary data is sourced directly from named publications: BIS Quarterly Reviews and Bulletins, FSB reports, S&P Global Market Intelligence, and major financial press (Reuters, Bloomberg, Financial Times). These figures are cited in-text and in Appendix B with direct URLs where available.
Icarus Asia estimates are analytical outputs, including interpolations between primary data endpoints, growth multiples calculated from published figures, and forward projections derived from named third-party scenarios. These are labeled "Icarus Asia est." in chart stat cards and in chart footnotes. They should not be treated as primary data or relied upon independently.
Analytical inferences are conclusions drawn by Icarus Asia Research from primary data and public reporting. These include scenario analysis, regulatory interpretation, and market structure assessments. Where a claim is analytical rather than factual, the basis is stated in the text.
Illustrative models (see Figure 3) are hypothetical constructions designed to explain a mechanism, not to project actual performance. They carry explicit "Hypothetical illustration" labels and are not based on any loan price data.
Scope and limitations: The Goldman TRS product described herein had not been confirmed as live as of the finalization date. All structural descriptions are based on public reporting from Reuters, Bloomberg, and the Financial Times, dated March 9–31, 2026. Reference basket composition, pricing methodology, and counterparty details are not in the public record and are noted as unconfirmed throughout. This report describes what has been publicly reported; it does not represent the views of Goldman Sachs, JPMorgan, or any other institution.
Appendix A
Analyst Note
The Goldman TRS concept makes sense given the market setup. Private credit has written $500 billion in loans against a software sector where the competitive moat assumptions were formed before large language models existed. The loans are illiquid, the CDS market is thin, and the equity hedge is noisy. Something like a loan TRS was going to get built.
What's interesting is the timing. The product is not ready as of late March 2026, and Goldman is pitching it informally rather than distributing term sheets. That suggests the infrastructure and risk-management tooling is harder to build than the derivative structure itself. Getting clean, consistent loan pricing for a basket of heterogeneous private-credit borrowers is not a trivial data-engineering problem. These loans weren't designed to be TRS reference assets.
The systemic risk angle is real but premature. Right now this is a development-stage product at one bank. If it scales to five banks and $50 billion in notional, that conversation changes. The FSB's May 2026 report correctly identified derivatives referencing private credit as a transmission channel worth watching. But the channel needs volume before it can transmit anything at systemic scale.
One thing the press coverage has consistently mischaracterized: Goldman is not "selling insurance against AI defaults." When Goldman warehouses the long-TRS position against a client's short, it bears total return exposure. In a scenario where software loan prices fall 15 points without any formal defaults, Goldman still loses on the mark. The exposure is broader than a CDS seller's, not narrower. That distinction has direct implications for how Goldman's FICC desk should be risk-weighted and how regulators should classify the product.
The hedge fund trade that makes sense here isn't a pure directional short. It's a relative-value position: long AI infrastructure (semiconductor-linked debt, data-center bonds with hyperscaler tenants) and short AI-disrupted software (LBO paper for legacy enterprise SaaS). The TRS is the instrument that makes the short leg executable. Whether the trade works depends on one timing question: does AI disruption compress software cash flows faster than private-credit marks allow portfolio companies to refinance? The answer to that question is not in the derivatives documentation. It's in Q3 and Q4 software earnings.
Appendix B
Primary Source Verification
All material claims in this report are sourced below. Figures labeled "Icarus Asia estimate" in the body are interpolations or calculations from primary data; the underlying primary sources are listed here. Analytical inferences are clearly distinguished from primary-data claims in the text.
- Goldman Sachs TRS product — development stage: Reuters, Mar 9, 2026 — "Goldman pitches hedge funds product to bet against corporate loans." reuters.com · Bloomberg, Mar 31, 2026 — "Goldman Tells Clients Eager to Short Loans Its Tool Isn't Ready." bloomberg.com · Financial Times, Mar 2026 — "Goldman pitches hedge funds on strategies to bet against corporate loans." ft.com
- $500B+ SaaS direct-lending balance; ~19% share of DL market: BIS Quarterly Review, March 2026 — "Private credit's software lending meets AI disruption." bis.org/publ/qtrpdf/r_qt2603v.htm Both figures are primary-sourced from this publication.
- $121.9B data-center credit committed in 2025: S&P Global Market Intelligence, February 2026 — "Banks meeting data center demand with billions in credit facilities, bonds." spglobal.com Primary-sourced.
- $200B+ private credit AI loans; $300–600B 2030 projection: BIS Bulletin 120 — "Financing the AI boom: from cash flows to debt." bis.org/publ/bisbull120.pdf The 2030 projection is a wide-range scenario estimate published by BIS; it is treated as directionally indicative with wide uncertainty.
- ~$115–121B public AI-related bond issuance: Loomis Sayles, 2026 — "Impact of AI-Related Debt Financing on the Corporate Bond Market." loomissayles.com · Mellon, 2025 — "Record-Breaking AI-Related Debt Issuance in 2025." mellon.com
- JPMorgan "structural reassessment of software credit risk": JPMorgan Asset Management, 2026 — "How AI is rewriting the software playbook for private credit." am.jpmorgan.com Direct quote sourced from this publication.
- Goldman research — "Cracks in Private Credit" and "Will AI eat software?": Goldman Sachs Research, Top of Mind series and published research, 2025–2026. goldmansachs.com (Cracks in Private Credit) · goldmansachs.com (Will AI eat software?)
- Hedge funds "doubling down" on AI, rotating away from software — May 2026: CNBC, May 25, 2026 — "Goldman Sachs says hedge funds are 'doubling down' on AI, moving toward semiconductor stocks and away from software." cnbc.com
- FSB private credit vulnerabilities report — May 2026: Financial Stability Board, May 2026 — "Report on Vulnerabilities in Private Credit." fsb.org/uploads/P060526.pdf Primary-sourced for all FSB claims in Sections 3 and 4.
- TRS mechanics and CDS comparison: Tavakoli Structured Finance — "Total Return Swaps." tavakolistructuredfinance.com/trs · "Introduction to Credit Derivatives and Credit Default Swaps." tavakolistructuredfinance.com/cds · Investopedia, "Understanding Total Return Swaps." investopedia.com
- Loan TRS legal documentation framework: Weil Gotshal, March 2025 — "Total Return Swaps and Repos: Financing Solutions for Alternative Funds." weil.com · PricewaterhouseCoopers Credit Derivatives Primer for ISDA documentation framework. pwc.com.tr
- JPMorgan parallel product development: Confirmed by Bloomberg, Mar 31, 2026 — "Goldman Tells Clients Eager to Short Loans Its Tool Isn't Ready" — naming multiple Wall Street firms including JPMorgan as exploring parallel swap-based structures. bloomberg.com
- Dallas Fed on AI data-center debt and duration supply: Dallas Federal Reserve, February 2026 — "How AI debt financing impacts duration supply and interest rates." dallasfed.org
This report is produced by Icarus Asia Research for informational purposes only. It does not constitute investment advice, a research recommendation, or a solicitation to buy or sell any security, loan, derivative instrument, or other financial product. Icarus Asia Research is an independent research publication. It has no commercial, advisory, or financial relationship with Goldman Sachs, JPMorgan, or any counterparty, issuer, or institution referenced herein. No compensation was received in connection with this report.
All Icarus Asia estimate figures are clearly labeled throughout and should not be relied upon as primary financial data. Illustrative figures (Figure 3) are hypothetical and carry no predictive value. Readers should conduct their own independent due diligence before acting on any information contained in this report.
Finalized: June 23, 2026