30-day average bubble signal from news (-2 cooling ~ +2 heating) · from 2365 articles
Bellwether of the AI hardware cycle
MSFT + GOOG + AMZN + META + ORCL; cash PP&E plus new finance leases per SEC filings (Epoch AI)
Source: Epoch AI
Crunchbase: VC into AI startups, one quarter
Source: Crunchbase
Stanford HAI AI Index
Source: Stanford HAI (AI Index)
Stanford HAI AI Index — directly comparable with the China series
Source: Stanford HAI (AI Index)
Sequoia's '$600B question' series
Source: Sequoia Capital
According to Sohu, a company announced a $220 billion investment in AI compute infrastructure and predicted that the AI compute shortage will last until 2028. This massive investment highlights strong demand for AI infrastructure but also raises concerns about returns and bubble risk.
The founder of Firmus has gone from prisoner to AI tycoon, with the company valued at over $10.5 billion. This highlights the wealth-creating effect in AI, but also raises questions about valuation sustainability.
Hebei Province has recently conducted a large-scale cleanup of energy storage projects, with nearly 85 projects cancelled or delayed, reflecting a phenomenon of 'circling but not building' in the domestic energy storage industry, where projects become 'paper wealth'. This move is seen as a measure to squeeze the bubble in the industry, while overseas markets become an outlet for energy storage companies.
According to sources, Anthropic has decided to abandon its acquisition of AI startup Decart AI, despite prior due diligence. The two companies may still seek other collaboration opportunities. Representatives from both Anthropic and Decart declined to comment.
AI compute scheduling company Gimlet Labs announced a $300 million investment, valuing the company at $3 billion. This investment indicates continued capital market interest in startups in the AI infrastructure space.
The article argues that the 20% drop in domestic computing power stocks is a market overreaction, possibly due to excessive concerns about an AI bubble, while ignoring the actual demand and policy support for domestic computing power.