AI at the Wheel: Portfolio Construction in a Capital Investment Boom

The artificial intelligence boom is unlike anything capital markets have encountered before. It is not simply a technology story, though the technology is extraordinary. It is an asset allocation story — one playing out simultaneously across every major asset class, rewiring correlations, compressing diversification, and demanding a new kind of discipline from allocators who thought they had done the work of building resilient portfolios.

That is the orienting argument of AI-Conscious Asset Allocation: Portfolio Construction in a Capital Investment Boom ¹, a major new research effort from J.P. Morgan Asset Management's Strategic Investment Advisory Group (SIAG). The report is direct about the challenge: "Asset allocators face an old challenge with a new twist: harnessing AI's extraordinary investment potential while maintaining appropriate levels of portfolio risk diversification."

The SIAG's framework rests on three pillars. First, lean into core AI technology with active and targeted equity investing. Second, use the full allocation to spread AI returns and risks across asset classes. Third, reconsider assumptions about diversification that were made before AI existed as a portfolio-level risk factor. These are not abstract principles. They are a response to the specific conditions the report lays out in careful detail.

The Boom in Historical Perspective

The SIAG draws on five historical capital investment booms to provide context: U.S. railroads, telephone networks, petroleum, interstate highways, and fiber optics. The pattern these case studies reveal is useful and sobering. Capital investment booms tend to follow a recurring arc: an origin story, aggressive build-out, a reality check, a shake-out of weaker players, a reorganization toward viability, and finally, broad adoption and productivity gains.

The lesson from the railroad and fiber optic cycles is particularly instructive. Both featured intense competition, parallel infrastructure build-out, widespread investor enthusiasm, and then cascading failures. In the railroad panic of 1873, the collapse of a single major issuer triggered a systemic crisis. In the fiber optic bust of the early 2000s, more than 90% of cable laid in the United States went "dark" for years. The winners from both cycles were ultimately those who held on through the wreckage to benefit from the residual infrastructure.

The AI boom shares structural characteristics with both cycles, but also departs from them in one critical way. "A unique feature of the AI boom is the presence of large, profitable firms with strong balance sheets willing to subsidize the costs of the build-out," the report states. That subsidy, funded through hyperscaler cash flow, debt issuance, and equity raises, has insulated the current cycle from the kind of capital scarcity that ended previous booms. It may also sustain overinvestment well past the point of adequate returns. Whether that subsidy holds through the projected capex peak is, as the SIAG observes, "a central question for investors in AI today."

The Scale of What Is Coming

The numbers are staggering. AI capital expenditure in the United States has risen from under USD 200 billion in 2023 to a projected peak of roughly USD 1.2 trillion before the end of the decade under the SIAG's baseline scenario. Total AI capex through 2030 could reach USD 5.5 trillion, funded across hyperscaler internal cash flow, public equity, investment grade bonds, leveraged loans, securitization, and alternative capital. This is not a technology sector story. It is a capital markets event.

Google now processes more than 3 quadrillion tokens per month, up roughly sevenfold in a year. ChatGPT reports one billion weekly users. AI patent grants reached 131,000 globally in 2024. These are demand signals, not projections, and they underpin the SIAG's view that the current cycle remains self-sustaining: "Today's high expectations are grounded in real demand for AI that should grow fast enough to outrun supply for several more years."

But the report is not sanguine. Vulnerability is becoming visible: the high cost of model training, end-user awareness of rising AI costs, competition from low-cost Chinese models, difficulty accessing power and water for data centers, and a public backlash that is already stalling more than USD 150 billion of data center projects annually. "We see a real possibility that such resistance could complicate AI's future growth in the U.S.," the SIAG states plainly.

Mapping Risk Across the Allocation

The SIAG's taxonomy of AI intensity is one of the report's most useful contributions. Asset classes sit on a spectrum from AI-defensive to AI-concentrated, and the key insight is that most allocators are more exposed than they realize.

At the concentrated end sit U.S. public equity, venture capital, and growth equity, where AI winners and losers are still being sorted. In the middle sit private equity, private credit, and closed-end infrastructure funds, each offering some AI exposure within a diversified portfolio. At the defensive end sit core bonds and core real estate, where direct AI exposure is low and duration or diversification provide natural hedges against a growth shock.

For public equities, the SIAG identifies the semiconductor manufacturing sector, and TSMC and ASML specifically, as possessing the most durable competitive advantages. "These firms command near-monopoly control over the critical chokepoints in chip manufacturing, with few viable competitors in sight." But the report cautions against assuming market enthusiasm for AI translates automatically into durable returns. Active managers using fundamental valuation frameworks will be able to identify entry and exit points "where the market price of a firm's shares does not fully reflect its fundamentals."

In credit, hyperscaler bond issuance has surged from USD 14 billion in 2023 to an expected USD 280 billion in 2026, with USD 300 billion projected for 2027. The investment grade corporate bond market is undergoing, as the SIAG notes, "an unprecedented composition shift." And the growing use of off-balance-sheet financing vehicles and contingent obligations adds layers of complexity that "sophisticated fixed income investors can assess and price" but that passive exposure cannot navigate.

China warrants separate attention. Chinese hyperscalers are spending at roughly 10% of U.S. hyperscaler levels but with greater financial discipline, drawing on internal cash flow rather than capital markets. Chinese AI models lag the top U.S. performers, but "the difference is modest and ultimately may not prove a barrier to commercial implementation." Since the beginning of 2026, Chinese models have accelerated past U.S. models in total token share on OpenRouter. The presence of competitive, low-cost Chinese AI "should limit the ability of any other model to achieve commercial dominance in any market where they are available."

Five Key Takeaways for Advisors and Investors

1. Passive strategies face elevated risk. With AI-related sectors now representing more than 50% of S&P 500 market cap, passive U.S. equity allocations carry concentrated AI exposure without the benefit of fundamental analysis or active risk management.
2. Diversification needs to be recalibrated. The breadth of AI exposure across asset classes means that traditional diversification assumptions may no longer hold. Stress-testing portfolios against scenarios such as model commoditization, a pause in capital flows, or barriers to data center capacity is now essential.
3. Core infrastructure and core fixed income are the portfolio's defensive anchors. Regulated power utilities under long-term contracts offer compelling risk-adjusted returns across AI scenarios. High-quality fixed income provides duration protection against a growth shock.
4. China is a portfolio consideration, not just a geopolitical concern. An allocation to Chinese public equity or venture capital could add genuine diversification relative to a U.S.-concentrated AI portfolio, with a parallel AI ecosystem that is rapidly maturing.
5. Active management has a demonstrable edge in this environment. Fundamental valuation discipline, rigorous accounting scrutiny of off-balance-sheet obligations, and sector-level AI disruption analysis are now prerequisites for navigating the cycle, not optional enhancements.

The SIAG concludes with a call for both conviction and discipline: "Recognizing the massive weight of capital flows into AI-related investments suggests that allocators must consider both diversification within AI and diversification from AI as they navigate the path ahead." That is, in essence, the definition of what it means to invest consciously in an era when AI is everywhere at once.

Footnote:

1 Cembalest, Michael, Jared Gross, et al. "AI-Conscious Asset Allocation: Portfolio Construction in a Capital Investment Boom." J.P. Morgan Asset Management, Strategic Investment Advisory Group, 2026, https://cdn.jpmorganfunds.com/content/dam/jpm-am-aem/global/en/institutional/insights/portfolio-insights/siag-ai-conscious-asset-allocation-portfolio-construction-in-a-capital-investment-boom.pdf. Accessed 30 Sept. 2026.