From AI transmission risk to real-world stress: six months on

At the start of this year, I mapped how stress could travel from an AI capex boom into equity, credit and rates markets if optimism faded. In this blog I’ll examine what we’ve learnt from six months of live data.

Earlier this month, the Bank of England (BoE) published the follow-up to the report that anchored my original blog. Its message was not reassuring: several of the risks it originally flagged deepened over the first half of the year.¹

Our Midyear Global Outlook, published last month, frames the year so far through two forces – AI's acceleration and an increasingly contested geopolitical order – and reaches the same prescription as our original piece: prepare for a range of outcomes rather than predict a single path.

The stress test arrived

We did not have to imagine a valuation reset, as markets briefly delivered one. 

On 23 June, the Nasdaq fell over 2%, while South Korea's KOSPI – up more than 90% for the year and roughly half-weighted to Samsung* and SK Hynix* – dropped almost 10% and triggered a circuit breaker.² Two days later, stronger-than-expected results from Micron* helped send the KOSPI back up more than 5%. It was a live demonstration of the concentration-amplifies-moves dynamic we described in January.

The valuation backdrop has not eased. The BoE notes that the S&P 500's excess cyclically adjusted price-to-earnings yield has edged towards levels not seen since the dotcom bubble.³ Forward earnings measures paint a less demanding picture, but that itself reflects high expectations for future profit growth.⁴

The four channels, revisited

Equities: Concentration risk is already evident, but markets have continued to set records: the S&P 500 reached an all-time closing high on 2 June 2026.⁵ That fits our outlook assessment that resilient growth and earnings upgrades – led by technology and energy – have pushed global equities higher even as leadership has narrowed. Repricing risk and record-setting markets can coexist.

Credit: In January, we cited approximately $300 billion a year of AI-linked bond-market absorption. Morgan Stanley* now forecasts nearly $570 billion of global AI-related debt issuance in 2026, more than double the 2025 level.⁶ Our outlook noted that AI-capex names accounted for roughly 20% of US investment-grade issuance over the preceding 12 months, while Alphabet's* sterling issuance is expected to make it 1%-3% of key UK investment-grade benchmarks. Amazon's* latest $25 billion offering reportedly drew a final order book of approximately $41 billion – around 1.6 times the deal size – versus average US investment-grade oversubscription of approximately 3.5 times through May.⁷

AI borrowers remain less concentrated in credit than in equities, but credit's asymmetric payoff makes issuer selection particularly important, in our view.

Rates: This may be the channel our original piece underweighted most. The Federal Open Market Committee has held its target range at 3.50%-3.75%, but its projections shifted from a 2026 cut towards a possible hike.⁸ 

Our outlook offered a two-sided interpretation: tighter policy raises financing costs for an increasingly debt-funded build-out, but can also restrain rate-sensitive demand elsewhere, creating room for AI investment without equivalent overheating across the wider economy.

History offers a parallel. The Federal Reserve raised rates six times between June 1999 and May 2000, yet the Nasdaq peaked before the final increase and subsequently fell roughly 78% by October 2002, continuing to decline after cuts began in January 2001.⁹ Monetary tightening was not the sole cause, but the episode shows how higher financing costs can interact with stretched valuations and capital-intensive investment.

Market plumbing: The BoE's newest addition is leverage: a pronounced rise in hedge-fund equity exposure, including synthetic leverage facilitated through prime brokers. Morgan Stanley projects an $800 billion-plus private-credit opportunity associated with data-centre investment, led by asset-based finance.¹⁰ Our outlook also flagged that software represents over a fifth of high-yield private credit, against roughly 5% of public high yield.

One part of the plumbing held up better: when the Middle East conflict began, ETF exchange volume reached a record 43% of total US exchange activity, versus a 28% long-term average, highlighting ETFs’ role in investor repositioning.¹¹

Themes behind the theme, revisited

The AI buildout’s operational reliance points remain intact, and several have become more pressing. On the challenge side, data-centre development can take six to 10 years, while land, zoning, supply-chain and financing constraints remain material. 

On the opportunity side, Microsoft* estimates that its newest designs could save approximately 125,000 cubic metres of water annually per facility by eliminating water use for cooling.¹² 

Together, these developments highlight both the constraints facing the build-out and the solutions emerging in response, potentially supporting opportunities across core infrastructure, grid reinforcement, clean energy, battery-energy storage, clean-water solutions, cybersecurity and asset-based finance.

Theory versus reality

The question still open is whether the multi-dimensional diversification¹³ playbook I set out in my original blog worked, and whether portfolios moved the way the framework would have suggested. 

My next blog will provide a scorecard against public benchmarks and a look at where ETF flows actually went in the first half.

 

*For illustrative purposes only. Reference to a particular security or index is on a historic basis and does not mean that the security is currently held or will be held within an L&G portfolio. The above information does not constitute a recommendation to buy or sell any security. Figures shown are approximate, point-in-time figures for illustrative purposes and are not a recommendation. Past performance is not a guide to future performance.

Sources
 
[1] Source: Bank of England, Financial Stability Report
[2] Source: Wall Street is getting trampled by an AI sell-off. South Korean market plunges 10% | CNN Business
[3] Ibid.
[4] Source: Bloomberg, S&P 500 constituent weights and historical concentration data, July 2026.
[5] Source: CNBC, Dow jumps more than 200 points, S&P 500 posts first close above 7,600, 2 June 2026.
[6] Source: Global AI debt issuance to top $500 billion in 2026, Morgan Stanley says | Reuters
[7] Source: Bloomberg News, “Big Tech’s $25 Billion Mega Bond Sales Are Pushing Market Limits”, 8 July 2026; Barclays, US Investment Grade Credit Strategy, 2026.
[8] Source: Federal Reserve FOMC statement and dot-plot coverage, 17 June 2026.
[9] Source: Board of Governors of the Federal Reserve System, FOMC historical materials and federal funds target-rate decisions, 1999–2001. Bloomberg, Nasdaq Composite Index historical price data, March 2000–October 2002.
[10] Source: Morgan Stanley, “Bridging a $1.5tr Data Center Financing Gap”, July 2025.
[11] Source: J.P. Morgan Asset Management, "U.S. ETF Midyear Report: Structural Shifts and Active Solutions", 10 July 2026
[12] Source: Microsoft, 2025 Sustainability Report, 2025.
[13] It should be noted that diversification is no guarantee against a loss in a declining market.