Many AI stocks performed very well in the first half, and we see double-digit return potential for the AI theme over the next 12 months. After the strong share price performance of many names, however, we believe investors now need to invest more selectively. We essentially see two trends moving markets: tailwinds from the rising spread of AI-based solutions and applications, and headwinds from the increasing risk of a capex taper tantrum. The term “capex taper tantrum” is used in reference to the term taper tantrum, which emerged in May 2013 when the Federal Reserve unexpectedly announced that it would scale back its multibillion-dollar bond purchases, causing nervousness and concern in markets.
Overall, the outlook remains solid, not least because the industry’s fundamental trends are robust. According to OpenRouter, weekly token consumption has already risen roughly eightfold this year. At the same time, according to METR analysis, the capabilities of AI systems are doubling about every four months. If this pace continues, leading AI labs could come close to recursive self-improvement within the next one to two years.
On valuations, too, the situation seems less dramatic to us than is often claimed, with some exceptions. In semiconductor stocks, for example, valuations do not look like a bubble. Forward price-earnings ratios are now around 26x—far from the roughly 150x seen at the peak of the dotcom cycle. How is this possible given the enormous gains of recent months? Quite simply, because earnings growth has so far largely kept pace with share price performance, and in some cases even exceeded it.
Nevertheless, the durability and scale of AI investment are likely to increasingly influence the industry outlook: While the rapid growth in new data center construction was the main driver of share prices over the past three years, bottlenecks and the resulting price increases are playing a larger role this year. This is unlikely to change much in the coming months: Memory prices could rise 30-40% quarter over quarter in 3Q26, again exceeding consensus expectations of 10-15%. With higher prices also likely in other areas, the cost of AI servers could continue to increase. We forecast AI capex of USD 820bn in 2026 and USD 990bn in 2027 and would not be surprised if the actual figures were once again higher. Overall, we therefore continue to expect high spending on AI infrastructure.
At the same time, however, the question is what can follow this investment cycle. We expect the hyperscalers’ operating cash flow to be overtaken by cash capex requirements as early as 3Q26. In other words, additional investments by these companies, and by many other smaller firms as well, would then increasingly have to be financed through debt and equity capital because they would no longer generate enough free funds to cover spending. This can create volatility, particularly during the summer months, when liquidity is generally somewhat lower.
At the same time, the recent weakness in hyperscaler stocks is likely to lead some of them to respond to pressure for greater capital discipline. Together with concerns about corporate spending on AI and token costs, this could weigh on valuation multiples for semiconductors and AI hardware. That is why, during the recent memory-driven rally at the end of June, we took some profits in semiconductor and hardware positions—especially in smaller and midsize companies whose valuations had become more demanding.
Despite this reduction, we remain overweight the AI segment by around 16 percentage points versus the Nasdaq 100. In other words, we remain constructive on semiconductors and AI hardware, but after the strong share price performance, we are becoming more selective. We have also shifted part of the proceeds into more defensive technology areas, such as payment networks and data center REITs.
Our conclusion: AI is likely to remain an attractive investment theme, but it should no longer be a position in which investors simply rely on momentum staying positive and passively follow the market. The gains are real, adoption is rising, and the long-term potential is significant. At the same time, however, calls for capital discipline and investor sensitivity are increasing.
In our view, these developments together argue for an active, disciplined investment approach with sufficient diversification. Investors should stay invested in the AI segment—but more selectively, on a broader basis, and with a clearer view of the challenges around the capex taper tantrum. It is also worth noting that we see numerous opportunities outside the AI area as well. With energy prices falling, the remaining stagflation fears should increasingly dissipate, creating room for the equity rally of the first half to broaden to other sectors.
