This Is My Favorite AI Chip Stock (By Far) -- and It's Not Nvidia, Intel, or Broadcom

Amazon’s combined chip business is now running at more than a $20 billion annual revenue rate, with CEO Andy Jassy saying it would be $50 billion if operated as a standalone business. Trainium demand is robust, with more than $225 billion in revenue commitments, Trainium2 largely sold out, and Trainium3 nearly fully subscribed, while AWS backlog reached $364 billion. The build-out is helping drive AWS’s fastest growth in 15 quarters, though free cash flow has fallen to $1.2 billion from nearly $26 billion a year earlier due to heavy capex.

Analysis

AMZN is increasingly functioning as a vertical integration story rather than a pure cloud proxy: custom silicon lowers AWS unit costs, improves inference gross margin, and lets Amazon price performance in a way that can force hyperscale customers to re-architect around its stack. The second-order effect is that every dollar of customer migration to Trainium/Graviton is also a dollar less revenue leakage to NVDA and, to a lesser extent, AVGO’s custom-silicon ecosystem. That makes AMZN the cleaner AI monetization vehicle for investors who want exposure to the build-out without paying for pure-play semiconductor scarcity premiums.

The market is likely underappreciating how backlog converts into a multi-year earnings bridge rather than a near-term revenue pop. Gigawatt-scale reservations imply long-duration capacity consumption, which should dampen cyclical fears and support AWS mix expansion over the next 4-8 quarters. The real tell is not chip revenue itself but the compounding operating leverage from internal silicon adoption: if management’s margin-vs-third-party inference claim holds, each incremental deployment can widen AWS profitability even if headline cloud pricing stays competitive.

The risk is timing mismatch, not strategic validity. Free cash flow compression means AMZN is effectively pre-investing against demand that has to materialize on schedule; if enterprise AI spend pauses for even 2-3 quarters, sentiment could de-rate the stock despite a strong strategic moat. The contrarian read is that consensus may still be treating AMZN as an expensive mega-cap retailer/cloud hybrid, when the more relevant multiple is a platform company with embedded chip optionality and manufacturing-like economics. That multiple expansion can persist for years, but near-term upside likely depends on whether the market continues to reward capex intensity as disciplined moat-building rather than as margin destruction.

Key Decisions for Investors

• Go long AMZN vs short NVDA in a 3-6 month pair trade: thesis is not that NVIDIA weakens outright, but that Amazon captures more incremental economics from AI deployment while trading at a lower multiple of the full stack; stop if NVDA re-accelerates order growth or AMZN capex disappoints.
• Add AMZN on any 5-8% pullback over the next 1-2 months, targeting a 12-month horizon; risk/reward remains favorable if AWS growth stays elevated and silicon adoption continues to expand operating margin.
• Buy AMZN Jan-2027 calls or call spreads instead of common for upside convexity to a re-rating toward a platform/semis hybrid multiple; this is attractive if the market starts capitalizing Trainium as a durable earnings stream.
• Short-term hedge: if holding AMZN long, buy a 1-2 month put spread around the next earnings date to protect against a capex/free-cash-flow reset; the near-term failure mode is a sentiment shock, not a thesis break.
• Relative-value long AVGO vs short INTC only if you want a secondary beneficiary basket; otherwise prefer AMZN because it monetizes the same AI spending cycle with lower direct pricing risk and better internal demand visibility.