3 Incredible AI Stocks That Aren't Too Late to Buy Now

The article argues Nvidia, Broadcom, and Micron remain attractive AI beneficiaries, citing Nvidia's projection for data center capex to rise from $600 billion in 2025 to $3 trillion-$4 trillion annually by 2030. Broadcom says its custom AI chip business could reach $100 billion or more in revenue by 2027, while Micron is benefiting from a memory-chip shortage and analysts project revenue of $109 billion this fiscal year and $171 billion next year. The piece is opinionated rather than event-driven, but it reinforces bullish sentiment around AI infrastructure stocks.

Analysis

The market is still underpricing how quickly AI capex can migrate from a hyperscaler spend story into a supplier-share story. The key second-order effect is that the infrastructure stack is bifurcating: NVDA remains the default for flexible training and frontier models, while AVGO is increasingly the pick-and-shovel winner for custom silicon tied to inference and workload-specific economics. That mix matters because it broadens the addressable spend pool rather than cannibalizing it, which supports multiple winners even if GPU unit growth normalizes.

The real medium-term pressure point is memory, not compute. MU’s upside is less about one-quarter pricing power and more about a multi-quarter constraint regime where DRAM/HBM tightness forces customers to pre-buy capacity and accept less favorable contracts to secure supply. If hyperscalers and model developers keep pulling forward deployments, memory can stay the tightest bottleneck in the stack, which tends to preserve margin upside longer than consensus expects; that also creates a lagged risk that capex announcements become less elastic if supply-chain lead times remain stretched.

What the consensus may be missing is that a broad AI capex boom can coexist with relative underperformance in adjacent beneficiaries if investors chase the obvious names too aggressively. The more interesting trade is not simply long AI beta, but long the companies with pricing power and bottleneck exposure versus the names where growth is already fully embedded. The near-term risk is a digesting phase if hyperscaler guidance pauses for even one quarter; these names are still sentiment-sensitive and can de-rate quickly on any hint of capex moderation.

Key Decisions for Investors

• Stay long NVDA but trim into strength; use it as the benchmark exposure to AI spend, while expecting near-term multiple compression if capex growth rates decelerate for even one quarter.
• Add AVGO on any post-earnings or sector pullback; the cleaner risk/reward is 3-6 months out as custom silicon revenue compounds and the market re-rates the mix toward durable inference demand.
• Initiate/maintain a long MU position with a 6-12 month horizon; the trade is a cyclical shortage super-cycle, but size modestly because memory is the first place where supply response can eventually cap upside.
• Pair long AVGO / short a basket of high-multiple AI infrastructure names with less direct monetization of workload shift; this expresses the view that custom silicon and bottlenecks will outperform generic AI beta.
• If NVDA rallies sharply on capex headlines, sell upside via call spreads rather than outright shorts; the long-term thesis remains intact, but sentiment-driven pullbacks are likely if guidance stops accelerating.