AI correction or warning sign? What investors can learn from the latest tech sell-off

Sharp declines in AI-related stocks have revived comparisons with the dot-com era. But beneath the volatility, the market may be telling a more nuanced story about valuations, investor psychology and the importance of portfolio construction.

The recent sell-off in technology and artificial-intelligence-related stocks has been dramatic enough to revive an old question: is the AI boom beginning to unravel?

For investors watching some technology shares lose 20%, 30% or more over a short period, the temptation to draw that conclusion is understandable. Yet the broader market picture suggests something more complex. Rather than a indiscriminate flight from risk, recent market moves have looked increasingly like a rotation. Capital has moved away from some of the most aggressively valued AI and technology names and towards value, defensive and more traditional sectors. That distinction matters.

A correction concentrated in one part of the market sends a very different signal from a broad-based collapse in which investors sell almost everything.

Political events can create volatility not only because of their economic consequences, but also because markets dislike uncertainty. Around the US midterm elections, equity futures initially reacted sharply. Once the result became clearer, however, markets stabilised and recovered. This illustrates an important principle: financial markets can often digest disappointing news more easily than uncertainty about what might happen next. Once investors have a clearer framework for evaluating policy, regulation and economic consequences, risk can be repriced. In the technology sector, however, political uncertainty arrived at a particularly sensitive moment.

AI-related stocks had already experienced substantial gains, valuations had expanded and expectations were extremely high. Concerns around future regulation, data-centre investment and the sustainability of AI spending therefore acted as a catalyst for profit-taking. The political event may not have created the vulnerability. It may simply have exposed it.

Large market moves rarely affect investor psychology in isolation. Several behavioural biases can reinforce one another.

One of the most important is recency bias.

Investors tend to give disproportionate weight to what has happened most recently. When markets fall sharply, the current correction can suddenly appear more important or more dangerous than previous episodes.

Past recoveries, earlier corrections and the longer-term investment thesis can quickly fade from view.

A second problem is information overload.

During volatile markets, investors often respond by consuming more information: news, social media, analyst opinions, market commentary and forecasts.

More information does not necessarily lead to better decisions.

In fact, it can make decision-making harder if much of that information has little relevance to the investor’s original investment thesis or time horizon.

Then comes confirmation bias.

Once investors begin to believe that AI regulation, excessive valuations or slowing investment will damage the sector, they naturally tend to search for information supporting that conclusion.

These behaviours can reinforce one another:

recent losses → more information → stronger conviction → emotional decisions.

Understanding that cycle can be just as important as understanding valuation multiples.

One of the most interesting features of the latest correction is what did not happen. The entire market did not collapse. Instead, the selling was heavily concentrated in technology and AI-related areas, while other parts of the equity market performed significantly better. That suggests sector rotation rather than wholesale capitulation. Money moving from highly valued growth stocks into value and defensive companies is very different from money leaving equities altogether. This is important because healthy markets frequently rotate. Leadership changes. Expensive sectors consolidate. Previously neglected companies attract capital. A market in which every asset rises at the same time can sometimes be less healthy than one in which investors discriminate between sectors and valuations.

Another lesson from the correction is that investors should be careful about treating “AI” as a single sector. The AI ecosystem contains very different businesses. There are semiconductor manufacturers, cloud infrastructure providers, data-centre operators, software developers, hyperscalers and companies applying AI to existing industries. Their business models, valuations, capital requirements and profitability can differ substantially. Even within semiconductors, some companies have delivered strong order growth and earnings while their share prices have still been caught in broader selling. That disconnect between fundamentals and short-term price action is worth watching. A sector can experience indiscriminate selling even when the underlying companies are not equally exposed to the risks that triggered the correction.

For investors, the question is therefore not simply:

“Do I believe in AI?”

A more useful question is:

“Which parts of the AI value chain have sustainable economics, and what valuation am I willing to pay for them?”

Whenever technology stocks fall sharply, comparisons with the late-1990s dot-com bubble tend to return.

There are similarities. Both periods involved extraordinary enthusiasm around a transformative technology, rapidly rising share prices and very optimistic expectations about future growth. But there are also important differences. Many of today's largest technology companies generate significant profits and cash flow. Their balance sheets are generally stronger, and substantial AI investment is being financed by some of the world's most profitable corporations. That differs considerably from the dot-com era, when many highly valued companies had weak revenues, little profitability and business models that had yet to prove themselves.

Valuations still matter. Strong companies can become overpriced, and even excellent earnings do not prevent substantial corrections. But comparing price performance alone risks ignoring major differences in profitability, financing and corporate maturity. The better comparison may therefore be less about whether AI equals dot-com and more about whether investors have priced future growth too aggressively.

Sharp corrections also reveal something about investors themselves. It is easy to believe that you have a high tolerance for risk when markets are rising.

It is much harder to discover that tolerance while watching a portfolio fall. This is why risk profiling should not focus only on the return an investor wants to achieve.

It should also consider:

1. Risk tolerance: how much volatility can you emotionally withstand?
2. Risk capacity: how much financial loss can you afford?
3. Risk need: how much risk do you actually need to take to reach your objectives?
4. Time horizon: how long can the capital remain invested?
5. Concentration: how dependent is the portfolio on one sector or theme?

Consider the difference between holding a modest allocation to AI-related stocks and building an entire portfolio around them. If AI represents 10% of a diversified portfolio and that allocation falls 30%, the direct impact is around 3% of the overall portfolio before considering movements elsewhere. If the portfolio is almost entirely concentrated in the same sector, the experience is completely different. Diversification does not eliminate losses. It can, however, make them manageable.

Fear of missing out is usually associated with buying after prices have risen. But the same psychological tension operates during corrections. An investor who is comfortable holding a defensive portfolio during a sell-off may later become frustrated when technology stocks rally much faster. That investor can then increase risk at exactly the wrong moment simply because other investors appear to be earning more. The objective of portfolio construction is therefore not to maximise participation in every rally. It is to build an allocation that an investor can realistically maintain through both rising and falling markets. The best portfolio on paper is of little use if its owner abandons it during the first major correction.

Volatile earnings seasons provide another useful reminder: risk management tools have limitations. A stop-loss order may help manage risk during normal market trading, but it cannot guarantee a specific exit price. If a company releases results after the market closes and its shares open sharply lower the next day, the price may gap below the stop level. A stock trading at 100 with a stop order at 90 could theoretically reopen at 75. The resulting execution would therefore occur around the available market price, not necessarily at 90. This does not make stop-loss orders useless. It simply means they should not be mistaken for complete protection against overnight gaps or event risk. Position sizing, diversification and portfolio construction remain essential.

Rather than focusing only on the size of a technology correction, investors can monitor whether weakness begins to spread. Several questions are particularly useful:

1. Is the sell-off still concentrated in AI and technology, or is market breadth deteriorating more broadly?
2. Are earnings expectations falling, or are share prices correcting despite resilient fundamentals?
3. Is capital rotating into other sectors, or leaving equities altogether?
4. Are credit markets beginning to show stress?
5. Are companies cutting AI investment, or does infrastructure spending remain strong?

The answers can help distinguish a sector-specific repricing from a more systemic deterioration in market conditions.

The first stage of the AI investment cycle has focused heavily on semiconductors, computing infrastructure, cloud capacity and data centres. But the next stage could look very different. If AI delivers the productivity gains companies expect, some of the biggest beneficiaries may eventually be businesses that use artificial intelligence rather than those building the infrastructure. Insurance, healthcare, defence, travel, financial services, retail, logistics and e-commerce could all potentially benefit from automation, better data processing and lower operating costs. That broadens the investment question considerably. The long-term AI opportunity may not ultimately be defined only by who produces the most advanced chip. It may also depend on which companies can turn AI into higher productivity, better margins and sustainable competitive advantages.

Is the AI boom over after the recent sell-off?

Not necessarily. A sharp correction can reflect stretched valuations, profit-taking and changing expectations rather than the end of a long-term investment theme. The key is whether company fundamentals, earnings growth and AI-related investment continue to support the sector.

Is today’s AI market comparable to the dot-com bubble?

There are similarities, particularly around enthusiasm and high valuations, but there are also important differences. Many of today’s leading technology companies are highly profitable, generate strong cash flows and finance AI investment from much stronger balance sheets than many companies during the dot-com era.

How can investors manage the risks of investing in AI stocks?

Diversification, position sizing and realistic risk tolerance are essential. Rather than treating AI as a single trade, investors can distinguish between semiconductors, infrastructure, software and companies using AI to improve productivity, while avoiding excessive concentration in any one theme.

Der Inhalt dieses Artikels dient ausschliesslich Bildungs- und Marketingzwecken. Er stellt keine Anlageberatung oder Finanzempfehlungen dar.