Barclays: "AI splurging" is a big factory "FOMO", some will retreat next year, but in the long term still in the early stage

The current artificial intelligence investment wave, the FOMO sentiment seems to be becoming the dominant force.

FOMO (Fear of Missing Out) describes a fear of missing out on investment opportunities, which can drive investors to make decisions without sufficient rational analysis.

Barclays' latest research report points out that although AI technology is still in its early stages, the capital expenditure of major companies on AI has shown an irrational prosperity, with FOMO sentiment dominating. As this sentiment fades, major companies will gradually reduce AI investments next year.

Barclays analysts such as Ross Sandler pointed out in a report released last week that FOMO sentiment is not uncommon in investment decisions, especially in rapidly developing technological fields. Investors fear missing the next major innovation, leading to substantial investments in emerging technologies without fully understanding their long-term potential. This sentiment was vividly demonstrated in the Internet bubble of 2000, and today in the field of AI, history may be repeating itself.

AI Capital Expenditure: A FOMO-Driven Frenzy?

Barclays' report mentions that there is a significant FOMO sentiment in the current market's capital expenditure forecasts for AI.

The report states that analysts expect cumulative capital expenditure in the AI field to reach $167 billion from 2023 to 2026, based on optimistic expectations for AI product demand.

However, in stark contrast, it is projected that by 2026, the incremental revenue from cloud services will only be $20 billion.

The significant gap between this capital expenditure and expected revenue has raised concerns in the market about overheated AI investments.

Barclays believes that while early AI companies may use user registration numbers and startup growth as evidence of demand, Wall Street's skepticism is rising.

This skepticism is based on a simple fact: in the past two years, there have only been two phenomenal AI products, ChatGPT and Github Copilot, but this is just the tip of the iceberg for large models. Currently, there are limited successful AI products in the market, and there is still a long way to go before widespread application of AI technology.

Future of Capital Expenditure: Some may retreat as early as next year

Barclays analysts expect that by 2026, AI capital expenditure ($167 billion) will be enough to support over 12,000 products the size of ChatGPT, with approximately $70 billion invested in training base models and the remaining $95 billion used for inference.

However, with technological advancements, the cost of AI inference may significantly decrease in the coming years, making the current capital expenditure forecasts appear overly optimistic Barclays stated that although it is expected that many new AI-based services will emerge and drive positive development in the market and industry, they are skeptical about whether there will be 12,000 AI products of the scale of ChatGPT in the market.

Based on the above reasons, Barclays wrote:

We believe that by 2025 or later, major players will retreat and cut (AI) capital expenditure plans.

We also note that breakthroughs in the field of smaller base models in the near term may shift a large number of products and queries from the cloud to edge technology (i.e., locally running on PCs or mobile phones) by 2026, which may further increase the demand pressure on the cloud for such large-scale AI capacity.

Long-term Perspective: AI Technology is Still in the Early Stage

Although the market may face adjustments in the short term, Barclays research report also emphasizes that from a long-term perspective, AI technology is still in the "super" early stage, and maintains an optimistic attitude towards NVIDIA in the medium term.

NVIDIA is unlikely to have any trouble in the medium term, as we expect AI capital expenditure to remain strong for several years before anyone retreats and cuts investment. The early stages of the new chip and technology cycle indicate that (tech giants) will continue to invest in AI in the foreseeable future.

It is worth noting that (the AI wave) is still in its early stages. Taking Apple as an example, it took nearly five years after the first appearance of the iPhone to have leading mobile native applications, and today, we are only 20 months into the AI wave.

Barclays also pointed out that the computing power required to run AI agents far exceeds today's software and internet services.

To illustrate this dynamic, Character.ai is a popular high-engagement AI company with only about 5 million users, roughly 1/600th of Google search, but recently stated that it processes nearly 1/5 of Google's daily request volume. This is because AI agent systems have no memory, need to cache interaction histories, and rerun them through large language models to make the next step or message accurate.

Barclays believes that the development potential of AI technology is enormous, but to realize this potential, time is needed to overcome technical challenges, market education, and the cultivation of user habits. In addition, the widespread application of AI technology also requires addressing a range of issues including data privacy, ethics, and regulation