Why I Just Bought Even More of These 3 Underappreciated AI Stocks

The article highlights improving fundamentals across three AI-linked stocks, led by Silicon Motion's 46% year-over-year revenue growth in Q4 2025 and guidance for roughly 80% growth in Q1 2026. IREN's 4.5 GW pipeline and Microsoft deal imply more than $40 billion in potential annual recurring revenue, though the path is capital-intensive and dilution risk remains high. Broadcom posted 28% Q4 FY25 revenue growth, guided to $22 billion for Q2 FY26, and announced a new $10 billion buyback, reinforcing a positive long-term setup.

Analysis

The cleanest second-order read is that the AI trade is broadening from compute to the less-glamorous plumbing: controllers, memory subsystems, and power-dense data-center infrastructure. That matters because the next leg of capex is likely to be constrained less by GPU availability and more by integration bottlenecks, which should support vendors like SIMO even if the market keeps treating them as a peripheral beneficiary. In that setup, the market often underprices the duration of demand because the controller layer benefits from every incremental memory deployment, not just headline AI chip shipments.

IREN’s upside is real but the market is likely over-anchoring on a simple ARR-per-MW extrapolation. The issue is not demand; it is financing intensity and execution latency, which means equity dilution and debt markets are the true swing factors over the next 6-18 months. If hyperscaler demand remains strong, IREN can rerate sharply on contract announcements, but the same operating model makes the stock highly sensitive to any stumble in occupancy, power delivery, or capex inflation.

AVGO looks like the highest-quality expression because it combines AI exposure with visible cash generation and capital returns. The buyback is important less for EPS optics than for absorbing volatility from a market that is still debating custom silicon’s longevity versus merchant GPUs; persistent repurchases signal management’s confidence that free cash flow will outrun even aggressive AI investment. The risk is that consensus is already comfortable with AVGO’s moat, so the stock may need another upward revision in AI revenue mix to sustain multiple expansion.

The contrarian miss is that NVIDIA may be the indirect loser in the custom-chip and memory-enablement narrative: as hyperscalers diversify into bespoke silicon, some marginal AI workloads migrate away from the most expensive general-purpose GPU stack. That does not break NVDA’s story, but it can compress the ceiling on incremental share gains while reinforcing winners deeper in the stack. Near-term, the most likely reversal trigger is not demand weakness but financing stress at neocloud operators or a digestion phase after hyperscaler capex expectations have been repriced too far ahead of revenue realization.

Key Decisions for Investors

• Long AVGO on a 3-6 month horizon; best risk/reward among the group because buybacks and FCF reduce downside if AI spending pauses, while upside continues if hyperscaler capex stays elevated.
• Trade SIMO as a delayed-picks-and-shovels beneficiary versus a basket of memory names over 1-2 quarters; the controller layer should outperform if AI memory attach rates keep rising, with lower valuation risk than the headline memory leaders.
• Treat IREN as a tactical long only into contract/news catalysts, not a core hold; consider buying pullbacks and using tight downside stops because dilution and capex can overwhelm the equity during quiet periods.
• Pair long AVGO / short NVDA over the next 3-6 months as a relative-value expression on custom silicon share gains and capital return support, with NVDA exposed to any slowdown in marginal GPU demand expansion.
• For more aggressive positioning, buy medium-dated IREN call spreads financed by selling out-of-the-money calls; this captures upside from contract wins while limiting damage from equity issuance or project delays.