By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel discussion highlights the risks and opportunities in AI capex spending by tech giants, with a net bearish sentiment due to concerns about ROI realization, rising costs, and potential misallocation of capital.
Risk: The risk that current AI capex spending may not generate sufficient returns, leading to stranded assets and margin compression.
Opportunity: The potential for successful monetization of agentic AI workflows by 2025, which could make current capex spending look like a bargain.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
"Big Short" investor Michael Burry's latest X post is once again predicting doom for the AI industry.
"The market has voted and the results are clear," Burry said in his July 27 post. The post also includes two screenshots of charts showcasing Magnificent Seven companies' latest stock performances and valuation changes.
Both charts are taken from a recent Bloomberg article. The first shows a negative correlation between each tech company's forward capex (or capital expenditure) estimates and its stock performance.
The second shows that big tech companies: Nvidia, Microsoft, Amazon, Alphabet, and Meta — five of the companies that make up the Magnificent Seven — have all seen their valuations decline under their 10 year average as of July 24.
Apple, which boasts relatively restrained AI spending, is by far the outlier in S&P 500 performance in the first chart. In the second chart, it's the only company with a valuation above its 10-year average.
Here's what could be behind this performance — and why Burry might be interested.
AI has been in an unprecedented boom ever since ChatGPT entered the picture in late 2022. Companies are spending hundreds of billions of dollars on AI infrastructure — and have mostly been rewarded for it.
That might change soon.
People are starting to push back on some parts of AI usage that were previously embraced. Before, companies tokenmaxxed by encouraging their employees to use as much artificial intelligence as they wanted — or more.
But tokens are getting increasingly expensive as AI companies pass more of their expenses on to their customers. In response, companies are starting to pull back on AI usage, especially as reports come back showing AI isn't as good at reducing workloads as previously thought.
At the same time, regular Americans are starting to push back on AI companies' accelerated data center creation. A Consumer Reports study found that data centers are making electricity more expensive for the people who live close to them, and people who live near data center construction are reporting ecological consequences.
Four leading AI models discuss this article
"Heavy AI capex is already showing up in valuation compression and relative underperformance, a leading indicator that the S&P 500's concentration premium is at risk over the next 12 months."
Burry's charts highlight a clear near-term divergence: the five Mag7 names with the heaviest forward capex (NVDA, MSFT, AMZN, GOOGL, META) have seen both stock performance and valuation multiples compress below 10-year averages as of July 24, while AAPL—with restrained AI spend—has outperformed. The article correctly flags rising token costs, disappointing productivity gains, and local pushback on data-center power and ecology. However, it downplays that capex is largely sunk-cost infrastructure already ordered; utilization and monetization lag by 12-24 months. The S&P 500's concentration risk remains high—Magnificent Seven still ~32% of index weight—so any sustained de-rating would pressure broad-market returns regardless of Burry's narrative.
If AI inference costs fall faster than expected via model compression and new chip architectures, or if enterprise adoption accelerates in 2025 once ROI data improves, the current capex-heavy names could re-rate sharply higher—rendering Burry's 'market has voted' call a classic value-trap signal rather than a trend reversal.
"The market is temporarily mispricing long-term infrastructure investment as a permanent earnings drag rather than a necessary foundation for future software-as-a-service (SaaS) scalability."
Burry’s focus on capex-to-stock-performance correlation misses the 'infrastructure build-out' phase of the AI cycle. While he highlights the negative correlation between spending and price action, he ignores that hyperscalers like Microsoft and Alphabet are investing in long-lived assets—data centers and power grids—that create massive moats. Apple’s current outperformance is a defensive rotation into a cash-rich balance sheet, not a fundamental rejection of AI. The market is currently punishing the 'uncertainty' of ROI, but if these companies successfully monetize agentic AI workflows by 2025, the current capex drag will look like a bargain. We are seeing a transition from hype to operational scrutiny, not a structural collapse.
If AI utility fails to translate into meaningful margin expansion by Q4 2024, the massive depreciation costs on these underutilized data centers will create a multi-year earnings headwind for the entire tech sector.
"Recent Mag Seven underperformance is more likely cyclical valuation reset than evidence that AI capex is fundamentally broken, but the burden is now on earnings to justify 2024 spending within 12 months or Burry's thesis gains real teeth."
Burry's thesis rests on a correlation that may be spurious. Yes, Mag Seven capex leaders underperformed recently—but correlation ≠ causation, and the timeframe matters enormously. Microsoft, Nvidia, and Amazon are still up 20-40% YTD despite July weakness. The article conflates short-term valuation mean reversion with structural AI capex failure. More concerning: the 'tokens getting expensive' claim lacks hard evidence. OpenAI's pricing has actually compressed for bulk users. The real question is ROI realization—not whether spending happens, but whether it generates revenue. Apple's outperformance may simply reflect iPhone cycle strength and lower AI execution risk, not proof that capex-heavy models break.
If enterprise AI adoption genuinely stalls due to token costs or disappointing productivity gains, and if capex doesn't translate to revenue growth within 18-24 months, Burry's mean-reversion thesis becomes prescient rather than premature—and the Mag Seven repricing could accelerate sharply.
"AI capex, while a near-term headwind, should boost long-run productivity and margins for mega-cap tech, justifying higher valuations over time."
Despite Burry’s doom note, the strongest read is that AI spend is a multi-year investment, not a one-quarter drag. The article leans on charts showing near-term valuation gaps, but it overlooks that AI infrastructure is supposed to lift long-run productivity, pricing power, and margin resilience for leaders like Apple and the Magnificent Seven if ROI materializes. Near-term headwinds—data center costs, energy concerns, token economics—could weigh on headlines, yet the market often prices forward cash flows, not today’s capex. The bigger risk is policy, regulatory shocks, or a mechanical capital-expenditure unwind that hits multiple expansion rather than earnings growth. Burry’s track record adds skepticism, but the setup isn’t doomed.
AI spend could prove more protractedly negative to near-term earnings than the article suggests, with ROI delayed or overrated. The correlation between capex intensity and stock performance may be misleading, and if demand or policy worsens, even AI-enabled megacaps could underperform.
"Token economics and depreciation create a 18-24 month earnings overhang that valuation compression has only partially discounted."
Claude's token-cost rebuttal is incomplete: while OpenAI bulk pricing has compressed, inference at scale for agentic workflows still shows rising marginal costs per Claude's own cited productivity shortfalls. This aligns with Gemini's depreciation risk—if utilization lags, those sunk costs become stranded assets by 2026. Burry's correlation may be noisy, but the ROI timeline mismatch remains the core unpriced risk nobody has quantified.
"The current AI infrastructure build-out risks becoming a massive, multi-year margin headwind due to a fundamental disconnect between high inferencing costs and actual enterprise productivity gains."
Gemini and ChatGPT are ignoring the 'utility' trap. We aren't just looking at capex; we are looking at a potential massive misallocation of capital. If these hyperscalers are building data centers for a 'field of dreams' scenario—where they build it and hope agents come—the depreciation cycle will crush margins long before 2026. The risk isn't just delayed ROI; it's that the current LLM architecture is fundamentally too expensive for the productivity gains it actually delivers.
"The real tail risk is guidance misses on utilization, not capex allocation per se."
Gemini's 'field of dreams' framing is sharper than the ROI-delay narrative—but it conflates two separate risks. Capex misallocation is real; depreciation drag is real. Yet neither proves current valuations are wrong if markets are already pricing 40-50% lower utilization assumptions than 2023. The unpriced risk isn't capex itself—it's that *management guidance* on 2025-26 utilization will disappoint, forcing downward revisions mid-cycle. That's the catalyst Burry's charts hint at but don't isolate.
"ROI monetization risk and deployment economics may fail to justify the capex moat, not merely the build-out itself."
Gemini's 'field of dreams' moat critique misses how sensitive the ROI line is to monetization pace, energy costs, and utilization uptime. Capex is front-loaded, but depreciation and power/efficiency headwinds bite margins long before 2026 if usage stalls. Even with data-center scale, a slow adoptions curve or pricing pressure compresses ROIC, undermining the assumes-impervious moat. The real risk isn't build-out—it's whether the economics ever justify the cash spend.
The panel discussion highlights the risks and opportunities in AI capex spending by tech giants, with a net bearish sentiment due to concerns about ROI realization, rising costs, and potential misallocation of capital.
The potential for successful monetization of agentic AI workflows by 2025, which could make current capex spending look like a bargain.
The risk that current AI capex spending may not generate sufficient returns, leading to stranded assets and margin compression.