The debate over whether US AI investments are rational or irrational hinges on balancing long-term technological advancement with pragmatic commercialization. As Broadcom plunges 12% in a single day, Oracle wipes out its annual gains, and NVIDIA faces mass sell-offs by major funds, Wall Street's skepticism shifts from questioning the existence of an AI bubble to predicting its collapse. This tech frenzy, ignited by ChatGPT, is undergoing a global reassessment after three years of capital exuberance.
### Structural Bubble The US AI bubble debate centers on the imbalance between high investment and low returns, manifesting in hardware, software, and applications.
**Hardware**: The "computing arms race" has led to runaway capital expenditures. NVIDIA, the linchpin of this race, enjoys near-monopoly status with its A100 and H100 GPUs, boasting a 210% YoY revenue surge in Q3 2025 and a 78% gross margin. However, its P/E ratio of 75x far exceeds the semiconductor industry average of 30x, and its $3 trillion valuation—equal to the combined market cap of the top 10 global semiconductor firms—relies heavily on sustained AI demand. Any slowdown could trigger a severe correction.
NVIDIA’s "ecosystem-driven growth" poses risks: clients like Microsoft and OpenAI prepay for chips, while NVIDIA reinvests in AI firms, creating a circular dependency. Yet, as global AI startup funding drops 32% in 2025, smaller clients delay orders, causing NVIDIA’s Q4 chip shipments to decline 15% sequentially.
Tech giants like Microsoft and Amazon plan to double capital expenditures to $470 billion by 2026, with 80% allocated to data centers and chips—60% of which flow to NVIDIA. If downstream applications underdeliver, upstream overcapacity could destabilize valuations. Oracle’s 136% capex hike to $50 billion (75% of revenue) has already pushed free cash flow to -$10 billion, while Broadcom’s $73 billion order backlog fails to shield it from stock declines amid weak AI revenue growth.
**Software**: Circular financing masks commercialization gaps. OpenAI’s $1.4 trillion spending plan projects losses until 2030, relying on NVIDIA and Oracle for capital and infrastructure. Valuations are stretched: Palantir trades at 180x P/E, Snowflake at 140x, and even Microsoft’s AI segments are priced far above traditional businesses. ChatGPT 5.2’s underwhelming performance exposed the limits of "compute-driven innovation," prompting a market reassessment.
**Applications**: Despite hype, scalable profitability remains elusive. Meta and Microsoft anticipate negative free cash flow by 2026, while enterprise adoption lags due to cost concerns and ethical debates. Q4 2025 saw Microsoft and Amazon cut AI server orders by 8% and 12%, signaling a shift from aggressive expansion to cautious procurement—a red flag for NVIDIA’s growth narrative.
### Real Value Labeling US AI as a "full bubble" oversimplifies. The Nasdaq’s 26x forward P/E pales against the 80x dot-com peak. NVIDIA and Alphabet’s dominance in AI chips and models reflects genuine technological moats. NVIDIA’s CUDA ecosystem and edge-computing bets, plus Alphabet’s TPU-powered AI pipeline, underscore strategic foresight. Government initiatives like the "Genesis Plan" highlight AI’s transformative potential in research and industry.
The bubble is *structural*—overheated infrastructure spending and inflated valuations coexist with legitimate innovation.
### Rationality vs. Overheating China’s AI landscape contrasts sharply: "rational but underheated." With 2025 capex at $40 billion (one-tenth of US peers), Chinese firms prioritize commercialization and domestic substitution. Alibaba’s GPUs run at full capacity, reflecting pragmatic demand. Local breakthroughs in specialized chips and models like DeepSeek demonstrate progress, though gaps in foundational research and high-end chips persist.
Yet, warning signs emerge: speculative startups and redundant local AI parks risk resource misallocation. Alibaba’s revised $380 billion AI infrastructure plan hints at underinvestment pressures.
### Divergent Paths The US’s "high-risk, high-reward" approach clashes with China’s caution. Both must balance innovation and sustainability. For the US, recalibrating spending toward efficiency and real-world applications is critical. China must boost core R&D while curbing local bubbles.
Globally, AI’s trajectory mirrors the dot-com era—after the bubble, true leaders emerge. Whether through US "correction" or China’s "catch-up," the winners will be those marrying innovation with viable business models.
In this pivotal moment, both nations must eschew "bubble anxiety" and "scale worship," focusing instead on long-term value creation. True revolutions stem not from capital euphoria, but from aligning innovation with market logic.
Comments