The Zhitong Finance App learned that global capital flow data and stock market trading conditions compiled by LSEG Lipper show that in the past two weeks, the global market has shown a pattern of “reducing total risk and finding local opportunities” in the stock market. Based on the closing price of August 28 to September 10, the Korean stock market benchmark stock index, the KOSPI index, which has the title of “AI computing power weather vane”, rose about 3.61%, and the two major stocks, Samsung Electronics and SK Hynix rose by about 4.67% and 12.10% respectively; the Philadelphia Semiconductor Index rose by about 1.26% during the same period, but first rose 3.37% on September 4, then fell 2.66% on September 10, showing that the global AI computing power theme and the rebound in the AI computing power industry chain are still actively pouring into Asian stock funds. and technology industry funds.
In line with these trends, the latest global equity fund redemption data compiled by LSEG Lipper shows that investors have drastically cut their exposure to US large-cap equity funds while continuing to buy European and Asian equity funds and global technology industry funds. The expansion of demand for cutting-edge high-performance AI computing power brought about by the introduction of OpenAI Astra and the RSI (recursive self-improvement) training paradigm beginning to dominate the profit opportunities brought about by the blowout expansion of AI computing power demand in the context of AI training are still receiving attention, but the market's tolerance for its valuation and financing costs is declining.
Wall Street financial giant Morgan Stanley's latest key trend and judgment around the OpenAI Astra model, which Nvidia CEO Hwang In-hoon called the “AGI Era,” is that the AI model's capabilities have greatly increased to make more workloads economically viable, thereby strengthening the supply constraints of AI computing power, data center power chains, carrier boards, and storage manufacturing. According to its scenario calculation, the power capacity corresponding to the computing power deployment of hyperscale cloud vendors will expand from about 35 gigawatts in 2025 to about 145 gigawatts in 2028, reaching about 4.1 times the original.
The GPT-6 Astra was released on September 3, 2026. The Astra release unprecedentedly boosted the market's buzz about the advent of the general artificial intelligence (AGI) era. In particular, the new “pay-for-performance” growth model is expected to bring stronger total computing power requirements, and the recent more direct evidence of computing power requirements comes from the AI R&D process itself — that is, the “recursive self-improvement (RSI)” development trajectory where AI began to “make AI.”
As the advanced and cutting-edge model led by Astra brings more and more strong demand for AI computing power, Morgan Stanley expects the data center comprehensive capital expenditure of the four largest supercloud computing and AI application vendors in North America to rise from US$917 billion in 2026 to US$1.47 trillion in 2027 and US$1.64 trillion in 2028. The deployment capacity is expected to expand from 35 gigawatts in 2025 to 145 gigawatts in 2028.
An important industrial signal brought by Astra is that more complex jobs are beginning to have commercial value for execution by AI. According to media reports on September 10, OpenAI launched ChatGPT products for the financial services industry, combining GPT-6 Astra with professional data sources to support research, financial modeling, and customer material production. As deduced from this, the growth variables of AI demand will be further expanded to the number of concurrent agents, task execution time, tool call frequency, and context scale: when the cost of completing a task decreases and the success rate increases, enterprises have reason to deploy more workflows. This opens up huge room for cloud-based AI inference computing power and AI-related high-performance storage requirements. It is also the latest basis for the market to re-evaluate the sustainability of AI infrastructure growth.
US stock funds lost a lot of blood, and technology funds bucked the trend and sucked in gold: global capital shrinks exposure, selected tracks
The latest fund redemption data shows that investors have drastically reduced their exposure to equity investment funds related to US large-cap stocks, while continuing to buy European and Asian equity funds and global technology industry funds. Value stocks, high-cash-flow stocks, and bonds have not been safe havens for general gains. In the same period, using the reweighted closing price of the corresponding index ETF as an observation, the IWD of the Russell 1000 Value Index fell by about 1.94%, COWZ, which focuses on high free cash flow yield stocks, fell by about 3.79%, and the AGG of the Bloomberg US Composite Bond Index fell by about 1.13%. During the same period, the S&P 500 ETF-SPY fell by about 1.50%.
The above data highlights that high cash flow can improve the ability of enterprises to cope with financing pressure, but it cannot eliminate the risk that the industry's cyclical nature is too strong and that the position structure and valuation will be lowered; the net purchase of bond funds does not mean that bond prices are rising at the same time. The current more clear allocation change is that capital increases selectivity and tends to control the overall interest rate sensitivity of equity and bond holdings portfolios.
According to LSEG Lipper data, for the week ending September 9, the net outflow of global equity funds was US$15.52 billion, the largest weekly outflow since March 18. The net outflow of US equity funds was 32.27 billion US dollars, which unexpectedly hit the highest since the net outflow of 52.45 billion US dollars for the week of December 17, 2025; European and Asian equity funds received net inflows of 11.16 billion US dollars and 3.03 billion US dollars respectively.
The scale of divestment in the US market exceeded the net outflow of global equity funds, reflecting a partial offset by net purchases from other regions. This reflects the fragmentation of regional allocations. It cannot be directly deduced that the same group of investors will transfer all of the proceeds from US assets to Europe or Asia; fund redemption data is also not equal to the cross-border capital flow of the entire stock market.
Within the US market, divestments were mainly concentrated in large equity funds: net weekly outflow of US$40.44 billion, a record; medium equity funds had net outflows of US$682 million, while multi-market capitalization funds and small-cap funds had net inflows of US$3.52 billion and US$274 million respectively. Industry selection coincided with overall divestment: the net inflow of global industry funds was US$2.92 billion, of which technology and finance attracted US$1.89 billion and US$1.25 billion respectively; the net inflow of US industry funds was US$1.46 billion, of which net inflow of technology was US$1.71 billion and net financial inflow was US$720 million.
The total net purchases of technology and finance surpassed the overall net inflow of industry funds, which largely meant that other industries had offsetting large-scale one-way divestments. As a result, capital shows the characteristics of reducing broad stock exposure and retaining opportunities in specific industries; it is worth noting that regions, industries, and market capitalization categories are different statistical dimensions of LSEG Lipper, and cannot simply be repeatedly added or subtracted.
Inflationary pressure rewrites funding points! Short-term debt builds a line of defense, and energy allocations are increased
The macroeconomic pressure driving this round of adjustments is still the energy supply shock and the consequences of interest rates: Brent crude broke through $100 per barrel on September 9 and hit $109.97 in the intraday session on September 11; after breaking through the July high of $93.50 during the week, WTI hit $104.46 on September 11, both of which hit four-month highs. The US PPI and CPI data released on September 10, as well as the previous rise in prices in Europe and Japan driven by energy inflation, further strengthened inflationary stickiness and concerns about interest rate hikes.
The most obvious change in bond allocation is that investors are more willing to take risks over shorter terms. Global bond funds had a net weekly inflow of US$8.95 billion, the lowest weekly inflow since July 29; of these, short-term bond funds attracted US$6.65 billion, making it the second largest weekly inflow in three months. Loan participation funds and government bond funds had net inflows of US$1.01 billion and US$743 million respectively, while corporate bond funds had net outflows of US$2.37 billion. US bond funds received net purchases for the 21st consecutive week, with inflows of US$6.56 billion in the week; short to medium term investment-grade bond funds attracted US$3.75 billion, the highest in nine weeks, and attracted US$2.78 billion to medium-term government bonds and US Treasury bond funds.
Overall, investors still need bond dividend income, but pay more attention to long-term control and credit quality; global corporate bond fund outflows and US short- and medium-term investment-grade fund inflows also reflect regional and maturity differences.
Money market funds have further revealed that “safe haven” is not a globally synchronized one-way transaction. Global money market funds had a net inflow of US$10.72 billion, receiving net purchases for the second week in a row; the net outflow of US money market funds was US$10.41 billion, with a net inflow of about US$48.76 billion in the previous week. As a result, large redemptions from US equity funds were not mechanically converted into net purchases of US money market funds in the same week.
The pace of redemption, use of funds, and regional distribution may affect the results. Existing data is insufficient to track the final destination of each redemption fund, but it can be confirmed that the rise in global liquidity preferences can occur simultaneously with the phased outflow of US cash funds.
The distribution of capital in commodities and emerging markets is equally selective. Global gold and precious metals funds ended eight consecutive weeks of net inflows, with a net outflow of $537 million for the week; energy funds attracted US$211 million. The increase in energy allocation is consistent with investors' logic of focusing on supply shocks and their potential profit beneficiaries, but it is impossible to assume that precious metal funds are directly converted to energy based on this. Data covering 28,984 funds also showed that emerging market equity funds ended eight consecutive weeks of net purchases and turned into a net outflow of US$1.56 billion; emerging market bond funds had net inflows for the sixth consecutive week, attracting $537 million in the week.