US stock market indices close with mixed results, large tech stocks generally down
The market opened with mixed red and green indicators, as technology stocks led the decline. The Nasdaq's nearly 1% drop looks jarring, but compared to these earnings results, what's more intriguing is the subtle shift in market sentiment. Corning fell 7%, Apple dropped over 3%, Tesla declined 3%... This is no longer a minor adjustment, but a collective pause and reassessment of the "everything is AI" narrative. Wall Street's gentlemen seem to suddenly be reevaluating: Are the astronomical sums
Analysis
Meanwhile, another piece of news appears especially "hardcore" and somewhat surreal. Broadcom, along with Apollo and Blackstone, has established a $35 billion AI computing platform called "AI XPV," aiming to deploy over 20 gigawatts of computing power for laboratories like Anthropic and OpenAI before 2028. What does this figure mean? It's close to the power generation capacity of two Three Gorges Dams. This is no longer just investment—it's stockpiling "strategic nuclear weapons" for a computing arms race.
On one side, the secondary market shows impatience and caution toward the profitability prospects of tech stocks. On the other, top private equity firms and chip giants are frantically betting billions on infrastructure. This very disconnect reveals the core contradiction in the AI industry today: the vast gap between expectations and reality, scale and efficiency, is being forcibly bridged by capital.
The initial $35 billion is targeting Anthropic's expansion plan of over 1 gigawatt. What does this imply? The "bottleneck" issue in AI has shifted entirely upward—from algorithms and data to the most fundamental layer: energy and computing infrastructure. Large model companies may verbally discuss the philosophy of AGI, but their actions speak louder: they are all stockpiling resources for the impending "computing drought." Broadcom provides customized XPU and networking solutions, while Apollo and Blackstone provide the firepower. This is essentially a vertical integration of the AI computing supply chain. Future competition may no longer be about whose model is smarter, but whose GPU clusters are larger, cheaper, and more stable.
This "infrastructure mania" style of investment, when you think about it calmly, carries enormous risk. Behind the $35 billion lie dozens of data centers—involving site selection, construction, operations, and the unimaginable costs of electricity and cooling. If AI application growth falls short of expectations, or if the efficiency of model evolution outpaces the growth in computing scale, then these pre-deployed gigawatts of computing power could easily become expensive "digital real estate." But capital is clearly betting on the opposite: that AI demand will be exponential, that computing power will never be enough, that the winner takes all, and that infrastructure is the final ticket.
This reminds one of the dot-com bubble era, when countless companies frantically laid fiber optics, ultimately resulting in "one mile laid, one mile bankrupt." But Broadcom and its peers clearly believe they are smarter than the telecom operators of that time: they aren't just providing generic pipelines, but highly optimized computing pipelines tailored specifically for AI. This is a bet on a higher dimension—not on applications, but on infrastructure.
Interestingly, this investment is directed at Anthropic and OpenAI. This means top AI laboratories are forging tighter alliances with top-tier financial capital and hardware giants. Once this "iron triangle" relationship solidifies, it will significantly raise the industry's entry barriers. Future large-model startups may need not only genius algorithm teams but also the "financial muscle" or "connections" to secure massive computing resources. The flame of innovation, in this case, risks being extinguished due to the monopoly of fuel—that is, computing power.
Returning to the U.S. stock market. The decline in tech stocks may not be entirely unhealthy. When Microsoft, Nvidia, and others start contemplating how to generate real revenue from AI—rather than just making promises—the market may offer more rational valuations. The real test lies ahead: When Anthropic and its peers possess 20 gigawatts of computing power, will they deliver applications compelling enough for the world to pay for and cover these astronomical costs? If not, today's $35 billion bet will ultimately become a tragic footnote in the epic of AI.
Capital never sleeps. It withdraws from crowded tech stocks while precisely directing ammunition toward the hardest infrastructure. This split feast tells us: The future of AI depends neither on model parameters in a presentation nor on revenue figures in an earnings report, but on who can truly transform electricity into intelligence, and then intelligence into profit. Until then, all prosperity and panic are merely preludes.
Disclaimer: The above content is generated by AI and is for reference only.