Analysts Expect AI Capex to Reach $800 Billion in 2026. That Number Could Actually Surpass $1 Trillion. Here Are 2 Stocks That Will Benefit.

AI infrastructure capex is expected to rise materially: consensus calls for ~$800B in 2026, but Goldman Sachs estimates the “true” figure is closer to ~$1T, with U.S. spending potentially lower while global and private-company AI investment is projected to be $200B+ above expectations. McKinsey projects AI capex could reach $7T globally by 2030. The article argues investors can express this theme via AI leaders with expanding international demand (e.g., Nvidia) and AI-exposed semiconductor names with Asia revenue exposure (e.g., Qualcomm).

Analysis

The investable takeaway is not that AI capex is large; it is that the spend is becoming geographically and organizationally broader, which extends the runway for the semiconductor complex and reduces the odds that this is just a U.S. hyperscaler trade. That favors the highest-share wallet capture names first: NVIDIA should keep converting incremental industry spend into revenue faster than anyone else because it still owns the switching costs and software lock-in. The second-order winners are the picks-and-shovels around power, networking, and memory; the losers are downstream software vendors that will face tougher ROI scrutiny as customers keep pouring cash into infrastructure before monetization is visible.

The main risk is a timing mismatch: capex can accelerate for 2-3 quarters while stock multiples compress if investors start asking where the cash returns are. That matters especially for quality-vs-optional AI exposure names like Qualcomm, where the AI narrative is still a future call option rather than a near-term earnings driver; its China concentration also makes it more vulnerable to any export-control or demand wobble over the next 1-2 quarters. If hyperscalers start signaling digestion pauses, or if NVIDIA lead times normalize faster than expected, the market can quickly rotate from "scarcity premium" to "earnings skepticism."

Contrarian view: the consensus is probably underestimating the duration of the build-out but overestimating how broadly the economic gains will accrue. In prior infrastructure booms, the early winner captured most of the margin pool while late-cycle entrants absorbed the depreciation. That argues for owning the dominant enabler rather than the "cheap AI beta" stories; the latter may work only when edge AI revenues become visible, likely 12-24 months out, not immediately.

Key Decisions for Investors

• Long NVDA on any 3-5% pullback over the next 2-4 weeks; target a continuation move into the next earnings cycle, but trim if capex commentary stops accelerating or gross margin guidance softens.
• Pair trade: long NVDA / short QCOM for a 1-3 month relative-value expression of 'clear monetization' versus 'optionality'; thesis breaks if QCOM starts showing tangible data-center or edge-AI revenue inflection earlier than expected.
• Do not chase QCOM here; treat it as a watch item for 2026+ edge-AI adoption. Reassess only after a quarter with measurable AI contribution to revenue or margin, not on narrative alone.
• Use any AI-capex headlines to add to a basket of infrastructure beneficiaries rather than software; if forced to stay concentrated, prefer the highest-quality enabler over second-order beneficiaries until ROI evidence appears.
• Alert level: if NVDA underperforms the SOXX/SMH basket by ~5% over 4-6 weeks while capex estimates keep rising, that would signal valuation saturation and warrant reducing exposure.