October 2026 Performance Report

US midterms: less political risk, same AI question

The biggest investment consequence of the US midterms may be the risk that has come off the table.

Going into the election, my concern was a close result followed by prolonged legal challenges or unrest. A decisive outcome makes that much harder. Disputes over individual seats matter less when they are unlikely to change control of Congress.

That is a positive for markets. But removing a political risk does not resolve the question sitting underneath share prices: how sustainable are the earnings investors are buying?

For that, we need to look beyond Washington and into the economics of the AI boom.

Washington looks headed for gridlock.

My expectation is that divided government will produce plenty of political noise and relatively little legislation.

Trump can veto Democratic legislation. Democrats can block his legislative agenda. Executive orders and court challenges are likely to remain prominent, with disputes taking months or years to work through the legal system.

For investors, that can be a reasonably supportive environment. Businesses have more room to plan when the range of possible policy changes narrows.

It does not mean every risk disappears. Congressional opposition cannot prevent every executive action. But it creates another constraint on policy changes that could disrupt trade, investment and business confidence.

There is also a longer-term benefit if stronger oversight improves confidence that businesses will face consistent rules. Companies are more willing to commit capital when commercial outcomes depend on the quality of their investments rather than political access.

Do not assume gridlock means spending cuts.

I am less convinced by the argument that divided government automatically delivers lower spending. Democrats have healthcare and social support priorities that require funding. Trump has shown an appetite for spending too. There may be areas where both sides can agree, even while tax increases face resistance.

Spending could be lower than it would have been under unified Republican control. That is different from expecting austerity.

State governments also matter. Changes in state spending, taxation and regulation can affect the economy even when Washington is stuck.

Most of those effects will take time. Incoming administrations need to take office, draft legislation and implement it before businesses and households feel the full impact.

For the AI investment cycle, however, state and local decisions deserve attention much sooner.

The AI policy debate is about what actually gets built.

Data centres sit at the intersection of several political concerns: electricity availability, water use, land use and the cost imposed on surrounding communities.

Those are legitimate issues. The investment question is how the response changes construction costs, completion dates and the amount of computing capacity delivered.

My assessment is that much of the opposition so far has had a limited effect on commercially credible projects.

A speculative development application is different from a project with funding, experienced operators and committed customers. Counting every abandoned proposal as lost AI investment can give a misleading picture of the construction pipeline.

The same distinction applies to regulation. To date, the political headlines have been performative posturing.

Water requirements can be easily addressed with added costs and different cooling systems or recycling. If you don’t make data centres install water recycling, they won’t. But water requirements are not a blocker.

Restrictions on grid connections seem prohibitive at first glance. But for 2027/28, 80% of projects are already bringing their own power supply because a timely connection was unavailable. So regulation to stop data centres connecting to the grid when they weren’t going to anyway is performative.

Texas is a particular area to watch, given its role in data centre development. But the useful evidence will be in the actual rules governments implement and the projects those rules affect.

Campaign promises alone tell us relatively little about how much capacity will reach the market. A slower build could extend the boom. A sharp, widespread halt to data centre construction would be a serious risk. It could hit investment spending and companies across the AI supply chain.

That is not my base case.

A more modest slowdown could have a different effect. It might reduce pressure on construction costs, equipment and energy infrastructure while giving investors more time to assess demand.

The industry is in an arms race to build capacity. Eventually, supply will catch up with demand. We do not know when. nIf capacity arrives more gradually, that point could move further out. Existing operators may enjoy favourable conditions for longer, and developers may have more opportunity to adjust before committing to the next project.

Slower construction would affect companies differently. But it is too simplistic to assume that anything slowing the build must be bad for every AI investment.

The key distinction is scarcity versus a lasting advantage.

The more important investment question is where today’s AI profits come from. Shortages allow businesses throughout the supply chain to charge more. That can benefit chipmakers, data centre operators, equipment suppliers and companies with access to power or approved sites.

Some of those profits may reflect durable competitive advantages. Others will disappear as supply expands. NVIDIA’s investment case hinges on the strength and duration of its technological and commercial lead, as well as the price investors pay for that advantage. There is an argument for years or decades of technological leadership.

Compare that with a supplier of ordinary earthmoving equipment used to prepare a data centre site. Strong construction demand might temporarily improve its pricing. Competitors will respond by supplying more equipment, bringing returns back towards normal levels.

Both businesses participate in AI spending. Their ability to retain the benefits may vary widely. That distinction matters far more than putting every company connected to AI into the same investment bucket.

Cheap shares can still have expensive earnings.

Valuation adds another complication.

A company can look cheap relative to current earnings while those earnings are unusually high. If shortages ease and margins fall, the apparent bargain can disappear.

Our valuation work has international shares both cheaper and growing earnings faster than Australian shares. But that comparison still depends on the earnings underneath it.

The questions we need to ask of each company in the AI chain are straightforward:

• How much profit comes from temporary shortages?
• How quickly can competitors add supply?
• What keeps customers from switching?
• How much earnings decline does the share price already anticipate?

A business that maintains strong margins for several years has a different value from one whose margins begin to normalise within months.

A low price-to-earnings ratio cannot answer that question on its own.

What this means for portfolios

Our approach is to work through those differences company by company.

The pullback over the last few weeks has created an opportunity to reassess which businesses had fallen further than their prospects justified, and which still depended on optimistic assumptions.

We continue to look for AI exposure where the competitive position, earnings outlook and valuation make sense together. As prices change, the relative attraction of those businesses changes too.

The midterms have reduced one source of uncertainty. Investors can give more attention to the businesses, their customers and the returns being earned on the enormous amount of capital going into AI.

The central question is how long each company can hold on to its profits—and how much we should pay for them today.