Stress testing AI’s ‘what if’ scenarios

The first scenario is macro. If AI adoption is seen as changing labor-market expectations faster than the economy can adjust, the transmission channel is not limited to the technology sector. It runs through labor income expectations, consumption, policy risk, and volatility. The market does not need proof of large-scale labor displacement to react. It only needs enough evidence to question whether the productivity story also carries a labor-market and broader economic adjustment cost.

We model this as a volatility shock. Current market risk is proxied by the VIX Index. The scenario shocks the VIX Index up by 300 percent from its end-June level of 18 percent. The calibration is informed by the February 2026 AI-related repricing, when concerns about white-collar workflow automation and service-sector business models moved from an academic debate into public market pricing.

The aim here is to estimate what happens to an AI-sensitive, risk-on portfolio when investors reprice AI from a productivity enhancer into a source of labor-market adjustment risk.

Stress-test results

The portfolio's exposure to this volatility shock is highly concentrated. Information Technology, Industrials, and Consumer Discretionary account for 74.6 percent of portfolio weight and contribute 85.6 percent of total portfolio risk. Under the VIX x3 scenario, those same three sectors contribute 81.3 percent of the expected loss. Technology remains the dominant transmission channel. It represents 50.1 percent of the portfolio, contributes 71.4 percent of total portfolio risk, and accounts for 64.4 percent of the projected loss. The scenario therefore does not behave like a broad market move spread evenly across sectors. It behaves like a concentrated risk-on repricing, with the AI-linked Technology exposure absorbing most of the initial loss and the two cyclical overweights adding most of the remainder. 

The second scenario is credit. Many SaaS and software-enabled service companies were financed during the zero-rate period in the aftermath of COVID-19, when growth was cheap and refinancing risk was, funded in large-part by private-credit loans that will begin maturing in 2028. Now, AI can automate workflows and change the value proposition of subscription-based software tools that were previously treated as durable business services.  

Yet, the exposure runs deeper than public software valuations. It sits in the revenue assumptions underpinning that private-credit financing. A business-model adjustment can then become a refinancing challenge, and a refinancing challenge can create pressure in parts of the private-credit market.

We model this through credit spreads. The stress test shocks selected credit curves: the All-Sector Single-B curve by 400 basis points (bps), the Software Industry Single-B curve by 400 bps, and the Software Industry Triple-C curve by 500 bps. During the initial 2025 software repricing, Software Industry Single-B spreads widened by 175 bps before stabilizing and partly retracing. As of July 14, 2026, spreads remain around 100 bps above their pre-repricing level. But, what happens if the market stops treating software-linked private credit as a diversified income exposure and starts treating it as concentrated AI adoption risk?

Stress-test results

The software credit shock again concentrates the expected loss in the same risk-on sectors. Information Technology, Industrials, and Consumer Discretionary account for 74.6 percent of portfolio weight. Under the Software BB +300 bps scenario, they contribute 75.5 percent of the projected loss, with Technology alone contributing 49.4 percent. Under the Software B2 +400 bps scenario, the same three sectors contribute 75.6 percent of the projected loss, with Technology contributing 48.4 percent. The result is consistent across the two software spread shocks. The loss is not evenly distributed across the portfolio, it is transmitted primarily through Technology exposure, then reinforced by Industrials and Consumer Discretionary. The broader credit stress therefore behaves less like a generic spread-widening event and more like a targeted repricing of AI-exposed software and business-service assumptions. 

The third scenario is valuation. The market has capitalized a large amount of AI-related future revenue before that revenue has fully appeared in earnings, cash flows, or enterprise adoption metrics. We use style factors from Axioma's Fundamental Factor Model to simulate a valuation shock: what happens if those revenue expectations prove too optimistic?

If a dominant assumption shared by many investors turns out to be wrong, risk-tolerant factors are likely to absorb the initial losses first. To model this scenario, we simulate a 5 percent negative factor return on Market Sensitivity, Downside Risk, Residual Volatility, Leverage, Investment, and Profit Growth. These are not "AI factors", they are the risk-tolerant exposures most likely to carry the initial loss if revenue timing comes into question.

The goal is to confirm the portfolio can absorb a slower revenue path, even if AI monetization arrives later than current prices assume.

Stress-test results

The valuation scenario is also dominated by Technology, but the transmission is factor specific. Information Technology represents 50.1 percent of portfolio weight, yet it accounts for most of the loss in the Market Sensitivity, Downside Risk, Residual Volatility, Investment, and Profit Growth shocks. Its projected losses are -6.73 under Market Sensitivity, -9.90 under Downside Risk, -9.98 under Residual Volatility, -13.46 under Investment, and -1.07 under Profit Growth. The only exception is Leverage, where the Technology exposure offsets part of the loss and the stress is instead concentrated in Consumer Discretionary, Consumer Staples, Industrials, and Real Estate. The result is consistent with the valuation framing. If AI revenues arrive later than valuations imply, the first-order loss is likely to come through risk-tolerant growth and volatility exposures, not through an even sector-wide de-rating. The portfolio is therefore most exposed where the AI narrative overlaps with high-beta, high-growth, and investment-sensitive factor exposure.