AI Stock Pullbacks: How Much of Your Portfolio Is Really One Bet?

Chip stocks have been grinding through their worst stretch in 16 months, and MarketWatch — a secondary report, not a filing — frames the cause as “AI fatigue”: Wall Street growing less patient with heavy AI spending, with attention turning to Anthropic’s investor day for signals on how long the capital expenditure cycle runs. That is the market story. The portfolio story is different, and it is the one worth your twenty minutes, because AI stock pullbacks do not test your view on artificial intelligence. They test your position sizing.

Here is the thesis I will argue: for most diversified retail investors, the risk exposed by an AI-led drawdown is not a stock-selection error but an unmeasured concentration that arrived through the back door — through index funds, sector funds, and a workplace plan nobody re-read. You cannot fix that by forecasting the next chip cycle. You can fix it by adding up what you already own.

Consensus: the dip is the entry point

The prevailing view deserves a fair statement before I take it apart. It runs roughly like this: AI infrastructure demand is a multi-year capital cycle, not a quarter; the companies supplying compute have real revenue and real order books rather than pre-revenue narratives; and sentiment-driven selloffs in a secular growth theme have historically rewarded buyers with patience. On that reading, a 16-month-worst stretch is noise inside a longer signal, and the main risk is being underweight.

This is not a straw man. It is a coherent, widely held argument, and parts of it may well prove right. The problem is that it answers a question about direction while most investors are carrying a problem of magnitude.

Three funds, one bet

Consider how concentration actually accumulates in an ordinary account. An investor holds a broad market index fund, because that is the sensible default. They add a technology sector fund, because that is where the growth has been. Their retirement plan’s default option is a large-cap growth fund. A fourth sleeve holds a thematic AI or semiconductor ETF bought in the last two years.

Four line items. Four different tickers, four different fund companies, four separate decisions made months apart. And potentially the same dozen companies underneath all of them, at escalating weights. The investor experiences this as diversification — four holdings, several asset managers, a mix of broad and targeted. The market experiences it as one position.

This is the specific mechanism that AI stock pullbacks expose, and it is why the drawdown feels disproportionate to people who believed they were diversified. Correlation does not care that you bought your exposure in four separate transactions. When a single theme drives a selloff, holdings that looked distinct on your statement move together, because they are substantially the same underlying companies responding to the same repricing of the same spending assumption.

I want to be precise about what I am not claiming. I am not telling you that the top holdings of a broad index now constitute some specific percentage of it, because I will not state a figure I cannot source to the fund’s own current documents. That number is knowable — it is printed in every fund’s monthly factsheet and in its filings — and it is your number, not mine. The discipline is looking it up.

📊 What a 25% drawdown costs at different weights

Concentration risk is often discussed as a vague virtue problem. It is arithmetic. The table below shows what a single theme falling by a given amount does to a total portfolio at different exposure levels. These are not predictions and not historical returns — they are multiplication, shown so you can locate yourself on it.

Read the bottom-right corner carefully, because that is where behaviour breaks. A 4% portfolio decline is an unpleasant month that most people hold through. A 20% decline is the point at which investors sell — not because their analysis changed, but because the loss became larger than the conviction behind it. The practical cost of concentration is rarely the drawdown itself. It is that deep drawdowns convert long-term holders into forced sellers at the worst available price, which permanently removes the recovery from their return.

There is a second, quieter cost. Recovering from a 20% loss requires a 25% gain, and recovering from a 40% loss requires roughly 67%. That asymmetry compounds in the wrong direction over a lifetime of saving, and it is why sizing discipline does more for terminal wealth than being early to a theme.

Measuring your own overlap

The useful work here is unglamorous and takes less time than reading one broker research note. Every fund you own publishes its full holdings list and its top-ten weights in official documents — the prospectus, the semi-annual and annual reports, and the factsheet, all of which are filed and searchable through the SEC’s EDGAR system for US-registered funds. That is the primary record. A fund’s own disclosure is the source of truth about what you own; a market commentary is not.

What you are looking for is not a sector label. Sector labels mislead here, because the companies driving an AI capex theme are classified across semiconductors, software, hardware, utilities, and industrials. You are looking for repeated company names across your holdings, and for the combined weight those names carry once you sum them across every account — taxable, retirement, workplace plan, and anything a robo-advisor is running on your behalf.

✅ Run four numbers before you add

If a pullback has you considering whether to buy more of a theme, average down, or trim, these four numbers decide the question more reliably than any view on the next earnings cycle.

• 1️⃣ Your true theme weight. Add every holding’s exposure to the same cluster of companies across all accounts, then divide by total portfolio value. Most people discover a number meaningfully higher than the one they assumed, because the index and sector sleeves double-count.
• 2️⃣ Your single-name maximum. Sum one company’s weight across every fund and every direct holding. A name that appears at modest weight in four funds can quietly become your largest single position.
• 3️⃣ Your drawdown tolerance in currency, not percent. Translate the relevant row of the table above into the actual amount of money you would see disappear. Percentages are abstractions; a specific sum is the thing you will or will not hold through.
• 4️⃣ Your income correlation. If your salary, bonus, equity compensation, or client base depends on the same technology spending cycle, your portfolio exposure is additive to your employment exposure. This is the factor most commonly left out entirely, and for people working in tech it is often the largest one.

Notice what this check does not require: a forecast. It does not ask whether AI capital expenditure decelerates next year. It asks whether your allocation survives the scenario in which it does — and if the answer is no, the fix is available now, at a price the market is still offering, rather than later under duress.

⏳ When the crowd is right

I owe the consensus its best case, because there is a real set of conditions under which buying into this weakness is the better decision and a diversification check still leaves you holding the exposure.

If your true theme weight comes in genuinely modest, if your income is uncorrelated with technology spending, if your horizon is long enough that a multi-year repricing is tolerable, and if the capital expenditure cycle the MarketWatch report describes investors as losing patience with turns out to be durable rather than front-loaded, then the sizing check simply confirms you have room. Sentiment-driven selloffs in sectors with real revenue have at times been exactly what the dip-buyers claim. Nothing in the arithmetic above argues against owning the theme. It argues against owning it accidentally, at a size you never chose.

The distinction matters because the opposite error is also expensive. Investors who respond to every drawdown by de-risking into cash systematically underperform, and “I reduced concentration” is a comfortable cover story for having panicked. The check is a sizing instrument, not a sell signal.

The question I can’t answer for you

What would flip my judgement here is evidence about the spending itself rather than the price action. The source of the current fatigue, per that reporting, is impatience with heavy AI outlays — which means the decisive disclosures are forward capital expenditure guidance in the operators’ own filings and earnings calls, not the direction of the sector index. If that guidance holds firm and order visibility extends, the pullback was sentiment and concentration will have been uncomfortable but survivable. If guidance is trimmed, the correlation you are carrying will reprice again, and the only variable you controlled was how much of it you held.

That variable is still yours this week. The weight is knowable, documented, and sitting in a filing you have not opened. Everything else about the next twelve months is a guess — including the confident ones.