The Most Obvious Bubble In History: Unlikely To Pop This Year

Hyperscalers plan about $725B of 2026 CapEx (roughly $800B including Oracle), up ~77% YoY, but the article argues the increase overstates incremental AI capacity because ~$25B of Microsoft’s spending is tied to higher component pricing rather than more servers—about 35% of Microsoft’s CapEx growth. It estimates roughly $95B of the four hyperscalers’ combined 2026 CapEx is “price inflation” (cutting growth to ~54% from 77%), while the financing risk is highlighted via claims that AI buildout debt could reach ~$4.1T through 2030 and may be partially off-balance-sheet. On the “adopter” side, surveyed CEOs report limited realized AI benefits, suggesting returns may be concentrated among “picks and shovels” rather than end users, keeping a cautious risk posture.

Analysis

The key market mechanism is that a growing share of AI capex is behaving like a supplier inflation tax, not a clean capacity expansion. That shifts incremental economics away from hyperscalers and toward memory, networking, and infrastructure vendors, but it also means the buy-side is paying more for the same marginal compute. In the next earnings cycle, the market should care less about headline capex growth and more about whether free cash flow, buybacks, and cloud margins are being diluted faster than AI revenue is scaling.

The first-order winners remain the sell-side monetizers and component suppliers, but the second-order winner may be relative-value longs in names with visible AI monetization and less balance-sheet strain. The bigger loser set is the hyperscaler cohort if capex inflation persists: even if revenue stays strong, the multiple can compress when investors realize the spend is partially price-driven and increasingly debt-supported. That creates a subtle but important asymmetry: upside from AI is still intact, but the market may start discounting the denominator before the numerator shows up.

The contrarian miss is that this is not yet a clean bubble-top call; it is more likely a phase change from "growth story" to "capital allocation story." Over the next 1-3 months, the catalyst is any further guidance revision on capex inflation and any sign that memory pricing is rolling over; over 6-18 months, the falsifier is sustained EBIT uplift at adopters. If adopter ROI does not show up, the eventual repricing will hit software, IT services, and the banks/enterprise names that are treating AI as an efficiency project rather than a revenue engine.

Key Decisions for Investors

• Long MU / short MSFT as a 1-3 month relative-value pair: MU benefits from AI component inflation, while MSFT absorbs the cost drag. Falsify if Microsoft guides capex down materially or DRAM prices flatten faster than expected.
• Add on pullbacks to NOW and CRM versus hyperscaler exposure: these are the cleanest monetization stories with less direct capex burden. Prefer them over MSFT/META if the market keeps rewarding AI spend without clear FCF conversion.
• No aggressive outright short in NVDA yet, but use strength to hedge with short-dated put spreads if SOXX/NVDA squeeze on capex headlines. The thesis breaks if AI revenue acceleration stays ahead of component inflation for another quarter.
• Watch JPM and LYG as adopter-lag proxies rather than core shorts: if they fail to show measurable AI-driven cost takeout by the next reporting season, it supports a broader "ROI gap" trade against enterprise adoption names.