The real risk of Amazon: is the capacity utilization rate in 2028. If the growth curve of AI demand even shows up for six months of pause, the fixed cost of 220 billion will instantly go from vision to overinvestment. Currently, the market has chosen Narrative to trust Andy Jassy.
The correct attitude against $SK海力士(SKHY)$ [傲娇] The problem with the South Korean government is not that it allows such products, but rather that it is a dilemma. The South Korean market originally had no leveraged products. Seeing that Hong Kong Korean stocks attracted a lot of money, the government believed that the fat should not go to other people's land, so it allowed the local market to launch in the name of "protecting investors". Individual investors are investing wildly, and the government is struggling to get out. When the stock price of Hynix quickly returned to high levels, the government said it needed to regulate and restrict, so the market panicked and retail investors stepped on people. Some analysts believe
The biggest troubling of Tesla isn't the short-term free cash flow (FCF) turning negative—the market actually expects to burn $3.25 billion, but actually only $10.9 billion is a "handsome number"; the real chronic poisoning is that the gross margin of the car business has dropped to 16.3%, closer to the level of traditional car manufacturers BYD and Toyota.
What does it mean to turn negative free cash flow? These companies will have to start borrowing heavily, issuing corporate bonds, and even increase capital to support AI infrastructure investment. Credit risk will gradually increase. Alphabet has already led this—almost $50 billion in equity financing plus $100 billion in debt raising. When the arms race of AI forced the world's most profitable group of companies from cash cow to borrowers, the whole valuation framework of technology stocks was being rewritten.
Musk's statement at the performance meeting was filled with his usual all-in gambler style: "This year is a year of super large capital expenditure. I am confident that everything we invest in will yield incredible returns. Perhaps the best return on capital expenditure we have ever seen. But the ruling given by the market was cruel: share prices plummeted 14.5% in a single day, evaporating about $200 billion in market capitalization.
Tesla Q2 recorded revenue of $28.5 billion and delivered a record high of 480,000 vehicles. But the adjusted EPS was only 0.33 dollars, significantly lower than Wall Street's expectations of 0.50 to 0.53 dollars. Operating expenses soared 47% to $4.35 billion, and operating profit margin plummeted to 1.4% from 4.1% in the same period last year. Capital spending surged 142% to $5.79 billion, and CFO Vaibhav Taneja confirmed that capital spending for the whole year would exceed $25 billion.
Alphabet Q2 revenue increased by 24% to $119.8 billion annually, and operating revenue increased by 30% to $40.8 billion. But free cash flow fell to minus $5.9 billion, because capital expenditure reached a record $44.9 billion.
To raise funds, Alphabet issued Class A and C shares and Mandatory Convertible preferred shares in June, raising $49.6 billion net. There was also a rights issue plan of up to $40 billion. For the first time in history, the world's largest digital advertising monopoly has had to reach out to capital markets because it spent too much money.
ASML Q2 system equipment sales reached 6.56 billion euros, with a total of 91 photolithography machines sold (86 new products, 5 second-hand). By application end, logic chips account for 51% and memory account for 49%, indicating that the demand for HBM on AI servers is crazily devouring production capacity
ASML CEO Christophe Fouquet's comments at the performance meeting were highly offensive: "Memory customers are stepping up their bookmaking production, and there is an extreme shortage of DDR and HBM, driving up 75% of memory revenue this year."
Starbucks announced that it would use AI-developed internal software to replace IBM's device maintenance management system, cutting its technical budget by about $30 million annually. Although Starbucks tried to build its own AI software, it recently was forced to cancel the project and resume manual inventory due to inaccurate inventory tracking system. This fully proves that non-tech companies have very little fault tolerance for trying to "full self-developed AI replacement", and that the enterprise-level services of traditional IT giants are not as fragile as the market expects.
IBM is actually the "macroscopic barometer" under the AI hardware frenzy. After companies buy servers and memory, they eventually need to integrate AI into their core business. The pre-front of hardware expenditure will eventually translate into deferred requirements for future software and service upgrades.
The moat is being rewritten by the underlying code. In February Anthropic claimed that Claude Code could modernize the ancient COBOL language, which directly threatened the survival foundation of IBM mainframes. Banks and insurers around the world rely heavily on hosts for mega-transactions, and if AI can easily refactor code, the cost of conversion for companies getting out of IBM systems will be dramatically reduced. The budget exclusion faced by AI hardware will persist long as the arms race does not end. $IBM(IBM)$