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

Amazon’s chip business exited Q1 2026 at an annual revenue run rate above $20 billion, with CEO Andy Jassy saying it would be $50 billion if operated as a standalone company. AWS backlog reached $364 billion, Trainium demand remains nearly fully subscribed, and the business is helping drive Amazon’s fastest cloud growth in 15 quarters. Free cash flow fell sharply to $1.2 billion from nearly $26 billion a year ago due to heavy capex, but the article argues the AI infrastructure build-out and margin benefits support the long-term bull case.

Analysis

AMZN is becoming the stealthest way to express the AI capex cycle because it monetizes both sides of the equation: demand capture through AWS and margin capture through self-supplied silicon. The market is still valuing it like a retail/cloud compounder, but the more important setup is that custom silicon converts AI infrastructure from an externally purchased input into an internal operating leverage engine, which should expand margins even if headline revenue growth normalizes. That creates a second-order benefit the semis don’t get: lower unit compute cost can be reinvested into more capacity, widening the adoption gap versus clouds that rely entirely on third-party GPUs.

The key risk is timing mismatch between capex and monetization. A collapse in free cash flow is not itself bearish if backlog converts on schedule, but it does raise the sensitivity of the stock to any delay in large customer deployments or a change in model economics that reduces training intensity. Over the next 3-6 months, the setup is more about guidance credibility and reservation conversion than current-quarter earnings; over 12-24 months, the real watch item is whether Trainium’s adoption stays differentiated enough to prevent hyperscalers and model labs from re-concentrating spend back into NVDA ecosystems.

The most underappreciated implication for competitors is that AMZN’s custom silicon can force a price/performance reset across cloud contracts, especially for inference-heavy workloads where utilization matters more than absolute peak performance. That pressures margins at peers with less internal chip stack control and could eventually cap premium pricing for GPU-backed cloud instances. It also strengthens the strategic moat around AWS by making switching costs more about software and data gravity than raw compute economics.

Consensus is probably underestimating how durable the demand visibility is, but overestimating how linear the payoff will be. The stock can still de-rate if capex intensity stays elevated without a corresponding margin step-up, so this is not a clean “buy the backlog” story; it is a call option on execution. If management continues to show both backlog growth and margin resilience, the multiple can expand; if not, the market will treat the silicon story as a cash sink rather than a moat builder.

Key Decisions for Investors

• Long AMZN on a 3-6 month horizon into the next two earnings prints; target is continued multiple support if AWS growth and margin commentary stay firm. Risk/reward is attractive versus pure-play semis because downside is partially buffered by retail and ads, but trim if free cash flow recovery stalls.
• Pair trade: long AMZN / short a basket of GPU-exposed cloud beneficiaries or high-multiple AI infrastructure names over the next 1-2 quarters. Thesis is that internal silicon lowers Amazon’s unit costs while peers remain hostage to third-party GPU pricing, creating margin divergence.
• Buy AMZN call spreads 6-9 months out to express upside from backlog conversion without taking full downside from capex volatility. This is preferable to stock if you think the market will reward evidence, not just narrative.
• Watch NVDA for relative underperformance versus AMZN if Trainium adoption broadens; if Amazon continues to take share in inference, NVDA’s pricing power in cloud-adjacent workloads could face incremental pressure over 6-18 months.
• Avoid shorting AMZN outright unless there is a clear sign of reservation slippage or backlog cancellations; the better expression of skepticism is a hedge via short-duration puts around earnings if you expect a guide-down on capex intensity.