AI and Infrastructure Investment – Insight and Oversights

Investment theory builds on an intuitively appealing proposition – our aversion to risk and enjoyment of reward. For investors in shares, the backbone logic for this trade-off is the Capital Asset Pricing Model (CAPM).

CAPM measures risk in terms of ‘beta’, defined by how a company’s share price co-varies with a broad market index. A stock with beta = 1 should be priced to offer a prospective return in line with the market’s return. Stocks with beta  >1 should have expected returns greater than the market, while the opposite applies for low beta (<1) shares.

From there it is a small step to a big result. Investors cannot expect higher returns for idiosyncratic company risks that are unrelated to beta, even when such volatility is substantial. Beta is the ‘single factor’ in the risk-return trade-off.

Why on earth would that be? 

The argument is deceptively simple. Investors can hedge out other stock-specific risks by holding a well-diversified portfolio. By assumption, this diversification is costless on efficient financial markets – a free lunch for investors.

All good then, apart from financial markets’ habit of charging for this lunch (including a tip).

Part of the problem is what the economist Joseph Schumpeter called the ‘Ricardian Vice’ – mathematically appealing models that assume away relevant context. In CAPM’s case, there is no room for bulls and bears, just super-rational CAPM automatons. There are no market frictions, no transaction costs, it rubs out dynamic factors. And so on.

There is even more trouble when applying the theory to large infrastructure investments. The underlying economic topology is also far from perfect, with complications including:

• Natural monopoly conditions.
• Local and global externalities.
• Long-term uncertainties on technology, demand, and political commitment.

These complexities undermine any summary of equity risk for infrastructure assets by beta alone. In practice, investors face a range of non-diversifiable risks.

Can Artificial Intelligence (AI) help us get a handle on those risks? This was a subject discussed at a recent Investec seminar led by economist Robert Ritz.

The answer is yes.

AI models can integrate live operational data on networks and individual assets – on asset vintage, utilisation, efficiency, maintenance and other factors. When an asset performs better than expected, its future cost of equity should fall. When unanticipated operational risks appear, financing terms can become tighter. AI makes this feedback loop between engineering and finance perfectly possible.

This is just the start. And it is not just an issue for the geeks.

Consider current concerns on energy affordability, something that affects each of us through our energy bills. If unaddressed, there is heightened risk for investors of later political and regulatory intervention. But the change in risk may not be the same for all. Electricity network franchises with major planned additions to their regulated asset bases face greater sensitivity to changes in political sentiment than those with modest capital expansion plans.

A conventional regulatory analysis might conclude the sector merits a single ‘sector’ cost-of-equity allowance estimated by CAPM. Whereas, an AI-enhanced approach could instead provide estimates of the cost of equity at the individual network level, even at the level of single assets.

Does all this matter in terms of hard numbers?

As an example, recent Vallorii research (available on their website) on the impact of rising political risk in the UK estimates cost of equity uplifts of 30-130 basis points in electricity network investments exposed to affordability-driven political risk. Such increases could easily be make or break for final investment decisions by Boards.

So, is that it then, with yet another crushing victory for Machine over Mind?

Well, not quite. Analysis brings insight but also comes with the danger of oversight - the possibility of missing something important.

Any pretence of certainty relating to the cost of equity, even in probabilistic terms, needs to be open to human challenge and ‘intelligence’ beyond logic and data.

Think of it this way. CAPM is like an Ordnance Survey map. It is static and its contours represent an unchanging system (at least if there are no earthquakes).

The map is certainly useful if you are about to go walking in the high mountains. But it is a (very) partial representation of reality. Today, with just a wristwatch as hardware, and armed with a few algorithms, the climber can access a host of other information to calibrate risk dynamically, including predictions for the isobars of weather, and much more. Decision-making improves.

Yet, as any mountaineer knows, when making a final determination on whether to summit, the ultimate arbiter is experience. 

In the end, all investment involves human judgement.

Moreover, value is more than financial value. Since the ancients, infrastructure has reflected higher values including creating a legacy and beauty, as much as about hard returns. Today, nature-based infrastructure solutions build on those foundations – with the potential to yield green not grey dividends and investment returns.

AI is a recipe to help us. The danger comes if we switch off our whole brains when we use it.

Further Reading:

Robert RiItz: Using CAPM to regulate infrastucture. Vallorii website, 2025.

Dieter Helm: The cost of capital, the regulatory asset base, and risk. D. Helm website, 2023.