---
title: "The Second Half of the AI Investment Arena: When Liquidity Tightens Meets a Cash Flow Black Hole"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/topics/43102875.md"
description: "The Federal Reserve decided to keep the federal funds rate unchanged at 3.50%-3.75% at its July 2026 policy meeting, in line with market expectations. However, beneath the facade of holding benchmark rates steady, a hawkish policy stance may cause the market to abandon optimistic expectations for easing. During this voting session, three committee members explicitly voted in favor of a rate hike, and the widening divergence confirms that inflation stickiness far exceeds predictions; the new Fed Chair Walsh anchored the main theme of inflation control at the post-meeting press conference, reiterating that price declines have not yet reached policy targets and there is no intention to prematurely initiate monetary easing. Compared to whether to raise rates once..."
datetime: "2026-07-31T02:21:58.000Z"
locales:
  - [en](https://longbridge.com/en/topics/43102875.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/43102875.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/43102875.md)
author: "[财华社](https://longbridge.com/en/profiles/11651030.md)"
---

# The Second Half of the AI Investment Arena: When Liquidity Tightens Meets a Cash Flow Black Hole

The Federal Reserve decided to keep the federal funds rate unchanged at 3.50%-3.75% at its July 2026 monetary policy meeting, in line with market expectations. However, beneath the facade of holding benchmark rates steady, a hawkish policy stance may cause markets to abandon optimistic expectations for easing policies.

During this voting session, three committee members explicitly voted in favor of raising interest rates, and the widening divergence confirms that inflation stickiness far exceeds predictions; newly appointed Fed Chair Walsh anchored the main theme of inflation control at the post-meeting press conference, reiterating that price declines have not yet reached policy targets and there is no intention to prematurely initiate monetary easing. More than whether to raise rates once, what truly drives global capital pricing logic is its official interpretation of the rise in long-end U.S. Treasury yields: continuously rising long-term market interest rates are a direct reflection of liquidity contraction across all domains and rising borrowing costs, which may also imply a preference for regulating interest rates through the market.

It is worth noting that against the backdrop of fiscal deficit pressures and large tech giants raising capital through bond issuance to fund AI development capex, long-end interest rates may face upward pressure, meaning the overall trend of funding costs is rising. This market interest rate level also serves as a reference benchmark for asset valuation pricing. No wonder U.S. stocks, especially AI upstream tech stocks that had seen significant gains previously, plummeted after the release of the rate decision.

**Upstream Computing Power Sector Collapses Collectively: Is the Hardware Prosperity Feast Ending?**

The liquidity inflection point first ignited the AI upstream hardware track, which had seen the most surging gains previously.

Over the past two years, chip, memory, and optical communication manufacturers were the direct beneficiaries of the AI capital expenditure boom. Cloud providers continued to increase computing power purchases, driving the Philadelphia Semiconductor Index (SOX.US) to hit record highs one after another. Memory and computing power chip individual stocks saw multi-fold gains. However, over the past 20 trading days, the Philadelphia Semiconductor Index has cumulatively declined by 26.67%, significantly shrinking its year-to-date cumulative gains and officially falling into a technical correction zone. Global capital market computing power hardware targets have simultaneously begun a valuation digestion phase.

U.S. storage leaders suffered the most severe declines: SanDisk (SNDK.US), which surged due to AI server memory expansion demands, saw its stock price fall by more than 50% in less than a month, quickly squeezing out the valuation bubble from previous speculation on computing power and storage shortages. South Korean storage giant SK Hynix (SKHY.US) was also not spared; despite operating profits growing significantly driven by AI demand, its stock price still experienced a significant decline, with a cumulative drop of up to 23.28% over the last five trading days. Investors are no longer simply chasing order prosperity but are turning to worry about the sustainability of downstream cloud provider computing power purchases declining, leading to 萎缩 in storage demand and downward pressure on product prices.

The wave of adjustments swept through both the A-share and Hong Kong stock markets simultaneously.

Domestic AI hardware weight stocks generally welcomed deep corrections, with valuations of hardware targets that had surged previously based on computing power and server concepts continuing to be revised downward. CSIC Special Gas (688146.SH), a core supplier of AI chips and advanced process electronic special gases for HBM, had a cumulative gain of 873% over the last 250 days, but fell cumulatively by 22.02% in the last 20 trading days; Yuanjie Technology ($YUANJIE SEMICONDUCTOR(688498.SH)), a core supplier of upstream optical chips for 800G/1.6T high-speed optical modules, saw its stock price repeatedly hit new highs earlier, with a cumulative gain of 611.89% over the last 250 trading days, but it retreated by 37.96% in the last 20 trading days; Yangtze Optical Fibre and Cable (601869.SH), a core supplier of high-speed transmission fibers and hollow-core fibers for AI computing centers, saw its stock price accumulate gains of over five times in less than a year, but fell cumulatively by 43.44% in the last 20 trading days.

This trend also transmitted to the Hong Kong stock market. Zhongji Innolight ($ZJ INNOLIGHT(03308.HK)), a core asset in optical communications, broke below its issue price on its first day of listing on the Hong Kong Stock Exchange, never reaching the issue price level of 980 HKD throughout the day. Popular stocks listed synchronously from the A-share AI upstream track to the Hong Kong Stock Exchange recently, such as Naxin Integrated Circuit (02249.HK), an AI server power management chip wafer manufacturer; Dingtai High-Tech (01377.HK), a core consumable supplier of precision drill bits for high-end PCBs in AI servers; and Sanhuan Group (06951.HK), a supplier of MLCC ceramic substrates and optical module ceramic components for AI servers, all broke below their issue prices. This reflects the market's realization that the valuations assigned to these popular stocks based on the AI track were too high. Under the prospect of rising funding costs, capital is no longer blindly chasing the AI hardware track and is beginning to distinguish between real order demand and speculative themes.

However, the more core reason lies in the fact that the rise and fall of the upstream hardware sector has always depended on the capital expenditure plans of mid-stream tech giants. When the confidence of these giants to burn money for expansion gradually wanes, the foundation of demand for chips, storage, and optical modules naturally shakes.

**Insight into Giant Earnings: Capital Expenditure Sprint Devours Cash Flow, AI Burning Money Model Reveals Endogenous Contradictions**

Microsoft ($Microsoft(MSFT.US)), $Meta Platforms(META.US), and Google (GOOG.US), the three major North American tech giants, recently released quarterly financial reports ending June 2026, exposing the financial concerns behind massive AI investments: while revenue for all three companies continues to grow steadily, operating profits are continuously eroded by R&D and capital expenditures. The growth rate of capital expenditures is significantly higher than that of operating cash flow, with free cash flow shrinking dramatically or even turning negative. The once ample cash reservoir is under continuous pressure from endless computing infrastructure investments.

Meta's financial imbalance is highly representative. In the first half of 2026, its capital expenditure scale reached $50.918 billion, a significant increase of 65.84% compared to $30.704 billion in the same period last year. Just in Q2 alone, capital expenditures were as high as $31.078 billion, a year-on-year increase of 82.68%. Meta's Q2 revenue grew 27.96% year-on-year to $60.801 billion. The revenue growth rate was far slower than the pace of capital investment expansion. Coupled with a surge in AI large model R&D investment, quarterly R&D expenses increased by 67.33% year-on-year. The proportion of R&D expenses to revenue expanded from 27.24% to 35.62%, causing the operating profit margin to shrink from 43.02% to 30.88%, compressing profit margins continuously.

Pressure at the cash flow level is even more fatal: Meta's net cash inflow from operating activities in Q2 grew 24.65% year-on-year to $31.862 billion, but capital expenditures surged by 82.68% year-on-year to $31.078 billion, equivalent to 97.54% of net cash inflow from operations. That is to say, almost all the cash earned from operations was invested in computing infrastructure, leaving only $784 million in quarterly free cash flow, a year-on-year plunge of 90.83%.

As shown in the figure below, although Meta's operating cash flow maintained growth, the magnitude of its capital expenditure increase was larger, almost devouring the net cash inflow from operating activities.

Even with significantly shrunken cash flow, Meta still raised its full-year capital expenditure guidance, increasing the lower bound of the 2026 capital expenditure range from $125 billion to $130 billion, while keeping the upper bound unchanged at $145 billion, demonstrating its determination to continue increasing investment in AI computing power.

Microsoft mirrors Meta. Latest earnings reports show that the company's operating cash flow for the quarter ending June 2026 grew 30.00% year-on-year to $55.4 billion, but capital expenditures surged 69.42% year-on-year to $41 billion. The speed of capital investment growth was more than twice that of cash generation growth, with 当期 free cash flow declining 23.34% year-on-year to $19.6 billion. Microsoft management stated that over two-thirds of its capital expenditures were used for purchasing computing hardware such as CPUs and GPUs. Rising component prices further increased investment costs, and they expect next quarter's capital expenditures to exceed $50 billion. Management not only maintained the full-year (calendar year) 2026 capital expenditure expectation but also predicted that capital expenditures for the fiscal year ending June 2027 would continue to rise.

It can be foreseen that its long-term computing power investment plan will further exacerbate cash flow pressure.

Google's free cash flow turned from positive to negative. In Q2 2026, Google's net cash inflow from operating activities grew 40.80% year-on-year to $39.069 billion, while capital expenditures increased by more than double to $44.924 billion. The investment scale exceeded its own operating blood-making ability for the first time in recent years, resulting in a quarterly net free cash outflow of $5.855 billion, forming a sharp contrast with the $5.301 billion net cash inflow in the same period last year. Management revealed that 60% of its capital expenditures were directed towards server construction, with the remaining 40% used for data center and network equipment deployment, all serving the global AI computing power layout.

Insufficient own cash flow to cover capital expenditures may explain Google's recent move to rely on large-scale external financing: In June 2026, Google issued Class A and Class C shares and mandatory convertible preferred shares, raising a total of $49.6 billion for general corporate purposes, including scaled AI infrastructure and global computing power capital expenditures; combined with $20.3 billion raised through senior unsecured notes.

In addition, the company also signed an equity distribution agreement, selling up to $40 billion of the company's Class A and Class C shares from time to time through an At-The-Market (ATM) offering program. The raised funds are mainly used to fulfill tax obligations related to employee equity grants.

It is worth noting that to maintain leadership in the AI race, the Seven Giants continuously attract talent through such equity grants. For such talent, equity may be more imaginative than cash packages. Continuous equity issuance dilutes shareholder equity. Coupled with simultaneous bond issuance and share increases, the balance sheet pressure on tech enterprises is accumulating day by day.

Even so, Google still raised its full-year 2026 capital expenditure guidance from the previous $180-190 billion range to $195-205 billion, and predicted that 2027 capital expenditures would rise significantly.

Thus, it can be seen that the inertia of the tech giants' computing power arms race cannot stop abruptly, but the boundaries of financial pressure are approaching step by step.

**Doubts About Commercialization Returns 叠加 Regulatory Implementation, Will AI Track Valuation Face Systematic Revaluation?**

Cash flow pressure on mid-stream cloud providers transmits upwards to upstream hardware and downwards to large model startups. Coupled with the emergence of two major industry expectation inflection points, the market's overall optimistic sentiment towards the AI track has cooled down comprehensively.

On one hand, the global legislative process for AI safety regulation continues to accelerate. As problems such as data privacy, content ethics, and algorithm risks exposed by the iterative advancement of large model capabilities constantly emerge—for example, OpenAI recently autonomously breached isolation during security testing and invaded external servers, exposing the risk of AI losing control—countries may begin to 酝酿 implementing AI constraint regulations, drawing strict boundaries for large model training, commercial implementation, and data collection. The costs for large model enterprises to iterate R&D and promote commercialization may further rise, and the implementation pace might be forced to slow down, prolonging the monetization cycle that was once highly expected.

On the other hand, the core contradiction of the entire industry has still not been resolved: massive computing power investment has not yet delivered matching commercial returns. Giants spend hundreds of billions of dollars building computing clusters and training general large models, but the revenue volume of AI paid products facing enterprise and consumer ends is limited. Small and medium-sized large model manufacturers are deeply trapped in a money-burning dilemma. Without cloud ecosystem support and lacking continuous cash flow replenishment, their survival space shrinks drastically in an environment of tightening financing.

Valuation feedback is the most intuitive: Domestic top large model enterprises Zhipu (02513.HK) and MiniMax (00100.HK) saw their market capitalization retract by about 50% from highs in less than a month.

**Investment Insights for the Second Half: Return to Cash Flow and Commercial Essence?**

Looking at the entire process of this round of AI track from 狂欢 to adjustment, liquidity cycle changes and imbalance between capital input and output may be one of the core inducements. This adjustment does not mean the end of the AI industry super cycle, but rather an inevitable stage of industry de-bubbling and survival of the fittest, possibly providing certain insights for capital market participants:

1) Macro liquidity may become the anchor for the long-term pricing of AI assets. The high valuation cycle driven by liquidity may have ended. Under inflation outlooks, the high-interest-rate environment will not disappear in the short term. The rise in forward discount rates will continue to suppress the valuation of pure theme, heavy-investment, and cash-flow-less assets. Future investments must prioritize anchoring interest rate trends.

2) Shift in the value focus of the AI industry chain. The emotion of unilateral soaring valuations in the upstream hardware sector in June tells us that extremes reverse. Although the dividends of the upstream sector have not yet faded, excessively high valuations will not last forever; there will always be a day of returning to rationality. On the other hand, mid-stream tech leaders with cloud ecosystems and AI monetization capabilities, and downstream AI applications with clear landing scenarios, although value release still takes time, present valuation depressions under unilateral trends, bringing opportunities instead. Avoid blind following. Even the best enterprises have valuation ceilings; rationally assess risks and returns.

3) Capital expenditure is no longer a bonus item; Return on Invested Capital (ROIC) is the core yardstick for screening tech enterprises. In a market environment of tightening liquidity, capital will more strictly calculate the marginal return of every bit of computing power investment. Giants who can control the pace of capital expenditure, match computing supply and demand, and gradually achieve positive profitability in AI business will sooner or later receive capital attention.

4) Internal differentiation within the AI track will intensify extremely, raising the survival threshold for small and medium enterprises. The general large model track is already very crowded, filled with giants like Google, Alibaba (09988.HK), Tencent (00700.HK), ByteDance, OpenAI, etc. Small and medium startups need to discover their unique advantages, avoid head-on competition in general large models, and only then can they obtain continuous survival space. Purely financing-driven, technology-less, scenario-less startups relying solely on burning money will be gradually cleared by the market.

5) After valuations return to rationality, hardware leaders with hardcore self-developed barriers and stable cash flows may welcome medium-to-long-term allocation value. The sector crash brought about by short-term emotional 宣泄 digested the vast majority of bubbles. Those chip and optical communication enterprises with core technological barriers, bound to long-term stable computing orders, and strong bargaining power upstream and downstream, will highlight cost-effectiveness after valuation corrections. When giants' capital expenditure rhythms stabilize and AI application demand drives a recovery in real computing consumption, they are expected to embark on a new round of performance-driven rallies.

**Conclusion**

The industrial wave of the AI revolution has not stopped, but the capital 狂欢 may temporarily come to an end. In the second half of AI competition, the contest is no longer about who dares to burn more funds to pile up computing power, but who can 守住 financial boundaries, open up commercial closed loops, and realize the long-term value of technological change through continuous positive cash flow amidst the macro wave of tightening interest rates.

Author: Wu Yan

### Related Stocks

- [03308.HK](https://longbridge.com/en/quote/03308.HK.md)
- [300308.CN](https://longbridge.com/en/quote/300308.CN.md)
- [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md)
- [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md)
- [GGLS.US](https://longbridge.com/en/quote/GGLS.US.md)
- [GGLL.US](https://longbridge.com/en/quote/GGLL.US.md)
- [688498.CN](https://longbridge.com/en/quote/688498.CN.md)
- [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md)
- [META.US](https://longbridge.com/en/quote/META.US.md)
- [GOOGN.US](https://longbridge.com/en/quote/GOOGN.US.md)
- [SKHY.US](https://longbridge.com/en/quote/SKHY.US.md)
- [SKHYV.US](https://longbridge.com/en/quote/SKHYV.US.md)
- [688249.CN](https://longbridge.com/en/quote/688249.CN.md)
- [02249.HK](https://longbridge.com/en/quote/02249.HK.md)
- [01377.HK](https://longbridge.com/en/quote/01377.HK.md)
- [301377.CN](https://longbridge.com/en/quote/301377.CN.md)
- [300408.CN](https://longbridge.com/en/quote/300408.CN.md)
- [06951.HK](https://longbridge.com/en/quote/06951.HK.md)
- [SNDK.US](https://longbridge.com/en/quote/SNDK.US.md)
- [688146.CN](https://longbridge.com/en/quote/688146.CN.md)
- [06869.HK](https://longbridge.com/en/quote/06869.HK.md)