Ask a chief investment officer about “system-level investing” and you will probably get a puzzled look.
Ask the same CIO about heavy concentration in a handful of AI stocks, thin governance at the technology companies involved and the possibility that an AI bubble becomes a systemic problem rather than a bad year for returns, and you will get a nod.
That gap is the problem. Sustainable investing professionals are talking a great deal about system-level investing and system-level stewardship. The terms are not always clearly defined, and the conversation often returns to the themes the field has worked on for years: climate, biodiversity, workforce. Those risks are real, and the work on them matters.
But one system-level risk has, arguably, gone beyond the rest in urgency, and the CIO already recognises it. It is moving faster than the slow-building risks the debate usually centres on and is concentrated in a small number of firms. Financial regulators, too, now treat its rapid adoption as a stability issue. If system-level thinking is to mean something, AI is where it should be tested.
None of this is an argument against AI. The potential gains, from productivity growth to drug discovery and earlier diagnosis of disease, are large and adoption will continue. As Goldman Sachs chair and CEO David Solomon has put it, the question is not whether AI will advance, but whether it advances well.
“The size of the opportunity is part of the problem: it makes the risks easy to discount”
For an investor, that starts with precision about what AI is. In our view, it is both a source of return and a source of risk, better understood as a risk factor than as a theme, and one that investors can choose to hold more or less of.
Every asset has some sensitivity to AI. Short-dated government bonds sit at the low end. Data-centre developers, AI-exposed growth equities and parts of private equity sit at the high end. Credit and real assets sit somewhere in between, in ways that are hard to see from asset-class labels. Measured against the total investable universe, every portfolio has an AI beta, whether or not anyone chose it.
AI: A systemic risk
The size of the opportunity is part of the problem: it makes the risks easy to discount. In the narrower sense that matters here, a systemic risk is a malfunction that can cascade through the system, not simply a risk that moves with the market.
In a letter to G20 finance ministers, Financial Stability Board (FSB) chair Andrew Bailey called the potential for advanced AI models to conduct cyberattacks the most immediate concern for a highly interconnected financial network. He also warned that many jurisdictions lack the protocols to manage the development, release and deployment of frontier models.
The Bank of England’s latest Financial Stability Report points the same way: a significant increase in cyber and operational risk, more stretched valuations and AI firms drawing on credit markets at an unprecedented pace.
Not everyone agrees. At a closed-door investor conference in October, US Treasury Secretary Scott Bessent reportedly said he sees no systemic problem with the AI boom and that it is not the dot-com bubble repeated. Those remarks speak mainly to valuations.
“A total portfolio approach, with an asset and liability lens, shows AI exposure across equities, credit, private markets and real assets together”
The FSB’s concern is different: how tightly AI is woven into a highly interconnected system. A sound boom can still carry systemic cyber, operational and concentration risk, and the case for treating AI as a systemic risk does not rest on calling a bubble.
For an asset owner, the exposure runs through both sides of the balance sheet. On the asset side, it reaches well beyond the equity weighting: investment-grade and private credit, data-centre construction, real estate and infrastructure. On the liability side, one plausible channel is lower contributions if job displacement falls unevenly on particular cohorts or sectors.
Beyond these sits a low-probability, high-severity tail of loss of control over powerful systems, which no portfolio can diversify away.
What asset owners can do
No asset owner can mitigate these risks alone, and neither can any one AI developer. A lab that slows down by itself cedes ground to those that do not. The FSB chair made the point from the regulator’s side: the risks will not respect national borders. Effective mitigation will need a system-level mix of industry self-regulation and public policy.
This is where the case for system-level stewardship is strongest, and where it is easiest to over-reach. Recognising that AI is a systemic risk does not tell an investor what to do about it. The answer differs by role.
- Treat the index as an active choice. Index rules decide what passive money owns. NASDAQ’s new fast-entry rule lets a newly listed company ranking in the top 40 holdings of the NASDAQ-100 join after 15 trading days, where the wait had previously been about three months. Early inclusion of AI listings, similar to index concentration more generally, creates active risk against the total investable universe. It may be the risk an owner wants. It should be a decision, not a default. A CFA Institute survey found strong support for keeping existing index-inclusion standards.
- Use a total portfolio view. A total portfolio approach, with an asset and liability lens, shows AI exposure across equities, credit, private markets and real assets together. Most funds measure it one silo at a time.
- Keep making the case for minority shareholder rights. Governance protections matter most where index inclusion removes the option not to own. Institutional investors are especially alert to the need for sunset clauses where dual-class structures exist. Large public pension funds called for the dual-class structure at the year’s largest IPO so far, SpaceX, to be eliminated. Sunsetting dual-class shares is a reasonable ask.
- Engage companies on board oversight of AI risk. Accountability sits with people, not models, and with companies that deploy AI as much as those that build it. Boards should be able to say who is accountable for what, and how they would know.
- Test your theory of change. Before pressing for a particular regulatory outcome, an owner should ask whether it has a credible route to influencing self-regulation or policy in a way that serves beneficiaries’ long-term interests, and whether other tools would do more.
What active managers can do
An active manager is paid for alpha over the horizon the client sets, from a quarter to five years, while looking through to business model strength over a longer period, perhaps 10 years.
For AI, three things follow: adjust AI beta deliberately, favour businesses with durable long-term economics over riskier ones and understand how self-regulation or policy might affect each company.
Debt markets are a good example: hyperscaler borrowing is now large enough that analysts argue it affects government bond yields, and a manager who understands who buys long-dated debt is better placed to judge that.
“The question for investors is whether they help it advance well and whether they are paid for the risk they take while it does”
The opportunity is real. The skill is in separating the businesses that will compound from those that will not and in knowing what you are being paid for the risk.
Because AI beta differs so widely across portfolios and clients, it is not an active manager’s place to push a specific policy goal on their behalf. Stewardship should aim to strengthen and de-risk the companies the manager owns, one by one, and to understand the risk those companies are actually running. That produces insight asset owners can use in their own decisions.
System-level investing will be judged by whether it helps investors with the system-level risk in front of them. On AI, that means asking narrower questions than the debate usually does. How much AI risk do we hold, and did we choose it? What are the companies we own actually risking? Where can we credibly move the system, and where should we leave that to others?
Investors who can answer those three questions will be better placed than those with a view on everything and a plan for none of it. AI will advance and the potential is large. The question for investors is whether they help it advance well and whether they are paid for the risk they take while it does.
Jonathan Bailey is global head of stewardship and sustainable investing at Neuberger.
