There’s a phrase I think advisers are going to hear a lot more this year.
“I’d quite like some exposure to AI.”
It sounds straightforward enough.
And to be fair, it’s not an unreasonable thing for a client to ask. AI is everywhere. It’s in the news, in workplace tools, in market commentary, in fund updates, and probably in half the conversations clients are having outside the advice meeting too.
But from an investment point of view, “AI exposure” is a much messier phrase than it first sounds.
Because what does it actually mean?
Does it mean buying the big technology names building the models?
The chipmakers supplying the kit?
The cloud companies hosting the workloads?
The data centres housing it all?
The energy infrastructure powering it?
Or the newer GPU cloud providers trying to rent out capacity to anyone who can’t get enough from the usual hyperscalers?
All of those could sit under the AI banner.
But they are not the same investment.
And that is where this starts to become a proper advice conversation, rather than just a topical investment theme.
The bit under the bonnet
For a while, most of the AI conversation was about what the models could do.
Could they write? Could they code? Could they analyse? Could they replace jobs? Could they make businesses more efficient?
That was the exciting bit.
But now the conversation is shifting slightly.
Because the more AI gets used, the more obvious the infrastructure problem becomes.
AI needs chips. It needs memory. It needs data centres. It needs cooling. It needs huge amounts of power. And it needs companies willing to spend a lot of money before the full commercial return is completely clear.
McKinsey has written about how AI workloads are changing the way hyperscalers think about capacity, power and data centre design, with inference expected to become a much bigger part of demand over time:
The next big shifts in AI workloads and hyperscaler strategies
That’s the part I think is easy to miss.
Training a model is one thing.
Running it constantly, for millions of users and businesses, is another.
And once you look at it that way, AI stops being just a software story.
It starts looking a lot more like an infrastructure story.
Big spending doesn’t automatically mean easy returns
The numbers being quoted around AI infrastructure are huge.
IDC has reported that AI infrastructure spending reached around $90 billion in Q4 2025, and expects spending to keep climbing sharply over the next few years:
That sounds exciting.
But big spending cycles are not always clean investment opportunities.
Someone has to fund the build-out.
Someone has to take the risk on demand.
Someone has to buy the chips, lease the space, secure the power, upgrade the network, and hope that future AI revenues justify what is being spent today.
And that is before you get into the question of valuations.
A client might hear “AI infrastructure” and think, “Great, that sounds like a sensible way to invest in the theme.”
But it may still come with a lot of the same risks as other specialist themes:
concentration,
high expectations,
expensive assets,
fast-moving technology,
and the possibility that the market has already priced in a lot of good news.
It might be a good investment.
It might not.
But it definitely needs explaining properly.
The overlap problem
This is probably the most practical bit for advisers.
A client may think they are adding AI exposure for the first time.
But they may already have quite a lot of it.
If they hold a global equity fund, a US equity fund, a passive index tracker, or a growth-focused multi-asset portfolio, there is a fair chance they already have exposure to the big AI names.
Microsoft.
Nvidia.
Amazon.
Alphabet.
Meta.
Maybe not directly in huge standalone positions, but enough for the theme to already be influencing performance.
So when a specialist AI fund, infrastructure fund, or data centre-related investment is added on top, the question is not just:
“Is this a good theme?”
It is:
“What does this add that the client does not already have?”
That is a slightly less exciting question.
But it is the one that matters.
Because without that check, “adding diversification” can quietly become “doubling up on the same story”.
Where the chain gets longer
The other thing I find interesting is how wide the AI value chain has become.
At one end, you have the familiar technology names.
But behind those, you have:
· chip designers,
· semiconductor manufacturers,
· memory suppliers,
· cloud providers,
· data centre owners,
· cooling specialists,
· energy companies,
· grid infrastructure,
and even private credit or infrastructure vehicles funding parts of the build-out.
That doesn’t make the theme bad. If anything, it shows how significant it has become.
But it does mean advisers need to be clear about which part of the chain a client is accessing.
Because buying Nvidia is not the same as buying a data centre operator.
And buying a listed infrastructure fund with some data centre exposure is not the same as buying a pure AI thematic fund.
They may all benefit from the same broad trend.
But they will behave differently.
They will carry different risks.
And they may suit different clients for very different reasons.
A better client question
Instead of asking:
“Do you want exposure to AI?”
I think the better question is:
“What part of the AI story are we trying to access?”
That changes the conversation.
Because then you can separate out:
the exciting bit,
the infrastructure bit,
the valuation bit,
and the suitability bit.
For example, a client might say they want AI exposure because they believe it will drive long-term growth.
That could be reasonable.
But the file still needs to show why the chosen investment is the right way to access that growth, why the level of risk is appropriate, and how it fits with everything else they already hold.
The FCA’s Consumer Duty work on consumer understanding is useful here. Clients need information they can actually use to make informed decisions, not just a neat label on a fund:
Consumer understanding: good practice and areas for improvement
And with specialist themes, that matters even more.
Because the client may understand the headline.
They may not understand what they are really exposed to underneath it.
A few simple checks
Before recommending or documenting an AI-related allocation, I’d want to be able to answer a few plain questions.
What does the investment actually hold?
Not the theme. The underlying exposure.
Is the portfolio already exposed to the same companies elsewhere?
If it is, are we comfortable with that?
What role is this playing?
Is it a long-term satellite? A growth tilt? A thematic allocation? Something else?
What could go wrong?
Not just “markets can fall”. More specifically, what happens if AI spending slows, energy constraints bite, chip supply improves, valuations reset, or revenues take longer to arrive?
And finally:
Could the client explain it back in normal language?
That last one is easy to underestimate.
But if the client can only describe the investment as “AI”, we may not have gone far enough.
The line I’d be careful with
I’d be cautious about simply saying:
“This gives you exposure to AI.”
It might be true.
But it is a bit too neat.
A clearer version might be:
“This gives you exposure to companies involved in building and supporting AI infrastructure. That could benefit if demand keeps growing, but it also increases exposure to a concentrated and fast-moving theme.”
Less catchy.
Much more useful.
And probably easier to stand behind later.
Final thought
AI infrastructure is a genuinely interesting area.
It may well create long-term investment opportunities.
But it is not as simple as “AI is growing, so this should do well”.
There is a lot sitting underneath the theme now: capital spending, energy, data centres, chips, memory, cloud capacity, valuations, and client expectations.
So when clients ask about AI exposure, the most useful thing an adviser can do may be to slow the conversation down slightly.
Not to put them off.
Just to make sure they know what they are actually buying.
Because “AI exposure” is not one thing.
And that is exactly why it needs advice around it.
