AI earnings and the 2025 lesson for investors

AI earnings

The question about AI has attracted most of the attention so far: how wide and deep will the benefits of AI be? We think the jury is still out.

The AI optimists point to a future in which AI eliminates large swathes of jobs throughout the economy, leading to massive productivity gains. The pessimists point to overreach in the belief in AI’s judgment and acceleration, and hence its ability to streamline tasks without significant supervision. The evidence to date is, at best, mixed.

There is also the second, again, multi-tiered question, which is arguably more important to investors: how, by whom and to what extent can AI be monetised?

AI is not the dotcom bubble, which was capital-light. The inconceivable amount of computing power and electricity utilised by AI, and hence the huge amount of capex being engendered, is more closely aligned, historically, to the telecommunications or the 19th-century railway investment splurges.

The 19th-century railway frenzy drew investors with promises of transformative technology and vast, untapped markets. Railway share prices roughly doubled within a few years, and investment peaked at 7% of GDP, half of all capital investment at the time.¹

However, in 1847, a financial crisis erupted, partly driven by capital being funnelled into unprofitable railway schemes. There was a massive overbuild and poor moats. So, while the development of railways led to a huge general benefit, initial investors lost much of their investment.

By the end of the decade, railway stocks had fallen 65% from their 1845 peak. Prior to the bubble bursting, writer Dionysius Lardner warned that to achieve the expected 10% returns, passenger traffic and revenues would need to increase more than fivefold, with the benefit of hindsight, an implausible outcome given passenger levels at that time.²

Given the desperate attempts by AI builders to cross-subsidise widescale adoption, a similar outcome for AI investors does not seem an implausible outcome.

Therefore, the dollar return on AI will need to be high to generate a decent return on the huge amounts of capital expenditure already spent or expected to be spent in the next few years. Not to mention depreciation on short-lived chips.

And it’s not yet evident what the revenue model will be.

At this stage, the hyperscalers seem to be playing for a winner-take-all advantage to be the preeminent player. Of course, if this works, it spells trouble and wasted capital for the other players. And what if the winner is the low-cost government-subsidised and controlled Chinese model?

Investors will have to assess and consider all of this.

So far, the growth factor has been a beneficiary of the AI boom, but what next?

There is little doubt that the ‘growth’ factor has benefited from the AI boom. The high risk-on sentiment has also contributed to unprofitable and speculative companies outperforming, and the ‘defensive’ quality factor being out of favour.

Comparing the past two instances of investors’ FOMO (Fear of Missing Out) driving up prices, being the lead-up to the Dot-com bubble in 1999 and the global COVID-19 reopening trade in late 2020. What followed was an unwind triggered by rising inflation and bond yields in 2021, adversely impacting longer-duration growth assets, and, in the case of the Dot-com bubble, a collapse in forward sales and earnings expectations, which adversely impacted speculative assets. The quality and value factor fared relatively well. It’s worth noting that the value factor has done well in 2025.

These episodes reinforce the notion that seeking companies with solid fundamentals is prudent. We experienced brief hints of an unwind during US Liberation Day and in early November this year, but these quickly reversed. However, 2026 may not be as kind, with equities priced for perfection.

There are several signals to watch that could cause a factor leadership shift. One is how US businesses decide to pass on tariff-induced higher input costs either through higher consumer prices or by reducing margins. Another is the trajectory of manufacturing activity, which has historically been a leading proxy for “timing” factor rotations. Thirdly, potential concerns about the conversion rate of high AI capex into earnings growth, considering decreasing free cash flow margins and rising financial leverage.

Growth hasn’t been the only factor: The lesson for 2025 and resilient portfolios.

While the growth factor, as measured by the MSCI World ex Australia Growth Select Index, outperformed broad global equity markets in 2025, so too did the value factor, as measured by the MSCI World ex Australia Enhanced Value Top 250 Select Index.

In 2025, therefore, as an investor, you did not need to try to time the AI trade; you just needed to widen it. For Australians building international equity exposure, shifting a slice from concentrated US growth towards World ex-US value was not contrarian; we think it is a way to build portfolio resilience.

As we noted in our recent ViewPoint, “Resilience is a system that can absorb shocks, adapt across regimes and stay the course. That takes conviction and diversification across styles, factors, fixed income and alternatives that respond differently to macro forces.”

This includes edited excerpts from the Q1 2026 global economic outlook, the VanEck ViewPoint: Resilience through selectivity. Read the full analysis

Sources: 

[1] https://www.focus-economics.com/blog/railway-mania-the-largest-speculative-bubble-you-never-heard-of/

[2] https://www.reuters.com/breakingviews/victorian-rail-mania-has-lessons-ai-investors-2024-07-12/