My 3 Favorite AI Stocks to Buy Right Now

The article highlights three AI-ecosystem beneficiaries: TSMC plans an additional $100B investment in Arizona (total $265B) for 2-nanometer processing and advanced packaging, which it frames as supply-chain strengthening in the U.S. Micron surged with fiscal 2026 Q3 revenue of $41.45B (+345% YoY) and net income up 1,400% to $28.24B (EPS $24.67 vs $1.68 a year ago), supported by improving capex signals from ASML. Nebius reported Q1 revenue of $399M (+684% YoY) while spending $2.5B on GPUs/hardware, supported by $6.3B raised in the quarter and major capacity commitments tied to Nvidia and Meta.

Analysis

The cleanest second-order winner is not the headline beneficiaries but the picks-and-shovels around AI capacity localization: ASML, advanced packaging, power delivery, cooling, and U.S.-based semi equipment names. TSMC’s U.S. buildout is strategically bullish for customer stickiness, but near term it is a margin mix headwind and a working-capital drag, so the market may overpay for the geopolitical option while underappreciating execution risk over the next 2-4 quarters. The other subtle winner is any supplier exposed to 2nm and advanced packaging qualification; the loser set includes offshore fabs that lack local incentive support and may face slower design-win conversion.

Micron is the more direct cyclical expression here. The market is still too anchored to legacy memory bust cycles, but AI-driven HBM and DRAM tightness can keep pricing firm for 2-3 quarters even if consumer electronics weakens, which supports estimate revisions and multiple expansion. The risk is classic memory: if supply responses from Korea accelerate or hyperscalers pause orders, the earnings slope can reverse quickly, so the thesis is strongest only while channel inventories stay lean and ASPs remain upward.

Nebius is the least clean fundamental long because it is trading more like a financed capacity call option than a self-funding operating business. Large upfront GPU capex can drive headline growth, but the market should discount any revenue that is pre-sold years forward unless utilization and incremental gross margin prove out; that makes it vulnerable to a de-rating if capital intensity outpaces contracted cash flow. Consensus is probably underestimating how often AI infrastructure “growth” is really balance-sheet expansion, not equity value creation, over a 6-18 month horizon.

Key Decisions for Investors

• Long MU on any pullback, 1-3 month horizon: favorable setup if DRAM/HBM pricing stays firm; target a continuation of estimate revisions. Falsifier: inventory days or memory ASPs roll over in the next earnings cycle.
• Add to ASML on weakness, 6-12 month horizon: beneficiary of broader fab and packaging capacity spend, with better quality of earnings than many AI beneficiaries. Falsifier: order intake or lithography backlog slows materially.
• Avoid chasing NBIS after strength; if already long, trim or hedge with puts, 1-3 month horizon. The market may be overpricing contracted capacity before proof of utilization. Falsifier: new customer wins materially improve implied payback on capex.
• Buy TSM on post-event dips rather than momentum, 6-12 month horizon: strategic moat improves, but near-term margin dilution from Arizona capex argues for patience. Falsifier: gross margin guidance or capex intensity worsens beyond expectations.
• Pair trade: long MU / short a basket proxy for AI capex exuberance via NBIS, 1-3 month horizon. This expresses the view that earnings power in memory is more monetizable than balance-sheet-heavy capacity growth; stop if NBIS shows durable positive free-cash-flow conversion.