Is AI 'one big bubble'? Behind the tech sell-off

A wave of skepticism is building around AI capital expenditure, with investors openly asking whether the $300B+ annual data-center buildout cycle will translate into returns that justify current valuations. The sell-off is hitting the entire AI stack — from chipmakers to software plays — as sentiment shifts from 'FOMO' accumulation to scrutiny of ROI timelines.

NVIDIA sits at the center of this debate. Its FY2026 revenue of $215.9B represents 65.5% YoY growth with a 71.1% gross margin and 55.6% net margin — numbers that are genuinely exceptional by any historical standard. AMD is growing at 34.3% YoY but operates at a much thinner 12.5% net margin, making it more exposed to any spending deceleration. MSTR is a separate beast entirely — a Bitcoin proxy masquerading in the AI conversation, with -806% net margins and essentially no operating leverage to AI fundamentals.

The bull case for NVDA specifically rests on the argument that its financials are not bubble-like — they reflect real, booked revenue from hyperscaler customers who are contractually committed to multi-year capex cycles. The bear case is that the stock's valuation already prices in continued hyper-growth, and any signal of capex moderation from Microsoft, Google, or Meta would reprice the multiple sharply.

AMD is the more asymmetric name in either direction: if AI spending holds, AMD's data-center GPU ramp could close the margin gap with NVDA; if spending slows, AMD has less cushion in its thinner margins and a less dominant moat. MSTR's inclusion in AI conversations is largely narrative-driven and its fundamentals offer no grounding for an AI-specific trade.

The key catalyst to watch is hyperscaler earnings guidance — any downward revision to data-center capex from the major cloud players would validate the bubble narrative and pressure the entire stack. Conversely, another quarter of NVDA beats and raised guidance would likely deflate the sell-off thesis quickly.