After High Crowding, High Expectations & High Valuations — The Market May Be Repricing AI Capex Returns

Over the past few trading days, the AI supply chain experienced a deep correction. Meta was interpreted as potentially selling excess AI computing power, and Samsung Electronics saw its stock price fall despite releasing better-than-expected earnings guidance. While this superficially looks like a rejection of the AI narrative, the underlying cause is a revaluation of the Return on Investment (ROI) for AI Capital Expenditure (Capex) following a period of high crowding, elevated expectations, and premium valuations.

Subscribe to Our  The Trend Catcher Topic / The Value Anchor Topic  Topics — Stay Tuned for More Trading Opportunities

1. Key Core Events Behind the Recent Shift

Meta's "Capacity Monetization" Disrupts the Scarcity Narrative: Reports indicate Meta is considering developing its cloud infrastructure business to sell AI model access and raw computing capacity. This has prompted investors to ask: Has the massive Capex over the past few years been driven by genuine undersupply, or temporary overbuilding?

Samsung's Earnings Beat Met with Sell-offs: Samsung’s Q2 operating profit is expected to grow significantly and beat market expectations, yet its stock price fell, reflecting a rapid cooling of risk appetite for AI hardware trades.

Asian AI Hardware Stocks Enter "Crowded Trade Unwinding": Goldman Sachs, in its 2H26 Asian Technology Strategy, pointed out that AI server/datacenter hardware stocks were the primary drivers in 1H26, with stock selection spilling over into peripheral sectors. The profit-taking in July is viewed as a healthy consolidation after a rapid rally, rather than a fundamental peak.

2. Underlying Economic Drivers: Shifting Focus from Growth to Returns

The core of this correction is a shift in investor logic—capital is now demanding tangible financial returns. The narrative that rising Capex from tech giants automatically benefits the entire supply chain is no longer sufficient at peak stock valuations. Investors are demanding concrete evidence: Can Capex translate into revenue? Are infrastructure utilization rates high enough? Will free cash flow pressures negatively impact valuations?

Bank of America (BofA) Viewpoint: BofA's report notes that the biggest debate regarding hyperscalers is the ROI of AI capital expenditure. Higher Capex increases fixed costs, depresses free cash flow expectations, and introduces long-term margin risks.

Asset Valuation Logic Shift: If computing infrastructure needs to be leased or sold externally to boost utilization, the market will reclassify AI infrastructure from a "scarce resource" to a "heavy-asset business," shifting valuation models from growth premiums to return on assets.

BofA's Optimistic Quantitative Estimates: BofA estimates that Amazon AWS will generate approximately $10.6 billion in revenue per GW of capacity in 2026, and Google Cloud will hit around $15.7 billion. If Meta successfully commercializes its excess AI capacity to enterprises by 2030, it could unlock an estimated $110 billion revenue opportunity.

3. AI Industrial Fundamentals and Structural Trends

Goldman Sachs remains fundamentally bullish on the Asian AI server and datacenter hardware chain in its technology strategy, citing two core reasons: there is no evidence that semiconductor and component supply exceeds demand; and supply tightness is spreading from memory to MLCCs and electronic materials, with potential extensions to silicon wafers and analog/power semiconductors.

Concurrently, BofA data shows that tech giants' Capex has not stopped, upwardly revising their capital expenditure forecasts:

Alphabet Inc. Class A(GOOGL.US) : Upgraded to $195 billion for 2026 and $290 billion for 2027.

Meta Platforms(META.US) : Upgraded to $145 billion for 2026 and $185 billion for 2027.

Amazon.com, Inc.(AMZN.US)  AWS: Upgraded to $230 billion for 2027.

Furthermore, BofA estimates that building a 1GW AI datacenter requires $25 billion to $45 billion, with the following cost structure breakdown:

4. Outlook and Stock Selection Framework

Short-Term Market Volatility

In the near term, market fluctuations are expected to persist. Investors will use upcoming megacap tech earnings to verify three metrics: whether AI Capex continues to be revised upward; whether cloud/advertising/enterprise AI revenues can match the computing investments; and whether free cash flow pressures remain manageable.

Mid-Term Trading Consolidation

Moving forward, stock selection should pivot away from speculative beta and align with Goldman Sachs' core 2H26 selection framework:

• Direct beneficiaries of price increases or Average Selling Price (ASP) hikes.
• Companies capable of alleviating supply bottlenecks through timely capacity expansions.
• Companies whose actual AI profit contributions are not yet fully priced in.
• Possession of independent catalysts rather than pure reliance on sector sentiment.

Long-Term Asset Bifurcation

In the long run, the AI landscape will differentiate into three distinct asset classes:

Computing Platforms: Evaluated on revenue per GW, utilization rates, depreciation pressures, and external monetization capabilities.

Hardware Bottleneck Providers: Covering HBM/DRAM, advanced packaging, ABF/PCB, MLCCs, optical interconnects, power, and thermal management—driven primarily by supply deficits and pricing power.

Defensive AI Beneficiaries: Enterprise IT services, cybersecurity, and systems integration. Goldman Sachs views certain Japanese IT services and software companies as a safe net during risk-off periods.

Conclusion

This correction marks the formal transition of the AI trade from Phase One (trading Capex upward revisions) to Phase Two (trading Capex ROI). The market is no longer willing to unconditionally buy into the AI vision. At this stage, the optimal strategy is to avoid chasing overextended valuations, preserve liquidity, and rotate allocations into structural plays backed by solid earnings, clear bottlenecks, pricing power, and attractive valuation gaps.