When AI Spending Became a Red Flag: The Correlation That Changed Everything

The Critical Infrastructure Costs Behind Woo's Thesis

The article's central claim—that hyperscaler capex is now viewed as a cost problem rather than a return signal—finds strong validation in the physical realities of AI infrastructure. Meta's recent announcement of a shift to Air-Assisted Liquid Cooling (AALC) reveals the magnitude of infrastructure pressure that Woo identifies. Meta's Vice President for Engineering explicitly stated that "increasing power demands and liquid cooling requirements are forcing fundamental rethinking of platform, rack, power, and data center design." This isn't incremental spending—it's architectural overhaul at unprecedented scale.

The article states hyperscalers will spend "more than 2% of U.S. GDP on AI-related capex this year," contrasting this with total U.S. advertising spending at 1.5% of GDP. This comparison is devastating when you consider that digital advertising—already 75% of all ad spending—represents the primary monetization channel for several hyperscalers. Woo is essentially arguing that the four largest tech companies are collectively spending more than the entire advertising market generates, betting on AI returns that remain largely theoretical.

The infrastructure vendors mentioned in available sources—ABB, Eaton, Schneider Electric, and Vertiv—are positioned in AI data center power infrastructure, suggesting a massive build-out underway. What Woo calls a "guarantee of over-investment" appears grounded in the physical reality that AI data centers require fundamentally different infrastructure than traditional data centers, including specialized cooling and power systems that weren't necessary for previous generations of cloud computing.

What the Market Behavior Reveals About Investor Confidence

Limited independent sources were found for this specific topic. The following analysis draws on established context where citations are unavailable.

Woo's observation about the correlation breakdown between capex announcements and stock performance represents a fundamental shift in market psychology. For three years, increased capex guidance sent AI stocks higher because investors interpreted it as confidence in future returns. The reversal—stocks falling on capex increases—suggests the market now sees these investments as mandatory defensive spending rather than offensive growth plays.

The claim that "only half of Nasdaq 100 stocks trade above their 100-day and 200-day moving averages" while the index itself remains flat indicates dangerous internal deterioration masked by index-level stability. This breadth weakness typically precedes broader declines, as fewer stocks carry the weight of the entire index. The Magnificent 7 breaking trend lines and sitting below 100-day moving averages confirms that even the strongest performers are losing momentum.

Woo's identification of retail investors as the "white knight" who saved the AI trade "at least five times over the past year" points to institutional distribution disguised by retail buying. When sophisticated money exits and retail money provides the bid, the setup typically precedes significant downside. The post-Microsoft earnings retail surge cited from Charles Schwab data exemplifies this pattern—professional sellers found willing buyers among less-informed investors.

The Revenue-Capex Gap Nobody Can Explain

The article's most important claim—that "AI-related capex rises far faster than AI-related revenue"—strikes at the core investment thesis. Woo notes that the four-quarter moving average of Magnificent Seven EBIT "slowed to its lowest level since 2023" despite massive capex increases. This divergence creates an unsustainable dynamic where spending accelerates while profit growth decelerates.

Woo's reference to Chuck Prince's infamous 2007 quote—"As long as the music is playing, you've got to get up and dance"—draws a parallel to pre-financial crisis excess. The comparison suggests hyperscalers are trapped in a prisoner's dilemma: no single company can stop spending without risking competitive disadvantage, even if collectively the spending makes no economic sense. This dynamic historically produces overcapacity and subsequent crashes.

The statement that hyperscalers "are motivated less by greed than by fear" and "don't know better than you or I if there is a pot of gold under the rainbow" reframes the entire AI investment narrative. If true, the capex surge represents defensive positioning against existential risk rather than calculated pursuit of identified returns. This matters enormously for valuation—markets price growth based on expected returns, not on fear-driven spending to avoid obsolescence.

Interestingly, while Woo has been short the Nasdaq since September 2025, prominent investors moved in the opposite direction earlier. Stanley Druckenmiller's Duquesne Family Office opened new positions in Amazon, Alphabet, and Meta during Q3 2024, suggesting elite investors were still buying the AI thesis at that time. The divergence between Druckenmiller's Q3 positioning and Woo's September short illustrates how rapidly the investment landscape shifted.

The Semiconductor Cycle Extension Problem

Woo's mention that "the semiconductor bull cycle is already historically extended" deserves emphasis. Typical semiconductor cycles run 3-4 years from trough to peak. The current AI-driven cycle, powered by demand for GPUs, high-bandwidth memory, and specialized AI chips, has now sustained elevated growth since early 2023. Historical patterns suggest these cycles end not with gradual slowdowns but with sharp inventory corrections when demand falters.

The specific cost pressures Woo identifies—high-bandwidth memory chips, transformers, cooling systems, construction materials, grid upgrades, land prices in power-rich regions, and specialized labor—represent a comprehensive bill of materials inflation unique to AI infrastructure. Unlike previous tech cycles where Moore's Law provided consistent cost deflation, AI infrastructure faces structural cost inflation across multiple input categories simultaneously.

The comparison to total advertising spending becomes even more striking when considering monetization timelines. Traditional cloud computing had clear revenue models from day one—companies paid for compute and storage. AI's monetization remains largely speculative beyond narrow use cases. Hyperscalers are spending 2%+ of GDP on infrastructure for applications that don't yet generate commensurate revenue.

What Woo's Positioning Means for Broader Markets

Woo's continued short position on the Nasdaq 100 despite recent market stability suggests high conviction in his thesis. His statement that "when the music stops, there will be no place to hide" implies he expects a broad tech sector repricing, not just weakness in a few names. Given the Magnificent 7's concentration in major indices, a tech bear market would likely drag broader equity markets lower.

The article's February 16, 2026 date places this analysis in a future context from the current perspective, but the structural arguments about capex-revenue divergence and infrastructure costs reflect real tensions visible in current earnings and guidance. Meta's actual infrastructure announcements about liquid cooling and the reality that specialized data center infrastructure vendors are positioning for massive AI buildouts confirm the spending surge is very real.

The trading volume data available for the Magnificent 7 stocks shows some interesting patterns. Apple's current trade volume (50.453M) sits below its trailing-twelve-month average (53.804M), and dramatically below its 10-year average (73.089M) and 20-year average (126.991M). This declining volume trend across long timeframes suggests diminishing retail and institutional participation in what was once the most actively traded stock globally—potentially supporting Woo's thesis that momentum is fading.

The Implications If Woo Is Right

If Woo's analysis proves correct and the AI trade enters a bear market, the implications extend far beyond tech stocks. The massive capex commitments already made create fixed costs that will pressure margins if revenue growth disappoints. Companies that spent billions on AI infrastructure with unclear ROI timelines will face difficult questions from shareholders about capital allocation.

The infrastructure buildout itself creates economic ripples. Power infrastructure vendors, construction firms, specialized labor markets, and land values in data center regions have all adjusted to AI-driven demand. A sharp pullback in capex would create overcapacity and stranded assets across this entire supply chain.

For investors, Woo's most important insight may be the shift from viewing capex as a bullish signal to viewing it as a red flag. This psychological shift, if it takes hold broadly, removes a key support for tech stock valuations. Without the assumption that spending equals confidence in returns, the market must evaluate AI investments based on actual revenue and profit generation—a much higher bar given current monetization levels.

The article presents a coherent bear case grounded in observable market behavior, earnings trends, and infrastructure economics. Whether Woo's timing proves correct depends on factors he acknowledges are unknowable—including how long retail investors continue buying dips and when institutional selling pressure overwhelms that support. But the structural argument about unsustainable capex-revenue divergence deserves serious consideration regardless of near-term price action.