The AI cycle reaches "overtime" and four capital expenditure bubble warning signs worth paying attention to
This week, Samsung Electronics' second-quarter operating profit surged about 18 times year-on-year, hitting a record high, but its stock price closed down nearly 7% that day. Analysts believe that this shows that investors' focus has shifted from performance to whether huge investments in artificial intelligence (AI) infrastructure can generate returns quickly enough and how long this boom can last.
This week, Samsung Electronics' second-quarter operating profit surged about 18 times year-on-year, hitting a record high, but its stock price closed down nearly 7% that day. Analysts believe that this shows that investors' focus has shifted from performance to whether huge investments in artificial intelligence (AI) infrastructure can generate returns quickly enough and how long this boom can last. Is this reckoning near for the huge investment in AI? Sima Jing, chief China investment strategist of BCA Research, an independent investment research institution, said at a strategy meeting on July 10 that the AI bubble does exist, but not on the valuation side, but on the profit side. Regarding the current stage, she used the World Cup as an analogy to say that the current AI investment cycle is in "overtime" and has not yet reached the "penalty kick" stage. An earnings bubble, not a valuation bubble Sima Jing said that AI is not a bubble on the valuation side, but a bubble on the profit side. "Bubbles on the profit side are not uncommon in history. On the eve of the financial crisis, the profits of banking stocks, real estate and other industries were also very strong. However, once this strength loses sustainability, the market will re-price downwards." She took the semiconductor sector as an example and said that its "profit margins are very high and the price-to-earnings ratio is not high." However, she believes that the reason why the price-to-earnings ratio remains low is "precisely masked by unsustainable profit growth." What supports the judgment of "overtime" is the three levels of profit, demand and supply-side competition. The hidden worries at the profit level first come from depreciation. Market forecasts quoted by Sima Jing said that by 2030, several ultra-large-scale cloud vendors will hold a total of about 2.5 trillion US dollars in AI assets. Calculated at a depreciation rate of about 20%, the depreciation expenses at the end of this year alone will reach about 500 billion US dollars, and the total profits created by these manufacturers last year were about 400 billion US dollars. In other words, depreciation alone has exceeded their total profits, and this part of depreciation has not yet been fully included in the statements. Demand-level issues are hidden in a kind of "circular accounting". Sima Jing explained that when a company spends US$100 to purchase an AI chip, the money is an income to the seller, but is included as capital expenditures to the buyer and is recorded as an asset rather than an expense, so it is also reflected as a profit in the buyer's financial report. “Your capital expenditure becomes my cash flow,” she concluded, and both buyers and sellers count the same money as profit. At the same time, the free cash flow of these ultra-large-scale manufacturers has peaked in 2024 and will be close to zero by the end of this year. On the one hand, cash flow is approaching deficit, but on the other hand, profit margins are pushed to record highs. In her view, the entanglement in the market since June stems from this. Value is shifting from the input side of capital expenditure to the recipient side, and this dynamic is "clearly unsustainable." Competition on the supply side is heating up rapidly. In the face of strong demand, manufacturers in storage and other sectors are accelerating their expansion of production. Citing experience from the Internet era, Sima Jing said that the efficiency improvements brought about by technological iterations often absorb a considerable part of the incremental demand, that is, "less investment can be exchanged for more output." This means that capital expenditures in the AI field may not continue to expand as market expectations, and once the expanded production capacity of all parties catches up with demand, a decline in prices and profits will be inevitable. Sima Jing does not rule out the relatively optimistic scenario that these manufacturers begin to prove that their investment does pay off. In her view, this possibility is greater than manufacturers taking the initiative to lower capital expenditure guidance because returns are not as good as expected. But she emphasized that even so, from the perspective of growth rate, it is unlikely that AI investment growth next year will exceed this year, and the investment scale in 2028 will be almost the same as in 2027. “Market pricing often looks not at the absolute level of investment, but how much it has increased this year compared to last year,” she analyzed, and the market usually reacts before the funds are actually invested. Therefore, she judged that unless capital expenditures next year greatly exceed expectations and truly generate profits, "otherwise investors will still be worried."
、“”. “, ., . ”, “, 、. 、, . ”, , . GPU., , chips, . “, ”, “, , . ”, “”, GPU, “”. chips, ., AI chips, , , rise, rise . 、, : “、, , . ” AI., , . Token, 20%80%, , 1%AI7000 US dollar, 11 US dollar ., AI., , 、AI, 、, AI., “AI”, 、, , ., 、, . US stocks 30%～50%, 、 Fed 、, AI25, 、、 chips ., AI2000., 5%, , . affects ., 70 US dollar, tech stocks, (GDP)230%; , 13 US dollar, GDP130%. “, , ”, 1 US dollar ( fall ), 4., 20%, 3%, GDP2%.
Sima Jing reminded that this time the impact may be greater on the richest 1% of families. The market generally describes the United States as a "K-shaped economy", with the richest families at the top and about 90% of families living mainly on wages at the bottom. But "this time, the upper end of the K-shaped economy may move closer to the lower end, eventually changing from K-shaped to L-shaped." Based on the above factors, Sima Jing believes: "If it is the downturn caused by the bursting of the AI technology bubble, the decline of US stocks may not be 10% or 20%, but fall in the range of 30%-50%." She added that this prediction "may sound a bit alarmist", but "once the herd effect occurs, 30% is not a particularly extreme prediction." (
、“”. “, ., . ”, “, 、. 、, . ”, , . GPU., , chips, . “, ”, “, , . ”, “”, GPU, “”. chips, ., AI chips, , , rise, rise . 、, : “、, , . ” AI., , . Token, 20%80%, , 1%AI7000 US dollar, 11 US dollar ., AI., , 、AI, 、, AI., “AI”, 、, , ., 、, . US stocks 30%～50%, 、 Fed 、, AI25, 、、 chips ., AI2000., 5%, , . affects ., 70 US dollar, tech stocks, (GDP)230%; , 13 US dollar, GDP130%. “, , ”, 1 US dollar ( fall ), 4., 20%, 3%, GDP2%.
Sima Jing reminded that this time the impact may be greater on the richest 1% of families. The market generally describes the United States as a "K-shaped economy", with the richest families at the top and about 90% of families living mainly on wages at the bottom. But "this time, the upper end of the K-shaped economy may move closer to the lower end, eventually changing from K-shaped to L-shaped." Based on the above factors, Sima Jing believes: "If it is the downturn caused by the bursting of the AI technology bubble, the decline of US stocks may not be 10% or 20%, but fall in the range of 30%-50%." She added that this prediction "may sound a bit alarmist", but "once the herd effect occurs, 30% is not a particularly extreme prediction." (