Blackstone quietly exits; Amazon bonds cool down; is AI about to recede?

In late June, Blackstone sold its stake in three data center sites in Northern Virginia for $3.5 billion ($1.2 billion in cash plus $2.3 billion in Digital Realty stock). Last week, Blackstone's QTS withdrew its final appeal to the Virginia Supreme Court, officially abandoning the "Prince William Digital Gateway," the world's largest data center campus. While the official reason given was that a residents' lawsuit dragged the project down, adults investing should never listen to what others say, but rather look at what they do. Before the 2007 financial crisis, Blackstone also accurately sold off many then-hot commercial real estate properties. Earlier today, Amazon issued $25 billion in bonds to continue building data centers, but the final subscription multiple was only 1.6 times, while the average subscription multiple for high-rated US bonds is 4 times. Although Amazon CEO Eric Garcetti has consistently maintained that the demand for AI computing power far exceeds the supply, and they will sell as much as they build. However, in order to sell these bonds, Amazon proactively promised underwriters that it would not issue any more dollar-denominated bonds this year. The lukewarm reception wasn't limited to Amazon; SpaceX's first bond issuance last month saw a significant weakening in the secondary market, surprising bond traders. If the bond market cools down first, then capital expenditures by US tech giants will peak this year and begin to decline next year. In the bond market, which is dominated by institutional investors, none of these fund managers publicly said, "I'm not optimistic about AI." They simply quietly withdrew their orders... Not only are the financial markets unable to function, but AI projects are also scaling back. Bloomberg reported earlier this month that Meta plans to sell its idle AI computing power. Combined with Musk's SpaceX, among the "Seven Sisters" of US stocks, Microsoft, Amazon, and Google are already selling computing power, with Nvidia being the leading player. Only Apple hasn't participated. Given that the entire market is dominated by sellers, how much purchasing power is left to support the assumption of "computing power scarcity"? Interestingly, according to a report by CICC, this wave of AI computing infrastructure construction is largely self-indulgent in the US, with neither China nor other countries following suit. Moreover, without significant investment, China is closely following the US in both large-scale models and hardware, threatening its dominant position.

Tether's CEO also made a public statement about Fud AI, listing four major accusations:

1. Circularly releasing funds to generate revenue

Giants like Nvidia and Microsoft are investing in AI companies like OpenAI and Anthropic, while simultaneously demanding that they use that money to buy their own chips and cloud services.

2. Chips bought with borrowed money become obsolete before the debt is even repaid. Most of these data centers were built with debt, while GPUs have a lifespan of only 3 years. When the debts incurred between 2023 and 2025 mature in 2026 and 2027, the collateral (old GPUs) will be worthless. 3. Three historical pitfalls, and this time AI infrastructure has fallen into them all at once. The 2000 dot-com bubble saw manufacturers lending money to customers to buy their own equipment; the 2008 financial crisis saw off-balance-sheet leverage and unclear financial situations; and the 2015 oil price collapse led to the construction of excessive production capacity that no one wanted. 4. Open source is the detonator that ignites the explosion. If Chinese open source models like DeepSeek and Qwen become stronger, large companies won't be able to sell tokens at high prices. Therefore, the US closed-source model must crush the Chinese open source model, or it will be finished. The figures presented by JPMorgan Chase's chief strategist, Cembalest, in his 2026 outlook are alarming: since the launch of ChatGPT, 65% to 75% of the S&P 500's returns, profits, and capital expenditures have come from 42 AI-related companies. If these 42 stocks are removed, US stocks have underperformed Europe, Japan, and China over the past three years.

Even more striking is the macro perspective: In the past three quarters, technology capital expenditure has contributed 40% to 45% to US GDP growth, compared to less than 5% three years ago.

What's even more striking is the macro perspective: In the past three quarters, technology capital expenditure has contributed 40% to 45% to US GDP growth, compared to less than 5% three years ago.