《Chiu's View》Since Anthropic raised concerns about AI safety and called for slowing down AI model development half a month ago, market discussions on AI safety mechanisms have intensified. U.S. President Trump not only 'suddenly' called NVIDIA CEO Jensen Huang during an AI summit to express that AI development would not be slowed, but recently hosted a gathering of leading technology and AI industry executives to discuss AI development and safety issues. After the meeting, these tech company executives agreed to support rapid data center expansion and signed an agreement on AI safety, committing to voluntarily implement internal safety measures, including mutual monitoring and third-party risk assessments of systems. This industry self-regulation approach, rather than new regulatory policies, reflects the Trump administration's stance of not restricting AI development. However, the industry is now facing the next factor affecting the pace of AI development in the United States: energy.
NVIDIA CEO Jensen Huang and SpaceX founder Elon Musk reiterated the importance of energy during the launch event of the U.S. government's official AI service website. Huang believes that without massive power supply to support data centers, AI cannot scale. Musk stated directly that the industry requires substantial electricity, and sufficient power is essential to maintain a leading position in AI development. Oracle's issuance of a 'force majeure' notice regarding one of its under-construction data centers shows that power supply and state regulatory approvals are indeed insurmountable obstacles to data center expansion.
To accelerate power supply for data centers, facilities such as Microsoft, Google, Amazon, and xAI have already established on-site power generation facilities (behind-the-meter), installing natural gas turbine generators or fuel cells that do not require connection to the public grid. This allows data centers to quickly obtain power to drive servers and serve customers, accelerating the return on AI investments. Institutions predict that by 2027, U.S. data center power capacity demand will double compared to 2025, which is a positive signal for companies like Bloom Energy, GE Vernova, and Caterpillar that provide on-site power solutions. However, investors should also note risks related to production capacity expansion speed and local policies.
Short-term risks in the U.S. stock market, apart from persistently rising bond yields, include potential increased market volatility due to quarterly portfolio rebalancing by institutional investors. Citadel strategists estimate that funding levels for major U.S. pension funds have reached 112%, the highest since 2001, increasing incentives for pension funds to reallocate equity and bond assets. Goldman Sachs estimates that pension fund rebalancing alone at quarter-end could trigger over $30 billion in U.S. stock sell-offs. After digesting short-term volatility factors, with the S&P 500's forward P/E ratio at 19 times, close to the 18.8 times observed on the 2025 tariff 'Liberation Day', a significant rebound in U.S. stocks may occur before the midterm elections, especially during the peak earnings season for large tech companies in late October.
《Wendy Chiu, Senior Analyst,华盛 Securities》
*The author does not hold any of the aforementioned stocks
*Articles published in Economic Information, signed or unsigned, represent the authors' personal opinions and do not reflect the position of Economic Information. The role of Economic Information is to provide a free speech platform.
NVIDIA CEO Jensen Huang and SpaceX founder Elon Musk reiterated the importance of energy during the launch event of the U.S. government's official AI service website. Huang believes that without massive power supply to support data centers, AI cannot scale. Musk stated directly that the industry requires substantial electricity, and sufficient power is essential to maintain a leading position in AI development. Oracle's issuance of a 'force majeure' notice regarding one of its under-construction data centers shows that power supply and state regulatory approvals are indeed insurmountable obstacles to data center expansion.
To accelerate power supply for data centers, facilities such as Microsoft, Google, Amazon, and xAI have already established on-site power generation facilities (behind-the-meter), installing natural gas turbine generators or fuel cells that do not require connection to the public grid. This allows data centers to quickly obtain power to drive servers and serve customers, accelerating the return on AI investments. Institutions predict that by 2027, U.S. data center power capacity demand will double compared to 2025, which is a positive signal for companies like Bloom Energy, GE Vernova, and Caterpillar that provide on-site power solutions. However, investors should also note risks related to production capacity expansion speed and local policies.
Short-term risks in the U.S. stock market, apart from persistently rising bond yields, include potential increased market volatility due to quarterly portfolio rebalancing by institutional investors. Citadel strategists estimate that funding levels for major U.S. pension funds have reached 112%, the highest since 2001, increasing incentives for pension funds to reallocate equity and bond assets. Goldman Sachs estimates that pension fund rebalancing alone at quarter-end could trigger over $30 billion in U.S. stock sell-offs. After digesting short-term volatility factors, with the S&P 500's forward P/E ratio at 19 times, close to the 18.8 times observed on the 2025 tariff 'Liberation Day', a significant rebound in U.S. stocks may occur before the midterm elections, especially during the peak earnings season for large tech companies in late October.
《Wendy Chiu, Senior Analyst,华盛 Securities》
*The author does not hold any of the aforementioned stocks
*Articles published in Economic Information, signed or unsigned, represent the authors' personal opinions and do not reflect the position of Economic Information. The role of Economic Information is to provide a free speech platform.
