In the semiconductor industry, presentation New Chinese Artificial Intelligence Models, which are alleged to be offer competitive performance with lower requirements in computing power and cost, has raised concerns as to whether the large capital expenditure of technological colossae in the US will be paid in time and at the expected rate.
This is one of the main catalysts that have fed in recent days a first move of investors from overpriced semiconductor shares and artificial intelligence to others, more traditional or undervalued branches.
Markets now price the possibility that future returns on huge American investments will be lower than expected, first of all because the appearance of cheaper and more efficient Chinese models suggests that AI software can be optimized rapidly by rapidly changing the landscape.
If less computing power is required for example, future demand for expensive AI chips may be slowed.
Such a scenario is expected to bring pressure on profit marginsAs strong price competition from China will probably force American Big Tech to reduce the prices of their own AI services, development that can launch and fears about «bubble» technology.
A treaty not easy for investors who are already concerned that American giants spend hundreds of billions on infrastructure, without even clear and immediate profitability.
But there are other factors that contribute to the recent shift of investors not exclusively due to China, but to a combination of macroeconomic factors.
In addition, after a huge rally of technological shares, it makes sense that several institutional portfolios should Locking profits, Moving on to portfolio restructuring aimed at moving to valuable sectors such as energy, banks, general interest and construction.
However, within the technology sector there is currently a serious rotation since there is now a widespread appreciation that the coalition of the seven technological giants – Amazon, Nvidia, Meta, Apple, Microsoft, Tesla and Alphabet – is no longer the only reference point for the placements in artificial intelligence, which Citigroup has recently pointed out and with extensive analysis.
Scott Chronert called on the investment audience to focus on a much wider set of businesses, which he calls «growth cluster», appreciating that it is more loyal to the protagonists of the new AI period.
What is the composition of this «growth cluster»?
An expanded investment basket that includes not only dominant technological groups but also the majority of companies involved in the construction and expansion of artificial intelligence infrastructures.
This expanded total represents over 50% of the total stock value of the S&P 500 index and produces nearly 48% of the index's total corporate profits.
The structure of AI Infrastructure
Investors no longer just look at the AI software, but thethe whole natural infrastructure chain needed to run these models.
Thus the centre of gravity of capital expenditure is shifted or rather better shared by chip designers such as Nvidia to network infrastructure, energy and heat management.
Starting with electricity and network, the basic assumption here is that artificial intelligence is energetic. Therefore, clean energy supply companies — including nuclear — and transformers are the new «must-have» institutional portfolios.
Thermal management is another interesting branch likely to emerge in the coming months. You see, as the new AI chips release huge heat, the water cooling systems for data centers are expected to have upward demand.
Data transfer speed among the chips is the next sector we expect the market to focus on, as it is just as important as the power of the chips themselves.
Through this perspective, fiber optic and switches companies – i.e. companies that have the equipment connecting computers, servers, etc. within a network –are directly favoured.
The making of data centers It's another big chapter. Companies holding the land, licenses and know-how to build infrastructure for data centers in time-record begin to come to the center.
So even if AI software becomes cheaper or if models require less computing power per question, the volume of data and the global adoption of AI increase exponentially. Data centers, energy and cooling will always be needed, regardless of whether China or the US win «Technological warfare».
Many of these industrial and energy companies have already started getting premium «Technological share» and evaluating the cost-performance relationship between physical infrastructure -Hardware/Infrastructure- and software -Software/AI Models- is expected to be the central field of controversy in the coming months on Wall Street.
The comparative advantages
Analysts now clearly separate the two categories, concluding that natural infrastructure offers much more «visible» and guaranteed ROI in the short term - this is the investment return rate that measures the profit or loss resulting from an investment in relation to its costs - while at the same time the AI software faces strong pressure on its profit margins.
Through this perspective the shareholder of natural infrastructure companies -data centers, energy, cooling- does not worry whether a particular AI model will succeed or not.
Companies providing water cooling systems or energy transformers for example sign multi-annual contracts and their profits are direct, as they are paid to build the infrastructure, regardless of the subsequent profitability of AI.
Another observation is that data centers and earth have residual value. Even if demand for AI slows down, these infrastructure can be used for traditional Cloud Computing.
Finally, the lack of electricity and delays in licenses give existing infrastructure companies enormous tariff power.
So the comparison with software shares and AI models is clearly in favour of infrastructure shares right now.
Although software has theoretically zero marginal reproduction costs – so high profit margins in scale – in practice faces the following paradox of competition: As OpenAI, Google, Anthropic, but also Chinese Baidu and Alibaba, constantly present new, cheaper models, costs per question are dramatically reduced. That compresses the margins.
Moreover, the industry always faces the risk of rapid depreciation. An AI model that cost $500 million to be trained, may be considered outdated in 6 months by a competitive open source model!
Also, while Big Tech spends billions on investments, traditional businesses such as banks and retail trade are slow to integrate AI in a way that produces measurable profits. The longer this delay, the longer the investments of Big Tech will be.
In summary, infrastructure has a much better revenue visibility based on signed contracts with Big Tech, low competitive risk and a long investment life cycle reaching all 30 years for energy/air infrastructure.
On the contrary, technological giants currently have low revenue visibility from investment AI, extreme competition especially from China which constantly displays new and cheaper models, while the life cycle of their investments is often so small that it barely touches 2 years before technological degradation.
Therefore, infrastructure begins to gain ground in the preference of investors in the light that at a time driven by «gold fever», it is safer to sell shovels – i.e. energy, data centers and cooling systems – than to look for gold, i.e. models AI etc.
Disclaimer of Liability: This material is provided for information purposes only. Under no circumstances should it be taken as a supply, advice or encouragement for the purchase or sale of the products mentioned. Although the information contained is based on sources considered reliable, no assurance is given that they are complete or accurate and should not be taken as such.
