Deep Dive: The AI Trade in Markets

Catch-up: Commercial Awareness Update (26th May) | Deep Dive: Why Hedge Funds Are Splitting Into Two Very Different Worlds

Hey {{first_name}} 👋!

In the previous commercial awareness update I asked readers to vote on what they wanted the next deep dive to focus on.

And the winner was…

…The AI trade in markets.

I really like this one.

Because this is one of those topics where the surface-level version is everywhere right now. You cannot open a financial news site without seeing something about AI stocks, chip companies, or hedge funds piling into tech.

But most of the coverage misses the actual interesting part.

The interesting part isn't that hedge funds are buying AI stocks.

It's which AI stocks they're buying, why they're buying those specific ones over others, and what that tells you about how sophisticated investors actually think about a theme as big and noisy as artificial intelligence.

There's a framework behind all of this that the best investors use. Once you understand it, you'll see it everywhere. And it'll make you sound considerably sharper in any interview where markets come up.

By the end of today's issue, you should understand:

Let's get into it.

Last week, Goldman Sachs published its Hedge Fund Trend Monitor. It analyses the holdings of 1,059 hedge funds managing $4.6 trillion in gross equity positions.

The findings were striking.

Hedge funds entered Q2 2026 with the highest ever exposure to semiconductor companies in recorded history. 10% of total long portfolio weight is now sitting in semis alone. To put that in context, the net tilt toward the information technology sector rose by 853 basis points in a single quarter. Goldman described that as the largest quarterly increase to any sector on record.

That's an extraordinary number. These funds aren't dabbling in AI stocks. They're making one of the most concentrated sector bets the industry has ever seen.

But here's the part that most coverage glosses over.

At exactly the same time as piling into semiconductor and hardware companies, hedge funds have been aggressively selling software. The weight in software stocks has fallen to just 6% of portfolios. That's the lowest since 2019.

So sophisticated investors are simultaneously going all-in on one part of the AI trade and walking away from another.

Why?

That's what this deep dive is about.

Before getting into the specifics, it helps to understand the mental model that experienced investors use when a transformative technology comes along.

Every major technology wave in history has followed a similar pattern.

There's a period of enormous excitement and capital investment as the technology is built out. Then, eventually, there's a period where applications built on top of that technology start generating real, sustainable profits.

The two phases don't happen at the same time.

And the companies that win in the first phase are often very different from the companies that win in the second.

The clearest historical example is the Gold Rush.

When gold was discovered in California in 1848, thousands of prospectors flooded in hoping to strike it rich. Most of them didn't. The people who reliably made money weren't the ones panning for gold. They were the ones selling the picks, the shovels, and the denim jeans to the miners.

Levi Strauss didn't bet on finding gold. He bet that people looking for gold would need durable trousers. That was the better business.

The same logic applies to every technology wave since.

During the railway boom of the 1800s, it wasn't the railway operators who consistently made money. It was the steel companies, the coal companies, and the equipment manufacturers who supplied the buildout.

During the internet boom of the late 1990s, most of the famous dotcom companies went bust. But the companies that sold server hardware, networking equipment, and fibre optic cables made a lot of money during the buildout phase, because every company building a website needed their products regardless of whether the website succeeded.

This is what investors call the picks and shovels approach.

Instead of trying to pick which specific AI application or AI company wins, you invest in the layer below. The companies that supply the infrastructure that everyone building AI needs, regardless of who ultimately comes out on top.

That's exactly what hedge funds are doing right now.

The infrastructure layer of AI is everything that needs to exist before any AI application can run.

It includes:

The semiconductors that power AI systems. Training a large language model requires enormous computational power. Running it afterward (what's called inference) requires even more at scale. This has created extraordinary demand for advanced chips, particularly graphics processing units (GPUs) and the specialised memory that sits alongside them.

The data centres that house the computing hardware. The hyperscalers, which is the industry term for companies like Microsoft, Amazon, Meta, and Alphabet, are projected to spend a combined $320 billion on capital expenditure in 2026 alone. A significant portion of that is going into building and expanding data centres capable of running AI workloads.

The power infrastructure that feeds those data centres. AI data centres consume enormous amounts of electricity. Utilities, power transmission companies, and grid infrastructure providers have become part of the AI infrastructure story in ways almost nobody anticipated two years ago.

The networking and optical components that connect everything together. Moving data between chips, servers, and data centres at the speeds AI requires has created a bottleneck that specialist companies are now racing to solve.

Here's what all of these businesses have in common.

Their revenue doesn't depend on any specific AI application succeeding. It doesn't matter which AI model wins. It doesn't matter whether the killer consumer AI app turns out to be a chatbot, a coding tool, or something nobody has built yet. As long as companies are spending billions building AI infrastructure, the picks and shovels businesses get paid.

That certainty of near-term demand is exactly what sophisticated investors are attracted to right now.

Goldman's own research adds to this. Agentic AI, which simply means AI systems that can take actions and complete tasks autonomously rather than just answering questions, could drive a 24-fold increase in the amount of computing power consumed between 2026 and 2030. Chipmakers may face supply shortages for the next 12 to 18 months as new manufacturing plants are built to keep up.

Visible demand. Supply constraints. Clear revenue growth.

That's a compelling investment case. And it's why the semiconductor index has gained roughly 71% over the last 24 months while software stocks have struggled to keep pace.

The application layer is everything built on top of the infrastructure.

It includes the software companies integrating AI features into their products, the AI-native startups building new tools and platforms, and the businesses trying to monetise AI directly through subscriptions, usage fees, or productivity gains.

This is where most of the public excitement about AI lives. The chatbots, the coding assistants, the image generators, the enterprise tools.

And right now, this is the layer that sophisticated investors are walking away from.

The reason is simple, and it's important to understand.

Excitement about what AI might eventually do is not the same thing as evidence that AI is generating predictable, sustainable revenue today.

Software companies have spent the last two years telling investors that AI features will drive higher prices, lower churn, and faster growth. Some of that is happening. But the timeline has been slower than markets initially hoped. Proving that customers will pay meaningfully more for AI-enhanced software, and that those customers stick around, has taken longer than the original narrative suggested.

This creates what one analyst described as a valuation desert.

Software companies are spending heavily to integrate AI. The payoff, when it comes, could be significant. But right now they're in the gap between investment and return, and investors sitting in that gap are essentially waiting.

Compare that to buying a chipmaker where demand is already real and already visible. The infrastructure trade has certainty behind it. The application trade requires patience and a willingness to be wrong about timing.

Hedge funds, which are judged on shorter time horizons than most investors, have voted clearly for certainty over patience.

I started my career at GS. I’ve helped tons of candidates break into GS and other top firms. If you want to maximise your chances of securing world class spring weeks, internships or graduate schemes in the 2026 application cycle (starting soon!) read on…

This month I have 2 slots for my 1-1 coaching programme where I personally coach you to get the best offers possible. It includes:

There’s a 3-month and a 6-month programme. Neither are free given the direct access to my time.

Previous candidates have landed front, middle and back office roles at firms like Goldman Sachs, Citi, J.P. Morgan, BlackRock, PwC, Millennium, Deloitte and others.

Interested? Reply ‘interested’ to this email and I’ll be in touch.

Here's where the story gets more interesting than the headline version.

The picks and shovels framework has a flaw that the best investors are already thinking about.

In the original Gold Rush, the shovel sellers and the gold miners were different people. The shovel sellers had no interest in mining gold themselves. They just wanted to sell shovels.

In the AI boom, the biggest buyers of AI infrastructure are the same companies that are also building AI applications.

Microsoft is spending tens of billions on chips and data centres. Microsoft is also building Copilot and integrating AI into every product it sells. Amazon is building data centres. Amazon is also selling AI services through AWS. Meta is buying chips. Meta is also building AI models for its own platforms.

In other words, the miners are becoming the shovel makers.

And that changes the investment picture in a subtle but important way.

As the hyperscalers build more of their own custom chips and run more of their own infrastructure, their dependence on external semiconductor and hardware suppliers may actually decrease over time. The hyperscalers are already developing their own custom AI chips. ASIC shipments from cloud providers are projected to grow at over 44% in 2026, compared to 16% growth for external GPU shipments.

This doesn't mean the infrastructure trade is wrong. The buildout phase could last years and the near-term numbers remain extraordinary.

But it does mean the simple version of the picks and shovels story, which suggests you just buy chipmakers and wait, misses a more complex dynamic that sophisticated investors are already pricing in.

The investors who will get this right aren't the ones who heard "AI is a big theme, buy chips." They're the ones tracking which parts of the infrastructure remain genuinely in demand from external buyers, which are being replaced by custom in-house solutions, and which application layer companies are actually starting to show the revenue proof the market has been waiting for.

That's the real level of this conversation.

Here's the practical part.

Understanding the infrastructure versus application layer distinction matters for your career in two specific ways.

First, it comes up directly in interviews.

Any interview at an investment bank, asset manager, hedge fund, or consulting firm touching on technology or markets right now is likely to involve AI in some form. The candidates who can explain this framework clearly, and who understand why hedge funds are buying chips and selling software rather than just knowing that they are, will immediately stand out.

The ability to say "it's a picks and shovels trade because near-term revenue certainty sits at the infrastructure layer while application monetisation is still being proven" is a different level of answer from "hedge funds are buying AI stocks."

Second, it teaches you how to think about any investment theme.

The infrastructure versus application framework isn't specific to AI. It applies to any transformative technology. It applied to the internet. It applied to mobile. It'll apply to whatever comes after AI.

Learning to ask "who gets paid regardless of who wins?" when a new technology emerges is one of the most useful instincts you can develop as someone going into finance.

That kind of thinking is what distinguishes candidates who have genuinely engaged with how markets work from those who've memorised facts about them.

Use these resources to get ahead before application season commences:

Breaking Into Banking | Finance Fast Track

If an interviewer asks:

A strong answer could be:

The AI trade is one of the most talked about things in markets right now.

It's also one of the most misunderstood.

Most people hear "hedge funds are buying AI stocks" and think they understand what's happening. But the real story is in the detail. It's in the distinction between infrastructure and applications. It's in the picks and shovels logic. And it's in the complication that the biggest infrastructure buyers are quietly becoming their own suppliers.

Understanding that full picture, rather than the headline version, is the difference between sounding like someone who reads financial news and sounding like someone who actually thinks about what financial news means.

That distinction is what this newsletter is built around.

Keep at it.

Afzal