My 3 Favorite AI Stocks to Buy Right Now

TSMC plans to invest an additional $100B in Arizona to support advanced packaging fabs and its 2nm process, bringing its total Arizona commitment to $265B—tailwinds for U.S. semiconductor capacity tied to AI demand. Micron reported fiscal Q3 revenue of $41.45B (+345% YoY) and net income up 1,400% to $28.24B (EPS $24.67, up from $1.68 a year earlier). Nebius posted Q1 revenue of $399M (+684% YoY) but spent $2.5B in capex on GPUs/hardware while raising $6.3B with $9.3B cash, alongside $12B+ of dedicated capacity agreements with Meta starting 2027.

Analysis

The clean read-through is not “AI is strong” but that the spend is migrating from a narrative trade into a capex-and-supply-chain trade. That favors the businesses with the most pricing power and the least execution risk: TSM’s moat widens if customers keep paying for geographic redundancy, and ASML gets longer-duration visibility as every new node and packaging line demands more tools. The less obvious loser is anyone selling commoditized exposure to the same boom without the same scarcity value.

Micron remains the sharpest operating-leverage expression, but also the most fragile. Memory rallies tend to look structural right before they revert, so the market can be right on demand and wrong on duration; the key question over the next 1-2 quarters is whether AI-driven DRAM mix can outrun any supply response. If pricing flattens or inventory turns up, the multiple can compress faster than earnings fall.

Nebius is the most interesting but also the most financeable-risky name: huge asset growth backed by external capital works until utilization or customer timing slips. The Meta backstop lowers demand risk, but not dilution risk, and that makes NBIS more of a balance-sheet/options trade than a pure operating story. Contrarian take: the market may be underpricing the infrastructure winners and overpricing the “emerging platform” intermediaries; quality of funding matters as much as quality of demand.

Time horizon matters: the first move is sentiment, the next 1-3 months are revisions and order flow, and the 6-18 month story is whether AI capex turns into durable margin pools rather than one-off buildouts. Falsifiers are simple: a softer DRAM pricing tape, a margin hit at TSM’s U.S. ramps, or any sign NBIS must re-fund growth before utilization catches up.

Key Decisions for Investors

• Buy TSM on weakness for a 3-6 month hold; use it as the highest-quality AI infrastructure exposure. Thesis breaks if U.S. ramp commentary implies a meaningful gross-margin drag or if advanced-node demand fails to reaccelerate.
• Overweight ASML versus the broader semiconductor basket over 12-18 months. This is the cleaner second-order beneficiary of multi-year foundry/packaging expansion, with less earnings volatility than memory or GPU-adjacent names.
• Hold MU only with a disciplined exit plan; best expressed via call spreads or a reduced-sized long after pullbacks. Take profits if DRAM pricing momentum stalls or inventory days start trending up in the next 1-2 quarters.
• Avoid or underweight NBIS as a core long; if long at all, size it like a venture-style equity story, not a compounder. The key risk is that financing remains available but equity holders get diluted before contracted capacity converts to cash flow.
• If you want a relative-value hedge, pair long ASML against short MU for 3-6 months. That captures AI capex persistence while fading the more cyclical, higher-beta leg if memory sentiment overextends.