Jensen Huang Said This Word Exactly 1 Time During Nvidia's Earnings Call, and That Was Enough to Put Artificial Intelligence (AI) Bubble Fears to Rest

The article argues the AI build-out is supported by scale: the top five AI hyperscalers are projected to spend about $800B on capex this year and $1.3T next year, with cloud backlogs around $2T. It highlights Nvidia’s stated “visibility,” enabling fiscal 2028 guidance of 70% revenue growth (floor given supply constraints) and frames Nvidia as moving from chip supply to “AI factory” full-stack planning across land, power, networking, optics, and memory. Overall, it treats the common “AI bubble” narrative as incomplete, implying secular demand rather than demand evaporation.

Analysis

The important shift is not “AI demand is real” — that is already consensus — but that the profit pool is migrating from isolated chip sales toward control of the full deployment stack. That structurally favors NVDA, but it also raises the value of adjacent bottlenecks: optical interconnects, custom silicon, and HBM/memory capacity. In other words, the next leg of alpha is likely to be in the constraint setters, not the headline winner, because those names will benefit from a longer inventory cycle and less pricing transparency.

The real risk over the next 1-3 quarters is that the market starts underwriting the build-out as a capex problem rather than a revenue story. If AMZN, META, GOOGL, and MSFT keep lifting spend while AI monetization remains back-end loaded, their free cash flow and ROIC optics can compress even if the demand narrative stays intact. That is a different failure mode than a bubble burst: the theme can remain secular while the stocks derate on capital intensity and depreciation.

Contrarian take: the market may be underestimating how much of this spend is already pre-committed, which supports NVDA’s shipment visibility, but it may be overestimating how much multiple expansion is left after that visibility is monetized. The cleaner expression is to own the supply-chain enablers with lower expectations and shorter payback periods. If memory pricing eases or hyperscaler capex guides flatten, the “visibility” premium should fade quickly, which would be the first falsifier for the bullish chain.

Key Decisions for Investors

• Tactically buy NVDA only on 5-8% pullbacks; this is a quality compounder trade, not a chase at any price. Falsify the thesis if the next guide implies gross margin pressure worse than expected or if hyperscaler capex commentary rolls over for two consecutive quarters.
• Long MRVL / COHR basket versus short a hyperscaler spender basket (AMZN, META) for 1-3 months. The long side captures higher attach rates to networking/optics, while the short side expresses risk that capex outruns near-term monetization; stop if ad/cloud revenue inflects enough to cover incremental depreciation.
• Accumulate MU on weakness as a second-order beneficiary of HBM and memory tightness, preferably via call spreads into the next earnings cycle. Risk/reward is attractive if pricing remains firm, but exit if spot memory pricing softens or lead times normalize faster than expected.
• If you want the cleanest structural expression, pair long NVDA with short AMZN or META rather than the broader index. This isolates the supplier-versus-spender dynamic and should work over 1-2 quarters if the market keeps scrutinizing free-cash-flow conversion.
• Set an alert on hyperscaler capex guides and AI-related power/interconnect lead times; those are the real catalyst variables. A flattening capex trajectory or easier supply would argue for taking profits across the whole AI infrastructure basket.