Am I overexposed to AI stocks? How to check your portfolio's AI concentration

If you are a passive index investor or hold a workplace pension – probably more than you think, and almost certainly more than you chose. The key global equities tracker is currently holding 72.03% in United States and 28.87% in Information Technology (MSCI World factsheet, 31 July 2026), and the Bank of England's July 2026 Financial Stability Report finds that AI-related companies now account for around half of the S&P 500, up from roughly a quarter in 2022. Because that exposure arrives through workplace pension defaults and index funds rather than through decisions, most people have never measured it. You can: list every account, look through each fund to the shares underneath, apply one consistent definition of what counts as AI, and total it across everything you own. The number itself isn't the answer, though. The answer is whether your plan still reaches your goals if that one theme has a bad decade.

What actually counts as an "AI stock"?

With all the news and recent market volatility around AI and whether it is a bubble, many investors ask themselves: how much of my portfolio is exposed to this one theme?

To measure it, you have to define it – and there is no clear definition. That's not a technicality. Where you draw the boundary changes your answer by a wide margin, so the useful thing is to pick a definition and apply it consistently. Three groups cover most of what UK investors actually hold.

Pure plays. Companies whose share prices are most directly exposed to expectations of AI demand – chip designers, AI-focused cloud providers, AI software firms. They move most on any change in the story about how much will be spent on AI, in both directions.

Hyperscalers and Big Tech. The very large platform businesses that are also the primary spenders driving the cycle. Two things are true here and it's worth holding both. These are genuinely diversified businesses – advertising, cloud, retail, marketplaces, devices – so calling them "AI stocks" flattens a lot of real revenue that has nothing to do with AI. And their share prices have nonetheless been moving on the same capital-spending narrative: in late July 2026, Alphabet's shares fell around 7% after it guided capital expenditure up to as much as $205bn (NBC News, July 2026). Diversified as businesses; correlated as holdings.

Supply chain and enablers. The picks and shovels – chip manufacturing and lithography equipment, memory and storage, networking, and the power and cooling that data centres run on. This group keeps widening. Goldman Sachs Research describes the AI trade moving through phases, from the first chip winners out to infrastructure, then to companies embedding AI in their products, then to broad productivity beneficiaries (Goldman Sachs). Each phase pulls more of the market into the definition.

Illustrative and non-exhaustive. Companies are named only as examples of what these categories contain. Nothing here is a recommendation to buy, sell or hold any investment.

Where you draw the line is a judgement call. Count only pure plays and your number looks modest. Count the hyperscalers and it roughly triples for most people. Neither is wrong. What matters is that you use the same definition every time you check, so the figure you get next year is comparable with the one you get tonight.

How much AI is already inside the portfolios you hold?

Here is what surprises people. You probably don't own three diversified funds. You own the same companies three times.

Most UK portfolios are built on a small number of index families – the S&P 500, the MSCI World, the Nasdaq 100. They sound like different propositions: one American, one global, one technology. At the top, they hold much the same companies. The pure plays and the hyperscalers from the table above sit near the top of all three.

The scale of that overlap is on the record. As at 31 July 2026, the MSCI World is 72.03% United States and 28.87% Information Technology (MSCI World Index factsheet) – so a "global" tracker is mostly one country, and close to a third one sector. And the Bank of England's July 2026 Financial Stability Report finds that AI-related companies now account for around half of the S&P 500, up from roughly a quarter in 2022 (Bank of England, July 2026). That is the Bank's view, not a forecast, and not ours.

The concentration has kept tightening since. Between 13 July and the end of August 2026, the S&P 500 added $1.75 trillion in market value – and nearly 80% of that came from the technology sector alone. Two companies accounted for more than 80% of the index's entire gain, while the other 71 technology stocks in the index lost value between them (Bespoke Investment Group data, reported 30 August 2026). An index at a record high can still be a narrower index than it was six weeks earlier.

Data shown in the chart above

Measure	Value
MSCI World – United States, 31 July 2026	72.03%
MSCI World – Information Technology, 31 July 2026	28.87%
AI-related companies as a share of the S&P 500, 2022	Around a quarter
AI-related companies as a share of the S&P 500, 2026	Around half

Now think about where those indices actually live. The global tracker in your stocks and shares ISA. The default fund in your workplace pension, which is usually a global equity strategy. The US or technology fund in the SIPP from an old employer. Three wrappers, three separate decisions, three different fund names – drawing on the same few dozen companies.

That is what the table in the previous section is for. Those companies don't sit in one holding you can point to and weigh. They sit across your funds, in every account you own, in a total nobody has ever shown you.

How to check your own AI concentration, account by account

This is the part no one does, because no single screen anywhere adds it up for you. It takes an evening.

1. List every account

   All of them. Your current workplace pension, and the ones from previous employers you haven't logged into for years. Stocks and shares ISA. Lifetime ISA. SIPP. General investment account. And the one most people forget, which is often the single largest concentration in the whole picture: employer share plans – SAYE, a Share Incentive Plan, vested RSUs. If you work in technology, that last line can dwarf everything else, and it sits in the same sector as much of the rest.

2. Get each holding's actual contents

   Four documents, in ascending order of detail. The fund factsheet gives you the top 10 holdings and a sector breakdown, usually monthly – enough for a first pass. The full holdings list on the fund manager's own website is the real answer; ETFs typically publish theirs daily, funds monthly or quarterly. The annual and half-yearly report carries the complete schedule of investments. And the key information document gives you the risk and cost summary – note that this is mid-transition in the UK: PRIIPs KIDs and UCITS KIIDs are what you'll see on most products today, while the FCA's new Consumer Composite Investments Product Summary has been in force since 6 April 2026 and becomes mandatory on 8 June 2027.

   One practical note, because it's where most people give up: for a workplace pension default fund, the factsheet is usually on the provider's main website rather than inside the app you log into. Search the fund's full name rather than hunting through the app.

3. Look through to the underlying shares, behind funds

   Two differently named funds routinely hold the same top ten. A global tracker, a US fund and a "technology-light" managed fund can all be leaning on the same handful of companies. Morningstar's Instant X-Ray, with the holdings-overlap view, does this free – you enter holdings manually and it has no UK wrapper logic, so it's a bit of work, but it does the job. Be careful with the X-ray your platform offers: it usually covers only what you hold with that platform, and your concentration is the sum across platforms, not within one.

4. Apply your definition consistently

   Use whichever of the three groups from Section 1 you decided to count, and count them the same way everywhere. Then express the total as a percentage of everything you own – not of one account.

5. Write the % weight down, with a date stamp

   When the AI theme rises faster than the broader market, its weight in your portfolio increases without you buying a thing. A number with a date on it is what lets you see that happening.

The total is the only figure that matters here, because hardly anybody's ISA is diversified against their pension. Four sensible, separately chosen funds can add up to one large, unchosen position.

How much is too much? Three honest approaches

Here are three approaches, catering to different risk appetites.

The conservative approach

Investors taking a conservative view want their total single-theme share brought back toward the market's own weight, or below it. The usual routes are directing new contributions elsewhere rather than selling, and adding exposure that behaves differently – equal-weighted rather than market-cap-weighted funds, value or smaller-company funds, or more outside the US, in developed markets and emerging ones. Some hold capped versions of an index, which limit how large any one constituent can become.

The true risk here is opportunity cost. If the theme keeps running, that approach trails the index, possibly for years, resulting in underperformance. That pill is often harder to swallow when it's right in front of you. However, those who want to limit downside of a sector shock, may want to consider this approach.

The moderate view

Investors taking a moderate view accept market-weight exposure as a reasonable default but refuse to let it drift too far. In practice that means deciding in advance what level would prompt them to act, and rebalancing when the boundary is hit.

The downside here is you often have to sell on a high – which is harder to do psychologically. In addition, if held outside an ISA or pension, rebalancing can lead to realising a gain and a tax bill in the short term.

The adventurous view

Investors taking an adventurous view can deliberately run an overweight position, on the judgement that the results will remain strong, and expectations met. They are ready to ride the wave down as well as up.

This is a double-edged sword. The same concentration that produces the returns is the concentration that exposes you to the falls, and July 2026 is a real preview of the volatility involved rather than an outlier to be waved away: the main semiconductor index fell 21% that month, its worst since October 2008, and closed up or down by at least 4% on nearly half the month's trading days (Fortune, 2 August 2026). It recovered much of that ground through August. That is the same point from the other side: a holding that can move like that in one direction can move like that in the other.

Here's the part that actually decides it, and it isn't a view on AI. It's how much of your financial plan depends on this theme working out, and how soon you need the money. A 34-year-old and a 61-year-old with identical portfolios have different answers, because one has thirty years to recover from being wrong and the other has three.

One thing worth saying: holding less of a concentrated position is not protection. It changes the shape of the risk, not the existence of it. A less concentrated portfolio still falls.

Keeping an eye on it: what to watch and when to rebalance

Three things are worth watching, in this order. Your own allocation, recalculated on a schedule – twice a year is plenty, because the drift is slower than the news cycle makes it feel. The index's own weights, published monthly by the index provider, which tell you whether the change came from the market or from you. And your fund factsheets, for changes in the top ten.

On rebalancing, three approaches, described factually.

Calendar rebalancing – a fixed date each year, back to your target weights. It is the simplest and, for most people, the hardest to talk yourself out of, which is its main virtue.

Threshold or drift-band rebalancing – you act when an allocation moves more than a set distance from its target, commonly around 5 percentage points in absolute terms, or the "5/25" relative rule. Research from Vanguard and Morningstar generally finds threshold-based approaches compare well against pure calendar ones, because you trade only when the drift is actually meaningful (Vanguard research).

Rebalancing with new money – you direct new contributions toward whatever is underweight instead of selling what's overweight. Slower, but it avoids realising gains outside a wrapper, and it's the method most people can actually keep up for a decade.

Two practical points for UK investors: Inside an ISA or a pension, rebalancing has no capital gains tax consequence. In a general account, selling to rebalance may crystallise a gain, which triggers a capital gains tax and is a real reason many people prefer the new-money route. And dealing charges and time out of the market are small but not nothing.

Then the behavioural truth, which is the actual reason to write any of this down now: on the day your rule applies, you will not want to follow it. Rebalancing means selling what has been going up and buying what hasn't, at precisely the moment that feels least sensible. A rule written in a calm month is easier to obey in a loud one.

The question that actually settles it: what does your concentration do to your plan?

"How much AI is too much?" has no general answer. "Does my plan still reach my goals if this theme has a bad decade?" has a specific one.

First, what Allocatewise will not do. It won't tell you what your AI exposure should be. It doesn't rate or value any company. It doesn't connect to your accounts and pull your holdings in for you – you enter them. And it isn't advice: if what you need is a decision about your own portfolio, MoneyHelper offers free, impartial, government-backed guidance, and a regulated financial adviser can advise you properly.

What it does is answer the question this article has been building toward. You enter what you hold across your accounts and build the picture yourself in the Portfolio Builder – it's user-directed, so you make every decision and construct every version. From there you can run up to 1,000 Monte Carlo simulations for a probability-based read on whether your dated goals are still reachable, stress-test the portfolio you actually hold against real historical crises – including the dot-com unwind, which is the closest thing we have to a single-theme re-rating playing out – and see all of it in real, inflation-adjusted terms rather than headline pounds. You can explore how a different asset mix compares against your current one, side by side, before moving any real money. And you can save both (up to 10 profiles), so the next time you run your audit it's a comparison rather than a fresh start.

It's arithmetic, not an oracle. Every figure is deterministic on the inputs you give it, the assumptions are inspectable, and none of it predicts what AI stocks will do next – because nothing can.

There is no universally correct level of AI exposure. There is only the level that still lets your plan work. That's not a matter of opinion, and it's not something you have to guess at – it's a number you can find. It's free to start, with no card required at sign-up; the goal-aligned allocation comparison sits on the premium tier.

If you want the method behind the projections, read the companion piece on whether you're on track to reach your financial goals. For the real-terms side of the picture, see how to think about your portfolio and inflation.

Not regulated by the FCA. Not financial advice. Any companies named are examples used to explain index composition. Nothing here is a recommendation to buy, sell or hold any investment. The value of investments can fall as well as rise and you may get back less than you put in. If you want advice on your own portfolio, speak to a regulated financial adviser; MoneyHelper offers free, impartial guidance.

Sources

1. Bank of England – Financial Stability Report, July 2026 (AI-related concentration; share of the S&P 500; amplification risk) – bankofengland.co.uk
2. Bank of England – Financial Policy Committee Record, July 2026 – bankofengland.co.uk
3. MSCI – MSCI World Index factsheet (United States and Information Technology weights, as at 31 July 2026) – msci.com
4. FCA Handbook – COLL 5.2, General investment powers and limits for UCITS schemes – handbook.fca.org.uk
5. FCA – Consumer Composite Investments regime (in force 6 April 2026; mandatory 8 June 2027) – summary
6. Bespoke Investment Group, via Yahoo Finance – S&P 500 market-value gains by sector since 13 July 2026, reported 30 August 2026 – finance.yahoo.com
7. Fortune – "Wall Street's favorite bet comes undone as chips whipsaw market", 2 August 2026 – fortune.com
8. NBC News – Nasdaq 100 correction and Alphabet capital-expenditure guidance, July 2026 – nbcnews.com
9. Goldman Sachs Research – AI infrastructure and the phases of the AI trade – goldmansachs.com
10. Vanguard – threshold-based rebalancing research – corporate.vanguard.com
11. MoneyHelper – free, impartial government-backed money guidance – moneyhelper.org.uk