Shade #1 documents the displacement. This shade documents what the financial system does with it. The concern is not that AI eliminates jobs. It is that the architecture through which Americans save, borrow, insure, and retire was built on income assumptions that AI is now degrading, and that the mechanisms connecting these systems amplify shocks rather than absorbing them. The financial crisis of 2008 demonstrated that a shock to one asset class (subprime mortgages) could cascade through securitization, credit default swaps, and interbank lending into a global systemic event. The structures have changed since 2008. The vulnerability has not. It has migrated.
The first link in the chain is private credit. The asset class grew from $46 billion in 2000 to roughly $2.5 trillion in assets under management by 2024, a fifty-fold expansion driven by post-2008 banking regulations that pushed middle-market lending out of traditional banks and into funds with lighter oversight. The IMF’s April 2024 Global Financial Stability Report identified vulnerabilities arising from fragile borrowers, multiple layers of leverage, stale and subjective valuations, and unclear connections between participants. Its October 2025 report found that U.S. and European banks hold $4.5 trillion in exposure to nonbank financial institutions, including private credit, and that stress testing showed these vulnerabilities can quickly transmit to the core banking system. The Federal Reserve Bank of Boston documented that bank lending to business development companies has been growing as a share of both banks’ total loan balances and BDCs’ balance sheets, meaning banks retain indirect exposure to private credit risk even though they do not originate those loans. Moody’s reported that U.S. banks have lent $1.2 trillion to nondepository financial institutions, $300 billion of it to private credit providers, as of mid-2025. The system is opaque by design: private credit assets rarely trade, are valued quarterly using internal models, and the IMF warned that this opacity could incentivize fund managers to delay the realization of losses.
Software is where the exposure concentrates. From 2015 to 2025, private equity firms acquired more than 1,900 software companies in transactions valued at over $440 billion, according to data compiled by Bloomberg. The thesis was simple: SaaS companies produce sticky recurring revenue, high margins, and predictable cash flows, making them ideal targets for leveraged buyouts. Private credit funds financed these acquisitions against the contractual nature of subscription revenue and the high switching costs that kept customers locked in. Software became, in the words of one SaaStr analysis, the engine of the entire unitranche loan market. Roughly 20-25% of all private credit deals are SaaS companies, according to 9fin; UBS puts the AI-disruption-exposed share at 25-35%. AI is now stress-testing every assumption in the model simultaneously. Seat compression, where AI agents do the work of multiple employees and reduce the number of licenses a company needs, undermines the per-seat pricing that generated the recurring revenue. Vibe coding, where users without programming experience build software through AI, lowers switching costs by making alternatives trivially accessible. Piper Sandler downgraded multiple software firms in February 2026, warning that seat compression and vibe coding narratives could cap multiples. Bloomberg Intelligence reported that more than $17.7 billion of U.S. tech company loans dropped to distressed trading levels in four weeks, the most since October 2022, swelling the total tech distressed debt pile to roughly $46.9 billion. In the leveraged loan market, a record $25 billion of software-sector loans now trade below the distress threshold of 80 cents on the dollar, according to Morningstar LSTA data. The contagion is already visible in specific names: Blackstone’s Secured Lending Fund marked its loan to Medallia, a Thoma Bravo-backed software company, down to 78 cents from 87 cents six months earlier. Deutsche Bank got stuck holding $1.2 billion in loans backing a software acquisition it could not sell to investors, a rare “hung deal” signaling evaporated lender appetite. Apollo cut its direct lending funds’ software exposure nearly in half during 2025, from about 20% to roughly 10%, while Arcmont Asset Management and Hayfin Capital Management hired consultants to audit their portfolios for AI vulnerability.
Isaac Kim, a partner at Lightspeed who previously led Elliott Investment Management’s tech private equity business, stated it plainly: “Technology private equity, in its current form, is dead.” The financial engineering that drove the model, buying a software business, improving margins, and adding leverage, “assumes the underlying product remains relevant long enough for financial engineering to work. AI has changed that assumption.” Blackstone’s Jon Gray identified the same risk on Bloomberg Television: the biggest threat is disruption risk, industries changing overnight, as happened to the Yellow Pages when the internet arrived.
The second link is the insurance-to-private-equity pipeline. Since the financial crisis, PE firms have completed over $900 billion in transactions by acquiring insurance liabilities, according to McKinsey, giving them roughly 13% of the U.S. insurance market, up from 1% in 2012. The model works as follows: a PE firm acquires or partners with a life insurance or annuity company, gaining access to a large, long-duration pool of capital that policyholders have deposited for retirement. The PE firm then manages that capital, deploying it into private credit and other alternative investments at higher yields than the insurance company would traditionally earn. Apollo’s merger with Athene is the paradigm. Athene, now a wholly owned subsidiary of Apollo with $331 billion in total admitted assets, provides what Apollo describes as “permanent capital,” decoupled from the fundraising cycles that constrain traditional PE. Athene collects annuity premiums from retirees and savers, Apollo invests the proceeds into private credit, and the spread between the yield on those investments and the rate guaranteed to annuity holders generates Apollo’s spread-related earnings. The arrangement has been widely replicated: KKR acquired Global Atlantic, Brookfield acquired American Equity, Blackstone acquired stakes in Everlake and Corebridge, and the list continues. Athene alone has reached at least 49 pension risk transfer deals covering about 535,000 people, converting corporate pension obligations into annuities managed by Apollo. Several of these transfers are being challenged in lawsuits by pensioners alleging their savings were placed at risk, and a federal court in the Lockheed Martin case denied a motion to dismiss after finding that plaintiffs had plausibly alleged increased risk of harm.
The connection between the first and second links is direct. When private credit funds deploy annuity capital into leveraged software buyouts, and AI disruption impairs the revenue assumptions behind those buyouts, the losses trace back through the chain to the retirement savings of ordinary households. The funds are structured with multiple layers of intermediation, BDCs, reinsurance vehicles, Bermuda affiliates, that make the chain difficult to follow. Bloomberg reported that many PE-backed insurers are shifting liabilities to offshore affiliates subject to less detailed disclosure requirements than in the U.S. The CFA Institute noted that the growing accessibility of private credit products to retail investors, often through interval funds and public BDCs, raises further concerns about liquidity mismatches. Blue Owl Capital, a direct lender with more than 70% of its loans to the software sector, sold $1.4 billion of its loans in February 2026 and simultaneously replaced voluntary quarterly redemptions with mandated capital distributions, a move that sent its shares and those of other alternative asset managers diving. Over $7 billion in redemption requests hit private credit funds in late 2025 and early 2026, according to TheStreet. The structural tension is that private credit assets are illiquid, often with average maturities of 4.4 years according to Federal Reserve data, but the capital funding them increasingly comes through vehicles with shorter redemption horizons.
The third link runs through the labor market to consumer credit. White-collar workers represent roughly half of U.S. employment. The top 20% of income earners drive over 60% of total consumer spending. Anthropic’s March 2026 labor market research sharpens the financial transmission mechanism: the workers most exposed to AI displacement earn 47 percent more than the unexposed group, are more educated, and are more likely to carry the mortgages, credit obligations, and discretionary spending patterns whose disruption triggers downstream financial effects. The University of Michigan Survey of Consumers shows labor market confidence among high earners near historic lows going back to the late 1970s. The New York Federal Reserve’s monthly consumer survey shows unemployment anxiety around record highs. UBS chief economist Arend Kapteyn attributed the trend partly to “AI fear, as white collar jobs are possibly at greater risk.” Economist Claudia Sahm, creator of the widely followed Sahm Rule recession indicator, warned that 2026 could produce a downturn concentrated among the professional and managerial class, as AI displacement, federal workforce cuts, and weakening demand for professional workers converge. If high-income professionals experience sustained income disruption, the effects propagate through the $13 trillion U.S. residential mortgage market, which was underwritten on assumptions of continued employment and income stability for exactly this demographic. Previous mortgage crises targeted subprime borrowers. AI displacement targets prime borrowers: the professionals with the largest mortgages, the highest property tax contributions, and the deepest exposure to service-sector spending in metropolitan economies. The mechanism is different from 2008. The vulnerability is comparable.
The same displacement also erodes the fiscal base. Individual income taxes and payroll taxes account for roughly 85% of federal revenue. Both depend on wages. When AI performs work that previously generated taxable wage income, and the resulting value accrues as corporate profit, stock appreciation, or platform revenue, the tax base contracts even as economic output grows. This is not a projection. It is the arithmetic consequence of labor’s share of income falling to record lows as of Q3 2025, the lowest in nearly eight decades. The public safety net that displaced workers need most becomes least funded precisely when they need it.
These links do not operate independently. They form a feedback loop. AI disrupts software revenue assumptions. Leveraged software companies miss covenants or default. Private credit funds, many of them funded by insurance company capital, mark down their portfolios. Insurance-linked vehicles face redemption pressure or liquidity strain. Banks with exposure to private credit tighten lending. Meanwhile, displaced white-collar workers reduce spending and struggle with mortgage payments. Reduced consumer spending further impairs the revenue of companies across the economy, including the very software companies whose subscriptions are the collateral backing the loans. Each step feeds the next. The SaaStr analysis mapped the sequence: companies miss covenants, but because many loans are covenant-lite the first sign of trouble is a missed payment rather than an early warning. Funds mark down the loans but delay as long as possible because they value their own assets. When markdowns arrive, limited partners want their money back. Funds sell performing loans and liquid assets to meet redemptions, depressing prices further. Banks curb lending. Credit tightens across the board.
This is not 2008. It is important to say why, and where the analogy fails. The June 2025 Federal Reserve stress test explicitly modeled a scenario in which private credit and hedge funds suffered severe losses. The Fed found that banks remained well-capitalized and could absorb shocks from private credit and other nonbank financial institutions without jeopardizing financial stability. The Office of Financial Research’s 2025 Annual Report to Congress concluded that leverage in BDCs and private credit funds is much lower than bank leverage, that losses would be borne primarily by equity holders, and that the total debt owed by private lenders is modest relative to the balance sheets of their creditors. “Taken together,” the OFR wrote, “these facts make it unlikely that distress at private lenders would transmit to the broader financial system.” J.P. Morgan Private Bank argued that fears of systemic crisis are overstated, noting that private credit default rates were around 2.4% in Q1 2025 and that senior direct lending’s starting yields provide a substantial cushion against losses. Brookfield CEO Bruce Flatt said on Bloomberg Television in late February 2026 that fears of a wider financial crisis from private credit’s software exposure are overblown. The closed-end fund structure that dominates private credit, accounting for over 80% of the market according to the IMF, limits run risk by design: long-term capital lockups prevent the sudden, large withdrawals that characterized bank runs in 2008. Plante Moran’s February 2026 analysis found that stress in private credit has been selective rather than systemic, with default rates remaining contained.
There are also strong structural arguments that AI displacement itself will be less severe than the worst-case scenarios suggest. The Jevons Paradox, the historical observation that making a resource cheaper tends to increase rather than decrease its total consumption, applies directly to cognitive labor. When the cost of intelligence collapses, organizations do not simply produce the same output more cheaply. They consume vastly more intelligence, creating new products, services, and roles that were previously uneconomical. The World Economic Forum’s Future of Jobs Report 2025 projected 92 million jobs displaced by 2030 but 170 million new ones created, a net gain of 78 million. Nvidia CEO Jensen Huang pushed back against displacement predictions at VivaTech 2025, arguing that greater productivity typically leads to more hiring, not less. Georgetown economist Harry Holzer noted that AI could reduce demand for college graduates in some roles while making workers without college degrees more productive, potentially narrowing rather than widening inequality. Richmond Fed President Thomas Barkin cautioned against assuming mass displacement: “We should also remember people are going to be enabled.” The actual unemployment rate in professional and business services stood at 4.5% in January 2026, down from a year earlier. IBM announced it was tripling entry-level hiring after discovering the limitations of AI, specifically in roles requiring customer interaction and contextual judgment.
These rebuttals have genuine force. But they also have a structural limitation that matters for this shade: even if the Jevons Paradox holds at the macro level and net employment eventually expands, the transition period is the danger zone for financial stability. Job losses are sharp and localized. Jevons-style job creation unfolds slowly and unevenly, often in different regions and occupations. The financial instruments connecting these systems, leveraged loans with five-year maturities, annuity contracts with thirty-year horizons, mortgages underwritten on current income, do not wait for the labor market to equilibrate. They reprice on today’s revenue, today’s employment, today’s sentiment. A financial crisis does not require permanent displacement. It requires a sufficiently large mismatch between the pace of disruption and the pace of adaptation, sustained long enough for the feedback loops to activate. The Fed stress test modeled credit and liquidity shocks to private credit. It did not model a scenario in which the fundamental revenue assumptions underlying an entire sector of the loan market are being structurally impaired by technological change, because that scenario has no precedent in the stress-testing framework. The OFR’s reassurance that losses would be borne by equity holders is accurate for idiosyncratic defaults. It is less reassuring if defaults are correlated across an entire sector because the same technological force is degrading all of them simultaneously. UBS estimated that default rates could reach 13% for U.S. private credit if AI disruption accelerates, more than three times the projected high-yield default rate. Bloomberg Opinion noted in February 2026 that investors in loans and private credit are still playing catch-up in assessing their AI exposure, and that some software companies are misclassified to hide the true extent of it.
The governed outcome carries a modest positive (+1) because the financial system’s defenses are genuinely stronger than in 2008. Banks hold more capital. Private credit’s closed-end structure limits contagion speed. Regulators are aware of the risks and have begun monitoring them. But governance cannot prevent the underlying disruption from repricing assets. It can only determine whether the repricing cascades or is contained. The specific interventions that would help are prosaic: mandatory mark-to-market for private credit valuations rather than quarterly model-based estimates, enhanced disclosure of insurance company allocations to alternative investments, stress testing that incorporates technological disruption scenarios alongside traditional macroeconomic shocks, and regulatory scrutiny of the growing practice of shifting insurance liabilities to offshore affiliates with lighter reporting requirements. The FSOC has called for enhanced data collection on private credit. The IMF has recommended more active supervisory and regulatory approaches, including improved reporting standards and attention to liquidity risk in retail-facing funds. These are the right directions. Whether they arrive before the stress does is the open question.
Key tension: The financial system was stress-tested for credit shocks, liquidity shocks, and macroeconomic downturns. It was not stress-tested for a technological force that simultaneously degrades the revenue assumptions underlying an entire asset class, displaces the high-income workers whose spending and mortgage payments anchor consumer credit, and erodes the tax base that funds the safety net. Each of these is manageable in isolation. The question is whether they arrive together, and whether the connections between them amplify the shock faster than institutions can respond.