No Data
Tokenomics in the U.S. and China: Sources of Profit, Premium Allocation, and Realization Sequence
Profits across the large AI model industry chain are being realized sequentially from upstream to midstream to downstream. The upstream computing infrastructure layer has already captured the full benefits of scarce resource premiums, while the midstream model layer is experiencing price deflation due to open-source competition leveling the playing field. In terms of where value accrues, there is a clear divergence between China and the United States: in the U.S., incremental AI value is being absorbed into existing high-priced software subscription models; in China, due to differences in willingness to pay, the low-cost token-based benefits are spilling directly into the downstream application layer. Going forward, whether downstream players can retain profits will depend on the 'switching costs' they establish within their specific business use cases.
Microsoft faces an AI compute shortfall—will Musk fill the gap? SemiAnalysis: SpaceX’s新增算力 in 2027 could exceed 10 GW, with ARR potentially reaching USD 300 billion.
After evaluating all pilot sites, SemiAnalysis considers SpaceX's goal of adding over 10 GW of new computing capacity by 2027 to be feasible. Microsoft, facing a substantial demand gap, is expected to sign a 3 GW contract with SpaceX worth approximately $150 billion, making it SpaceX’s largest computing capacity customer. If half of the newly added capacity is allocated to commercial inference, SpaceX’s annual recurring revenue (ARR) could reach $300 billion by the end of 2027.
"The new king" takes the throne with "three bold moves": Berkshire Hathaway has started spending, net purchasing $20 billion in stocks.
Since Greg Abel took the helm at Berkshire Hathaway, he has for the first time demonstrated a capital allocation approach markedly different from that of his predecessor. The cash-rich conglomerate, holding over $360 billion in cash, is accelerating its investments. In the second quarter, Berkshire net purchased approximately $20 billion in stocks, ending former CEO Warren Buffett’s streak of net stock sales that had lasted more than three years. Meanwhile, the company repurchased about $4.5 billion of its own shares and acquired homebuilder Taylor Morrison Home for $6.8 billion. Its cash reserves declined by roughly 4% from the previous quarter to $364.7 billion—the lowest level in four years.
Waiting days to access CPU capacity? AWS cracks down on computing waste, plans mass shutdown of virtual servers
① Amazon's AWS is conducting a thorough review of engineers' wasteful use of its internal computing resources; ② An employee disclosed that CPU server resource requests, which previously received approval within hours, now often take several days—a rare bottleneck that directly slows down project timelines; ③ Industry analysts attribute the current CPU resource constraints to the widespread adoption of AI agent applications.
AI Inference Outlook: From Infrastructure Development to Commercial Monetization
The investment cycle in AI infrastructure may not yet have peaked. Citi believes that supply-side constraints—including power, permitting, labor, HBM, and interconnects—will extend what was originally a three-year capacity expansion cycle to five to ten years.
Weekend Reading | A Review of the Trump Administration's Latest Seven AI Investments, Spanning CPO, Storage, and Semiconductor Equipment Across the Entire Supply Chain
The U.S. government is becoming a 'super venture capitalist' in AI infrastructure. Based on publicly disclosed transactions counted individually, the Trump administration has announced approximately 37 equity or quasi-equity arrangements over the past 14 months.