Recent extreme volatility in the gold and silver markets has triggered cross-market chain reactions, with a core catalyst being the distortion of assets' intrinsic logic. Due to excessively rapid price appreciation, highly concentrated trading, and an influx of leveraged capital previously active in cryptocurrencies, precious metals have transformed from traditional safe-haven assets into narrative-driven assets. This "Meme-ification" trend, causing severe fluctuations, has extended beyond precious metals, evolving into a global deleveraging chain reaction. Cross-asset selling pressure eventually circled back to US equities, forming a complete loop: capital withdrawal from overly crowded trend-following assets triggered a systemic adjustment from commodities to equities.

Global markets have now entered a new phase of fully unleashed volatility. Even after significant corrections, the implied volatility of gold and silver remains at historically extreme highs. Coupled with Bitcoin's 25% year-to-date decline, this intense two-way volatility signals a market shift from irrational exuberance to emotional博弈.

The narrative surrounding the US technology sector has deeply benefited over the past two to three years from the dual drivers of AI vision and abundant liquidity. However, as the phase where high valuations could be supported merely by capital expenditure scale has passed, the market is now facing an emotional backlash from prior over-optimism. Currently, tech stocks are in a period of pressure release following extreme crowding. Significant corrections in core sectors like software and hardware represent a sharp market correction to previous valuation premiums. Regarding the AI narrative itself, factors such as earnings pressure on tech stocks, interlinked dependency stories, and potential changes in the liquidity environment are being re-examined with a more sober perspective.

First, AI industry development and investment performance are two different matters. The AI industry continues to advance and deliver surprises, but the investment side may be overpriced, exhibiting excessive optimism about long-term returns. In reality, industrial progress is not always beneficial for investments. However, an excessively large investment bubble is not necessarily bad for the industry itself, as it can leave behind substantial cheap infrastructure. Often, genuine industrial technological advancement occurs after a bubble bursts. Therefore, it is advisable to separate industrial development from investment decisions. Between industrial logic and market narrative, the industry usually provides only a logical starting point. Once a logic forms and spreads, the market releases volatility through self-fulfilling mechanisms. Recent movements in software and hardware alone are insufficient to cause such dramatic market reactions; the shift is more attributable to changes in investment sentiment, liquidity, and narrative pace.

At the industry level, the recent focus has been on the series of applications released by Anthropic's Claude. Beyond the earlier ClaudeCode, the recently enhanced Cowork integrates models with office software, already capable of replacing entry-level office staff in vertical applications, posing a direct challenge to industries like software SaaS, legal, and financial data analysis. In fact, the software sector has underperformed hardware for two years, and the logic of large models replacing software has existed since GPT's launch. Although companies like Adobe have improved efficiency through AI, general-purpose large models demonstrate strong substitutability due to their programming, debugging, and programmer-replacement capabilities. The recent decline in software is more an excuse for the market to release pent-up high-volatility sentiment using industrial logic as a pretext.

Regarding hardware, the recent decline is more influenced by sentiment transmission. Companies like AMD and Qualcomm, due to unimpressive earnings performances, have seen amplified negative market sentiment. The real pressure on hardware stems from rising memory chip prices. Significant price hikes by Samsung and SK Hynix severely squeeze the profit margins of other hardware manufacturers. Phones, laptops, and cars are highly dependent on chips, and memory costs are fixed. If manufacturers cannot raise end-product prices amidst doubling memory costs, they must compress their own profits or hardware gross margins. This chain reaction extends from AI data center computing investment and HBM to price increases in other DDR memory, ultimately impacting sales of durable goods like phones, cars, and laptops.

This series of phenomena essentially reflects the US's K-shaped economy: the divergence between the AI sector and non-AI sectors is widening. The expansion of the AI industry creates a noticeable "crowding-out effect" on other industries, commandeering more financing, power, and storage resources. This resource competition from tech to non-tech sectors is a long-standing issue, and the underlying pressures on the US economic fundamentals will ultimately cap the sustained upward trajectory of the AI narrative.

Second, observations from the US Q4 earnings season show declining tolerance for capital expenditure, with liquidity becoming a core constraint. We are currently in the concentrated disclosure period for US Q4 and full-year reports. The weight of earnings signals in the market has significantly increased, with heightened focus on actual performance and rapidly declining tolerance for capital expenditure. This is a result of sustained "cash burn" over the past two to three years. Since 2023, the pace of capital expenditure by tech companies has accelerated noticeably for 2024 and 2025, with investment scales不断扩大. Whether liquidity can remain persistently loose has become an unavoidable question.

Two to three years ago, the return path for AI-related investments was still unclear, and the market was willing to grant time.但现在, the market's punishment mechanism for "over-investment" has significantly strengthened. The question of whether sufficiently large business models can emerge to cover both past and future expanding capital expenditures often becomes a market focal point when marginal liquidity changes occur. The market craves certainty of "persistent easing"; once this certainty is缺失, skepticism rapidly returns.

Amazon's Q4 report showed a full-year CAPEX plan as high as $200 billion, exceeding Google and Meta, leading to a roughly 13% single-day stock drop. Oracle has long been in a "cash-strapped" state, with negative free cash flow for three consecutive quarters. Whenever Oracle faces funding紧张, its CDS prices rise. Recently, Oracle announced plans to raise $45-50 billion in bonds within the year, a模式 similar to real estate firms or城投 platforms, requiring constant "refinancing old debt with new."一旦融资 channels are受阻, the market quickly punishes its valuation. Therefore, in a phase where marginal easing is questionable, Oracle is increasingly viewed as an indicator of denominator risk, with its CDS performance being a key metric to watch.

In terms of overall scale, the combined CAPEX of Amazon, Google, Microsoft, and Meta for 2026 is approximately $660 billion, a 60% surge from 2025 and more than double the 2024 expenditure. However, the combined revenue growth of these four companies in 2025 is only about 18%, and the vast剪刀差 between expenditure and revenue needs to be sustained. This is what truly frightens capital markets. The greater concern is that投入 is being提前透支 before returns materialize. Signs of software companies being "washed out" are emerging, and hardware faces similar issues; previously deployed hardware hasn't yielded truly high returns yet, with阶段性 profit反而 coming more from hardware price hikes squeezing profits in other industries, leading to accumulating fundamental worries.

Regarding liquidity, the current consensus overseas,围绕 the topic of Kevin Warsh's nomination, is that the rate-cutting path must continue, at least in the short term. Rate cuts are a prerequisite; the pace is the variable. As for quantitative tightening (QT), it depends more on the fiscal and political博弈 between the US economy itself and the Trump administration. Its logic isn't entirely determined by personnel changes and may not be highly correlated with Warsh personally.

The factor truly主导 liquidity changes remains inflation. Despite cumulative rate cuts of 175 basis points so far, the US Treasury yield structure shows significant divergence: short-end rates follow policy rates down, while long-end rates remain persistently high. In late January, the 10-year yield touched 4.3%, and the 30-year even approached 5%. From PPI to core CPI, from goods to services, US inflation levels have not substantially declined. Manufacturing industrial goods prices continue to face supply-side constraints, and AI-related investments further push up energy and basic raw material prices. On the CPI front, accumulated wealth effects support high-end consumption on one hand, while consumer goods prices face transmission mechanisms from tariffs to supply gaps on the other. These factors collectively cause inflation to remain sticky.

In this environment, whether it's a new chair or incumbent Chair Powell, the Fed is compelled by现实 conditions to maintain the general direction of rate cuts, with the only difference being the pace. Current inflation stickiness and rigidity persist, with the policy rate in the 3.5%–3.75% range. Objective conditions dictate that the maximum room for future rate cuts is only two or three more. Against this backdrop, US AI giants still need continuous financing for expansion, while secondary market patience is gradually wearing thin. Recent rises in US Treasury yields and a rebound in the dollar index, coupled with unclear marginal easing, make the capital expenditure scales revealed in large tech companies' earnings reports appear格外刺眼. The core question of "whether the math ultimately works out" has重新 surfaced.

Third, employment contraction and the K-shaped economy are accelerating internal differentiation within tech stocks. Macro employment data further intensifies market urgency and unease. With non-farm payroll data delayed due to the government shutdown, unofficial employment data becomes particularly crucial. However, the "small non-farm" ADP data was quite weak, showing only 22,000 new jobs in January, far below the expected 45,000. Excluding a surge of 74,000 hiring positions in the education and health services sectors, overall new employment would be negative. The AI penetration rate in health services is extremely low, so the new jobs primarily reflect non-AI sectors or employment挤出 by AI. Employment expansion in essential sectors like education and healthcare恰恰 highlights contraction in other areas, with professional business services, internet, and manufacturing employment all declining. Examples like layoffs at Amazon and chemical giant Dow, the latter being an important weight indicator for US manufacturing, clearly reflect the疲态 of the lower part of the US "K-shaped" economy. Services, manufacturing, real estate, and employment are all being挤出 by AI, overall presenting a state of "jobless growth."

Judging from recent earnings reactions, if performance isn't significantly above expectations, stock prices often react tepidly.但只要 results fall slightly short, the punishment is extremely severe, with almost no intermediate state of "no reward, only penalty." For instance, Microsoft's stock fell about 18% within roughly a week after its earnings report, because its cloud revenue growth was only 26%, below expectations, while capital expenditure growth was as high as 66%, significantly exceeding expectations, creating a stark contrast. Meta, with 99% of its revenue来自 advertising, showed超预期 revenue in Q4, demonstrating AI's ability to enhance ad efficiency, but its lofty capital expenditure plans also worried the market.

Advertising business is highly correlated with the real economy's景气度. If the real economy weakens, reducing demand for services like phones, cars, offline consumption, accounting, and legal, the advertising business will inevitably face pressure. Meta's use of AI to improve ad efficiency本质上 constitutes more cost reduction than revenue growth; its revenue growth still relies on the real economy. Even if luxury goods and high-end consumption related to the "K-shape's" upper end can contribute ad budgets, once pressure spreads to phones, cars, and even high-end consumption, impacting the advertising industry is only a matter of time.

Furthermore, the part of the US economy being挤出 is precisely the foundation nurturing tech giants' cash flows.一旦 the real economy is continuously squeezed, revenue from advertising, e-commerce, software, and subscriptions will directly affect the free cash flow of Meta, Google, and Amazon, in turn constraining their capital expenditure capabilities. If AI investment directions are wrong, or returns fall short of expectations, the stock price adjustment faced will be far from minor.

Among major tech giants, Google, with its "cloud," "chips," and models, and both revenue and profit exceeding expectations, can be called the "top student" in the class.但即便如此, the market remains highly挑剔 of its doubled CAPEX. Companies with poorer performance need not be mentioned, like AMD with its weak guidance, leading to a roughly 17% single-day drop.

The only relatively "refreshing" case is Apple. Previously, the market worried Apple would be marginalized by the AI narrative, with Siri迟迟未 finding a suitable AI载体. However, Q4 phone sales, especially in China supported by trade-in policies, provided crucial support for its performance. More importantly, Apple's capital expenditure continues to decline, with a full-year figure of only about $12 billion, appearing格外克制 among trillion-dollar市值 tech giants with CAPEX often reaching hundreds of billions. This makes Apple, even without necessarily大幅超预期 earnings, a temporary emotional safe haven for funds.

Nvidia has not yet reported earnings, but its narrative differs from others. As a "shovel seller," Nvidia only needs to持续 sell computing power. Jensen Huang's narrative始终围绕规模化扩展, from language models to the physical world, autonomous driving, embodied intelligence, even宇宙级 physical modeling, with almost limitless story space. Because its business model本质上 involves selling computing equipment, other companies cutting or expanding capital expenditure has limited impact on its short-term performance. Thus, Nvidia has become one of the few giants truly profiting in this trend.

Fourth, the interlinked financing and circular承诺 act as an emotion amplification mechanism. The "interlinked dependency" story is another catalyst for the tech stock adjustment. Previously, the core figure was Sam Altman, around whom giants made mutual investment承诺, binding themselves through orders, equipment procurement, and cooperation, forming circular financing and快联 transactions.

From September to October last year, Altman, to realize his grand narrative, broadly sought funding from Nvidia, Saudi Arabia, the UAE, Japan's SoftBank, etc.,不断扩大 the AI story, forming a capital承诺—financing—orders闭环. Giants continuously联合强化 the narrative, like巨额 CAPEX, models接近 AGI, and futures通向宇宙级 applications. During periods of liquidity泛滥 and optimistic sentiment, a promising story vision alone could drive valuation increases.但一旦 liquidity begins to change, the market starts scrutinizing and rationalizing.

Since late October last year, under pressure from Google, the entire OpenAI chain has underperformed. Microsoft, Nvidia, Oracle are高度绑定 to OpenAI. For example, if OpenAI is surpassed, Microsoft faces triple risks: 1) potential writedown risk on its OpenAI investment; 2) impact if other models surpass OpenAI; 3) a significant portion of Microsoft's revenue comes from Office and cloud, with nearly 45% of future orders reportedly来自 OpenAI. The previous interlinked narrative around OpenAI放大 order expectations layer by layer. Now, with the core loosening, order fulfillment faces uncertainty. If叠加 with an external environment of liquidity tightening, the amplification effect could reverse into逐层 valuation compression.

For OpenAI, current technological leadership doesn't equate to a commercial闭环. It hasn't truly earned operating revenue, only the faith premium from investors, while costs and投入 are solved through continuous financing, so it doesn't prioritize cost control for application diffusion. Now, the competitive landscape for models has changed. Claude performs well, and Anthropic has一直与 OpenAI并驾齐驱. Google advances on dual tracks, with both the world model Gene and the language model Gemini. In areas like office software, accounting software, legal software, and financial analysis, GPT hasn't established an absolute advantage. OpenAI is no longer the sole leader today, finding it difficult to claim "far ahead" status in ratings and comprehensive performance.

This also highlights the difficulty of investing in tech—the AI industry might indeed be the "vast星辰大海" long-term, potentially comparable to the iPhone revolution or even接近 the internet revolution's level. But a good company isn't necessarily a good investment; good companies serve the industry diligently, with success not necessarily belonging to them.回顾 the internet era, from search to social media, only one or two winners remained ultimately. Google found its business model through ad ranking, Facebook formed scale effects through social ads, finding genuine commercial闭环 and building moats amidst the cheap infrastructure after the bubble burst. Therefore, one way to invest in tech is广泛投资, capturing high returns from the few surviving entities. In the AI era, model differences are巨大 initially but gradually收敛, similar to autonomous driving. After directions融合, gaps缩小. Those who can persist, find funding, find conversion paths, and manage the enterprise well will be the true winners.

Fifth, the loosening of narratives may催生 greater volatility in US stocks. US markets often correct consensus expectations early in the year, as the start of each year is a window for concentrated corporate disclosure of earnings, guidance, and capital expenditure expectations, while liquidity often experiences marginal changes. For example, market sentiment was very optimistic early in 2025,普遍认为 Trump would强势回归 and continue his first-term policy approach of "everything for the stock market"—monetary easing, fiscal expansion, policy support—making US stocks better and fatter.然而, Trump's first step upon return was fiscal tightening, not expansion, dealing a direct blow to US stocks highly dependent on monetary easing and fiscal expansion environments. Entering 2026, this rhythm of "eliminating consensus expectations" has reappeared. In this context, market expectations for full-year rate cuts and monetary easing are unstable.

Especially as this is a midterm election year, politics can hardly承受更大规模的 monetary easing and fiscal expansion, with opposition forces明显增强. Therefore, this year's US liquidity environment, fiscal expansion expectations, and the trajectory of long-end Treasury yields hold extreme uncertainty. In such an environment,只要 companies provide guidance or capital expenditure plans容易引发分歧, it can trigger a re-correction of expectations in the early part of the year.

Looking ahead, numerous variables affect this year's liquidity environment—Fed leadership transition, inflation prone to rising rather than falling, midterm election pressures—all potentially bringing political conflict. US government shutdowns and declining fiscal spending efficiency could lead to marginal liquidity tightening. In this environment, the market will question how the gap between exponentially growing capital expenditure and relatively sluggish earnings can be bridged.

From a fundamental perspective, the core issue requiring attention is earnings pressure. The K-shaped divergence in the US economy is intensifying, reflected in rising electricity prices, basic raw material prices, and storage prices, indicating structural疲软 in the employment situation with signs of扩散. Capital markets will increasingly demand clear profit returns from tech companies, especially regarding whether AI-related business models can truly deliver. Simultaneously, the punishment mechanism for excessive capital expenditure deserves持续关注.

From an industry perspective, theoretically, the emergence of large models could change software forms or even replace some software functions, potentially leading to a major reshuffle in the software industry.但从更长期看, the path from large models to Agents to embodied intelligence remains a long-term vision; the core industrial logic hasn't fundamentally changed. Current volatility stems more from liquidity and risk appetite disturbances than a剧变 in industrial direction.

Meanwhile, the "interlinked dependency" narrative is loosening. In the past, it was a crucial link for storytelling around earnings, liquidity, and capital expenditure. If this link disintegrates, unless new narratives and new liquidity expectations fill the void, companies on the chain will inevitably be affected.

Early this year, high-volatility beta assets like gold, non-ferrous metals, cryptocurrencies, and Bitcoin experienced联动 deleveraging. Various assets are高度相关 at the liquidity level, and deleveraging triggered declining risk appetite,爆发 a chain reaction. The earnings environment for US stocks in 2026 is not轻松. Market patience is接近极限. Valuation demands on earnings will become more严苛. Emotionally catalyzed increased volatility is almost certain, but whether tech stocks have reached an inflection point仍需观察. What is certain is that market skepticism will增强 this year, requirements for earnings delivery will显著提高, and tolerance for capital expenditure will继续下降.

Risk提示: Uncertainty regarding AI commercial prospects;超预期 monetary policy changes during the Fed Chair transition year;超预期 changes in Trump's midterm election odds;超预期 geopolitical developments.