The recent performance of the largest US tech companies has raised concerns about the risk of a stock market bubble. Using a three-stage pricing model applied to the top ten US tech firms, this column finds that current market prices do not appear implausible given the underlying drivers of both recent and expected performance. The adoption of AI technologies will be a major engine of growth, which is already transforming markets and displacing jobs. However, risks will remain around the ability to identify reliable AI applications and the growing competition from abroad.
Innovations in artificial intelligence (AI) have fuelled substantial stock price gains for listed US technology companies. Nvidia, the most focused on AI technology, has been the first company to reach a $4 trillion capitalisation at the beginning of July, followed by Microsoft three weeks later. The top ten US tech companies (henceforth, Mag10) now account for one-third and one-sixth of total US and global stock market capitalisations, respectively (Figure 1).
Alongside rising concentration, the elevated and growing valuations of US companies (as measured by their price/earnings ratios; Figure 2) have been a top-of-mind concern for financial analysts, prompting them to debate whether the US stock market is currently experiencing a bubble like the dot-com era of the late 1990s (Marks 2025, Stacey and Novik 2025, Duguid et al. 2025).
In this column, we contribute to the debate by first presenting descriptive evidence on the valuations and fundamentals of the largest US tech companies. We then apply a simple three-stage pricing model to derive their market-implied expected profit growth rates. Finally, we discuss the plausibility of such expected growth rates.
Descriptive evidence
Price/earnings (P/E) ratios for the US stock market have been on an increasing trajectory for more than a decade, with a notable surge during the Covid period, and are indeed approaching the levels observed during the dot-com era (Figure 2). In contrast, the P/E ratios of the largest tech companies remain well below the dot-com extremes, also because stock prices have not skyrocketed as spectacularly as they did in the late 1990s (Figures 3 and 4).
By focusing on the Mag10, we find significant heterogeneity in their current and forward P/E ratios (between 20 and more than 600; Table 1). The smaller companies (Broadcom, Oracle, Palantir, and Tesla) are those that have the higher multiples. These are also the companies that have recently made – or are in the process of making – inroads into new, mostly AI-related businesses that have high growth potential and are expected to significantly change the level and composition of their earnings. 3 The larger companies, instead, have much lower multiples. This may reflect investors’ expectations that their profits are unlikely to experience spectacular growth rates, given their already huge size and reliance – at least in part – on fairly traditional and saturated markets. Finally, Google’s relatively low P/E ratio (around 20) seems to indicate that investors are not indiscriminately pouring money into AI companies: while Google has developed frontier and, in some cases, best-in-class AI models, it is not yet clear that its AI strategy is able to counter threats to its most lucrative business (Search), which might be displaced by the rapid change in users’ behaviour (LLM use instead of Internet navigation).
Table 1 Mag10 market capitalisation and price/earnings multiples
Source: LSEG.
Note: Average values for 1-13 August 2025.
This preliminary descriptive evidence suggests that investors are still significantly discriminating tech and AI-related firms by their idiosyncratic earnings prospects. Such discrimination tends to vanish during bubble episodes and other periods when sentiment and herd behaviour substantially shape valuations (Baker and Wurgler 2006, Chang et al. 2000).
Evidence from an equilibrium dividend discount model
A stock’s price should be equal to the present discounted value of its future dividends. Elevated P/E ratios may signal optimism about future profit and dividend growth, optimism that could prove either well-founded or misplaced, thereby exposing the market to the risk of abrupt corrections. In Albori et al. (forthcoming), we use an equilibrium dividend discount model (Molodovsky et al. 1965, Fuller and Hsia 1984, Sorensen and Williamson 1985) to compute the abnormal growth rates of earnings that would justify current equity valuations, that is, make them compatible with rational pricing in line with historical norms. Table 2 shows the model-implied growth rates, together with recent earnings growth rates and analysts’ expected growth rates, all inflation-adjusted.
Table 2 Mag10 recent, expected, and model-implied growth in earnings per share (percent)
Source: LSEG, our estimates.
Note: Real (inflation-adjusted) annual growth rates in earnings per share (EPS). For Tesla, historical EPS growth is over the past four years; for Palantir, the past year. Growth expected by I/B/E/S analysts refers to the average growth rate over a period of three to five years. Model-implied growth rates are derived from P/E ratios as recorded on 28 July 2025.
For the six largest companies in our sample, the average model-implied annual growth rate is 12%, with moderate dispersion. This indicates a significant deceleration from the 33% average growth observed over the past five years and is also more conservative than the 15% growth expected by analysts.
For the remaining four companies, the mean model-implied growth rate is 41%, which compares with 39% in the previous five years and 17% expected by analysts. As discussed in the previous section, these four companies are radically different from the other six, in that they are much smaller and have recently entered fast-growing AI-related markets. Their implied growth rates are compatible with scenarios – far from being guaranteed, but also not utterly unrealistic – in which they manage to scale up their new business lines at a pace similar to that kept in recent years.
In summary, current equity valuations for the top US tech companies are rational if one expects high, but not historically unusual growth rates for the coming years. Apart from historical comparisons, determining if these expected growth rates are realistic is challenging. This assessment relies heavily on subjective judgments about market developments, particularly the future demand for AI-related services and improvements in productivity and profitability driven by AI adoption. In the following section, we present information that may help inform these judgments.
Factors affecting future earnings growth
The main factors that contributed to the spectacular earnings growth of US tech companies in recent years (Figures 5 and 6) were:
- Strong increases in global spending on digital advertisements (by 15% annually on average between 2019 and 2024; Kemp 2025), often driven by the displacement of traditional advertisements (e.g. on TV and newspapers) and by the rise of e-commerce at the expense of brick-and-mortar stores. Amazon, Google, and Meta are market leaders in this space and have continued to expand their market shares (in 2024, they accounted for 51% of global ad sales; BestMediaInfo Bureau 2025).
- Increased corporate spending on both traditional and AI-related cloud services (up by 21% annually on average between 2019 and 2024; Slingerland 2025). Amazon, Google, and Microsoft are the main providers of these services, with a combined global market share of 63% (Brindley and Zhang 2025). Significant fractions of the hardware bought by these players to scale up their data centres is provided by NVIDIA and Broadcom.
- Still rapid global expansion of the number of smartphones (+5% average yearly growth between 2019 and 2024), coupled with the continued increase in the amount of time spent by users in the interaction with these devices (Howarth 2025). This has sustained the sales of phones (Apple, Google) and consumer-oriented cloud services (Apple, Google, Microsoft), and has helped fuel the previous two trends (spending on digital ads and cloud infrastructure).
- Gains in efficiency driven by economies of scale and AI adoption. For example, software development represents a significant fraction of the cost of offering the products and services mentioned above, but its cost tends to increase less than proportionally as the user base expands (the same piece of software serves more users). Moreover, AI has already facilitated the automation of labour-intensive tasks, such as content-moderation on social platforms (Meta). Reportedly, AI has also improved the economics of digital advertising, by allowing for a better targeting of users.
According to the latest quarterly reports of the Mag10 (which on average showed still sustained growth), as well as to the expert sources cited above, none of these trends is expected to end abruptly in the near future, although some of them are expected to slow down. For example, smartphone adoption is expected to be sustained mostly by emerging markets, as advanced ones have already been saturated (Bellan 2025). Similarly, future growth in digital-advertisement sales may be hindered by the already large market shares held by the Mag10, although also in this case emerging markets represent a significant expansion opportunity.
The growth prospects of cloud services (and related hardware) are probably the most controversial ones. While some analysts expect AI adoption to keep propping up cloud spending for the foreseeable future (Goldman Sachs 2024), others deem that current AI spending by corporations is bubbly and likely to slow down, as the economics of AI applications are often unproven (Widder and Hicks 2025). Whatever the stance on this issue, most analysts seem to agree that AI spending is the key factor that will determine whether tech companies are able to achieve the high growth rates envisaged by financial analysts and embedded in stock prices (e.g. our model-implied estimates).
While taking a position in this debate is complex, we note that some recent trends seem to support the view that the demand for AI services will keep increasing at a brisk pace:
- As noted in a recent Economist article, the use of large language models (LLMs), such as ChatGPT, is quickly displacing Internet navigation (15% search traffic drop in the past year), as users find it more convenient to receive responses to their queries from an LLM rather than sifting through multiple websites (The Economist 2025a). This trend has just started, and it is likely to continue. The consequence is that a large portion of the revenues and profits previously made by millions of web-site owners and content creators will be shifted to cloud-service providers and hardware vendors (Godoy 2025, Weiss 2024), as on-the-fly content creation by LLMs requires enormous amounts of computational resources. At present, some of the major AI players still report difficulties in meeting these computational requirements, despite the huge investments in infrastructure made in previous years.
- Job destruction by generative AI technologies (not only LLMs, but also image and video generators) is reportedly gaining traction and economic significance not only in the website-ownership space, but also in adjacent occupations (ad creatives, illustrators, writers and copywriters, photographers, models, translators, moderators, etc.) (Bartholomew 2025). For example, according to a 2024 survey conducted by the Society of Authors (SOA Policy Team (2024), 26% of illustrators and 36% of translators report having already lost work due to generative AI, and three-quarters of them expect future income reductions. As explained in the previous point, this kind of displacement creates demand for the hardware and cloud services sold by the Mag10.
- AI technology is being increasingly adopted in the automotive sector (not only in autonomous vehicles, but also for driving assistance and infotainment). This trend – which will also significantly prop up the demand for cloud services, given the huge size of the automotive sector – has begun only recently, and it seems unlikely to abate, given that most auto-industry executives expect vehicles to become increasingly ‘software-defined’ (IBM 2025).
These are just some examples of large markets where substantial AI adoption and, in some cases, the consequent displacement of jobs are not just a future possibility, but something that is already happening at scale. Mechanically, this adoption generates demand for compute, which is mostly offered by large US tech firms.
Source : VOXeu
