news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Too long to read:
The U.S. stock market style changed this week, with the Nasdaq underperforming the S&P 500 and the Dow Jones Industrial Average, and technology heavyweights such as Nvidia and Meta fell.
The bears, represented by Goldman Sachs, are concerned about the high cost of large models and that AI has not yet had a “killer application,” which has affected the commercialization of AI technology.
Although there are differences in the realization of AI, there is a certain consensus that Nvidia is relatively invincible as a "shovel seller". In addition to Nvidia, judging from the recent stock price trend, Apple may become another winner by relying on hardware.

Nasdaq underperformed

The U.S. stock market has changed its style this week. As of July 19, the Dow Jones Industrial Average rose 1.7%, the S&P 500 fell 1.3%, and the Nasdaq fell 2.9%. Among the heavyweight stocks, Nvidia fell 6.3%, Amazon fell 5.5%, and Meta fell 4.6%.

Recently, Goldman Sachs' global head of equities also poured cold water on AI stocks, the main driver of this year's US stock market rally, with a report saying how painful it would be to go short against the ever-expanding tech bubble. The market always has a way of making money month after month, even if it's clear that the latest technological breakthroughs are not progressing as expected.

In his view, the hundreds of billions of dollars spent by companies in the field of AI will not trigger the next economic revolution, and will not even be as effective as smartphones and the Internet.

Driven by AI, the Nasdaq has been leading the market all the way. Looking back at the first half of this year, the S&P 500 closed up 14.5%, the Nasdaq rose 18%, and the Dow Jones Industrial Average performed slightly worse, closing up only 3.8%.The "Seven Sisters" contributed more than 60% of the S&P 500's returns, while the equally-weighted S&P 500 returned less than 5%.

But the situation changed suddenly in the second half of the year, and the debate about whether AI is a bubble has once again attracted attention.

Long and short divergence

The bears, represented by Goldman Sachs, are concerned about the high cost of large models, and that AI has no "killer application" so far, which has affected the commercialization of AI technology.

In this global AI big model arms race, technology giants including Meta and Microsoft have increased their bets. According to FactSet statistics, in the first quarter of this year, the total capital expenditure of Amazon, Google, Microsoft and Meta hit a new high of US$44 billion.

Microsoft's capital expenditures increased 80% year-on-year to $14 billion in the first quarter.Capital spending for the full fiscal year will increase by about 50% to more than $50 billion.Microsoft is seeking to triple its GPU supply to 1.8 million this year.

Google's capital expenditure in the first quarter increased by 91% year-on-year to US$12 billion, and is expected to be about US$50 billion for the full year.Year-on-year growth was approximately 55%.

Meta raised its full-year capital expenditure to $35 billion to $40 billion.33% YoY growth。

Amazon expects full-year capital expenditures to exceed $60 billion.At least 24% year-on-year growth.

Judging from the first quarter earnings, the revenue growth of large technology giants has indeed exceeded market expectations with the help of AI.Meta, which has been the most active in turning to AI, saw its revenue grow by 27%. Microsoft and Google also saw revenue growth of more than 15% in the first quarter. However, this growth is still far from enough compared to the expansion rate of AI-related costs.

The technological revolution brought about by AI is still in the early stages of infrastructure construction, and technology giants are all looking for monetization models with huge initial investment burdens.

Current ways of delivering results include Meta's reliance on AI to improve advertising effectiveness, cloud vendors increasing demand for cloud services based on users' use of AI technology, and Microsoft's subscription to Copilot for Windows.

The large-scale capital expenditure at this stage may take a long time to be monetized, which forces the market to start worrying about the company's performance prospects.

In fact, after experiencing the initial enthusiastic investment, the technology giants themselves may have begun to calm down and pay attention to costs.

Recently, OpenAI officially launched a new generation of entry-level artificial intelligence models with significantly reduced prices.GPT-4o mini, this model is more than 60% cheaper than GPT-3.5 Turbo, and expands the competition in the low-price market by combining "capability and cost-effectiveness".

Figure: In horizontal comparison, GPT-4o mini performs as well as many large models in terms of key indicators such as reasoning ability and speed, but its price is much lower than Llama3 and Gemini.

If a small model is sufficient to meet the needs of users for functions such as text reasoning and digital coding, how can the advantages of a large model be reflected? And how many high-performance chips purchased to support the large model will be wasted?

Figure: Meta leads the way among tech giants in the number of Nvidia chips used