The winners in the AI decade won’t be who you expect

For all the noise surrounding AI over the past two years, we’re still only in the foothills of what the next decade of compute, cloud, and intelligence will look like. And unlike many of the tech narratives of the past: advertising, mobile, social, this cycle is not starting with the consumer. It is starting with infrastructure.

Across conversations with investors, CIOs, and technologists this year, one theme has cut through: AI is no longer a feature. It is an industrial-scale input. And that’s forcing a complete rebuild of the digital economy from the ground up, from silicon and data centres to power requirements and enterprise productivity.

To understand where durable returns may actually emerge, I spoke with Dr David Walsh and Dr Wang Chun Wei of RQI Investors, a quantitative firm that has deeply integrated AI into its investment process.

They argue that the next decade will still be dominated by foundational layers that make AI applications possible. 

"A lot of the more boring productivity components of AI are already getting monetised, and that’s the clearest opportunity over the next five years — because they have real scale and applications across every sector,” Wei says.

And as Walsh added, the power dynamics of this cycle will be defined by whoever controls the infrastructure layer: 

“It's a gold rush. You own companies that are building shovels. That's the way you'd do it.”

What follows is a breakdown of where in the AI value chain economic value is being created, where speculation still dominates, and how long-term investors can position for the infrastructure powering the next decade of AI.

The three-layer stack that will define AI economics

RQI breaks the AI ecosystem into three distinct layers:

1. Semiconductor design and manufacturing (Nvidia, AMD, Broadcom, TSMC)
2. Cloud infrastructure and hyperscalers (Amazon, Microsoft, Google)
3. Software and applications

According to Wei, today’s economic value is overwhelmingly accruing to the bottom two.

“We kind of feel that most of the value add is going to be at the bottom two layers. We're not seeing things to be as attractive at the top layer. Primarily, it's quite competitive and also at the moment slightly commoditised.”

While the application layer will eventually mature, RQI believes the pace will be slower than many investors expect — and far more competitive.

Where monetisation is already happening

Amid excitement around robotics, autonomous agents and generative AI, RQI sees the clearest near-term monetisation in enterprise productivity. These are the tools that automate the “boring” but time-consuming white-collar tasks that exist inside every business.

Wei emphasised, “AI tools to help programmers better code… AI productivity tools to do better document reading… These things are already getting monetised… and it has applications across all sectors.”

This category has three advantages:

• It delivers immediate efficiency gains
• It scales across industries
• It requires no behavioural change from the end user

By contrast, categories like robotics and agent-based systems remain interesting — but still “not very mature or likely to have real applications in the next five years.”

Risks: valuation, capital intensity and the ‘hunger games’

Walsh argues that the biggest risks to the AI thesis are not conceptual but practical. They revolve around over-investment, energy constraints, and stretched valuations.

On capital intensity, he warns, “You might find that some kind of race to the bottom and some kind of hunger games is going on where you've got all these guys continuing to spend: eventually the winner takes all.”

He also flags the extreme multiples now embedded in the megacap AI trade: “These things are trading on 30 to 40 times forward earnings and earnings growth is small… that's clearly valuation risk.”

His guidance for investors is blunt: “Hand waving about disruption or about CapEx is probably too general a question to be answered properly.” Investors need to drill deeper than narratives if they want to survive this cycle.

Walsh continues, “Understand in some detail… what those risks actually represent… If you don't understand it, you shouldn't [take the exposure].”

Why stock-picking won’t work in a 10-year AI horizon

When asked to identify the companies most likely to outperform through 2035, both Walsh and Wei pushed back. The speed of technological change makes long-range stock picking extremely unreliable.

Wei noted, “The companies that might succeed in 2035 probably don't even exist right now.”

RQI’s own portfolios rebalance weekly, and their signals focus on measuring risks rather than forecasting technological winners.

They also caution against the assumption that the best AI applications will dominate the value chain. Vertical integration can work — but only after control of the foundational compute and infrastructure layers is established.

“No company can sustainably own the application layer without first securing the compute and infrastructure layer beneath it.”

What this means for investors: positioning signals in the AI decade

While RQI avoids long-dated stock calls, their framework offers clear signals about where the strongest structural tailwinds lie:

Most attractive (high conviction structural tailwinds)

• Semiconductors and advanced chip design — the shovel-sellers of the AI boom
• Hyperscalers and cloud infrastructure — control access to compute and pricing power

Attractive now (clean near-term monetisation)

• Enterprise productivity software — coding assistants, workflow automation, document AI

Speculative (high uncertainty, long runway)

• AI agents and robotics — exciting, but lacking scale or maturity

Challenging (competitive + commoditised)

• Application-layer AI businesses without infrastructure advantages

The unifying message: returns will come from understanding which risks you choose to own — not from betting on long-dated predictions.

The most important insight: portfolio construction beats thematic conviction

While AI will dominate the market structure for the next decade, RQI argues that thematic correctness is secondary to portfolio construction.

Wei is clear: “Having good portfolio construction and risk modelling is probably even more important than just picking the thematic AI.”

Walsh reinforced the same idea: diversified exposures, disciplined risk management, and clarity around what you actually own will matter far more than chasing the headline AI winners.

The implication is simple: the AI opportunity is real, but durable returns will flow to those who understand the plumbing, not just the promise.