ING: AI investment is still supportive and the market is beginning to shift from capital expenditure to verification of investment returns

ING: AI investment is still supportive and the market is beginning to shift from capital expenditure to verification of investment returns.

AI is still the main line of long-term growth for large technology companies, but the capital market’s pricing logic for this investment boom is changing. ING pointed out in its latest research report that investors do not deny the demand for AI, nor do they believe that technology giants have fallen into an unsustainable capital expenditure bubble.

On the contrary, the adoption rate of AI services is still increasing, the demand for computing power still significantly exceeds the supply, and most large technology companies can still rely on operating cash flow to support infrastructure construction.

What really triggers market volatility is when investors begin to reassess: when huge AI investments will translate into revenue, profit, and earnings per share growth, and how free cash flow, stock repurchases, and depreciation expenses will affect valuations in the process.

This means that AI trading is moving from “buying the future narrative” to the stage of “verifying return on investment”. In the past, the market was willing to pay high valuation multiples for AI-related companies, but now it pays more attention to capital expenditure efficiency, profit margin resilience and EPS growth quality.

AI infrastructure investment itself is not a problem ING first emphasized that current investment in AI infrastructure does not lack economic justification.

Taking Microsoft as an example, the company's investment in cloud and AI infrastructure reached US$65 billion in fiscal year 2025, and its annualized revenue run rate for AI-related services has reached US$37 billion. This shows that AI investment is already supported by relatively clear business needs, rather than simply relying on long-term imagination.

Technology companies also generally say that the deployment speed of AI computing power still cannot keep up with customer demand. In other words, the core contradiction in the current industry is still that AI computing demand exceeds supply, not oversupply.

For cloud computing, data centers, semiconductors and AI model service providers, this constitutes a basic reason to continue investing. What's more, most big tech companies' balance sheets remain solid.

Hyperscale cloud service providers such as Microsoft, Google parent company Alphabet, Amazon, and Meta have strong operating cash flow and can fund AI capital expenditures without significantly increasing debt.

ING believes that this is also an important difference between current AI investment and the traditional bubble period: industry leaders do not generally rely on highly leveraged financing for expansion. But the problem is that being able to invest does not mean that the market will continue to give high valuations.

The financial impact of AI investments is shifting from the balance sheet to the income statement and cash flow statement, and market focus is also changing accordingly.

Falling free cash flow affects repurchase ability The report mentioned that the performance of large technology stocks in the first half of 2026 was obviously divided: Microsoft's stock price fell by 20%, Oracle fell by 27%, and Alphabet rose by 14%.

This differentiation shows that investors are re-pricing based on the company's cash flow capabilities, capital expenditure pace and AI commercialization prospects. In the past few years, one of the important supports for the valuation of technology stocks has been strong EPS growth.

EPS comes not only from profit growth, but also from equity reduction due to share repurchases. Big tech companies in particular have long engaged in large-scale buybacks, spreading the same profits over fewer shares, thereby pushing up earnings per share. Alphabet is a typical example.

The report pointed out that from the end of 2020 to the end of 2025, Alphabet reduced its share capital by 9.5% through buybacks. This has provided significant support to EPS and helped the company maintain a high valuation.

(Stock repurchase data of the five major cloud giants in the past three years, orange is Microsoft, light blue is Oracle, burgundy is Amazon, light pink is Alphabet, and purple is Meta) However, as AI capital expenditures rise rapidly, free cash flow space may be compressed.

ING uses "EBITDA minus capital expenditures" to measure a company's cash flow margin after investment, pointing out that this indicator will significantly narrow for many large technology companies in the next few years.

That is, even if companies can still afford AI investments, they may have less cash available for buybacks, dividends or other shareholder returns. The impact on valuations cannot be ignored.

If repurchases decrease, EPS growth may slow down; if EPS growth slows down, the market will naturally be unwilling to continue paying the high P/E ratios that it has in the past.

The report pointed out that based on the price-to-earnings ratio of the past 12 months, Nasdaq is currently about 35 times, which is higher than 24 times at the end of 2022, but lower than the peak of 40 times in 2021 and 2024.

Looking at the two-year forward price-to-earnings ratio, Nasdaq is about 20 times, which is close to the low end of the historical range. So looking at forward valuations alone is not extreme. However, ING reminds that whether future valuations can be maintained depends on the quality of EPS growth.

If buybacks decrease, depreciation increases, and profit margins come under pressure, investors may no longer be willing to pay high valuations of around 40 times for technology stocks.

High capital expenditure will gradually translate into depreciation pressure The report specifically emphasizes that AI investment will eventually be converted into depreciation expenses.

The industry is currently in a stage of heavy capital expenditure, and capital expenditure is usually higher than depreciation; but in the future, as built assets enter the depreciation cycle, the pressure on the income statement will gradually appear.

AI infrastructure is particularly at risk because server hardware refresh cycles are typically shorter than traditional data center assets and depreciate at higher rates.

The report lists several sets of key data: According to consensus expectations, Alphabet's capital expenditures in fiscal 2026 account for about 44% of sales, and depreciation accounts for about 14% of sales; Microsoft's corresponding proportions are 35% and 14% respectively; Amazon's are 24% and 13%.

This means that margins may come under pressure if depreciation charges move toward higher capital expenditure levels in the future.

Therefore, investors are paying close attention to two questions: first, whether these AI investments can generate returns above the cost of capital; second, whether the revenue growth brought by AI is enough to cover depreciation, operating costs and ongoing hardware refresh expenses.

Company risk differentiation: Oracle, Nvidia, OpenAI, and Anthropic each have pressures ING believes that the credit status of most technology giants is still good, but the risks of individual companies are more prominent. The most notable exception is Oracle, which is involved in Project Stargate, a large U.S.

AI infrastructure project, but its internal cash generation capabilities are weaker than those of Microsoft and Alphabet. According to Refinitiv's consensus forecast, Oracle's future EBITDA minus capital expenditures may turn negative.

At the same time, Oracle's credit rating is already at a lower investment grade level, and its financing space is not as generous as that of larger technology companies. To maintain an investment-grade rating, the company may even need equity financing in the future.

This means that Oracle's stock price fluctuations not only come from uncertainty about AI returns, but also from capital structure and financing pressure. The problem facing Nvidia is different.

According to consensus expectations, Nvidia's operating profit margin in fiscal 2026 may exceed 60%, which reflects its strong demand for high-end AI chips and strong pricing power.

However, major customers such as Microsoft, Alphabet and Amazon are developing self-developed chips to improve capital expenditure efficiency and reduce dependence on external suppliers. If these cloud giants continue to increase their self-developed chip capabilities, NVIDIA's pricing power and ultra-high profit margins may face challenges.

As for the two major AI companies, OpenAI and Anthropic, the risk lies mainly in business model verification. The two companies are investing heavily in developing cutting-edge generative AI models and developing customized services around customer workflows.

OpenAI seems willing to endure greater cash consumption in exchange for industry dominance, similar to Netflix's early bets on streaming media and Amazon's early bets on e-commerce and cloud computing.

However, the long-term economics of generative AI are still not fully clear, especially the cost of model training and inference, customer willingness to pay, competitive barriers, and profit margins, which are all still in the verification stage. (

#Stocks #Nvidia #Microsoft #Meta #Amazon