Amazon Just Revealed a Hidden $50 Billion Business -- Is This the Buy Signal Investors Needed?

Amazon disclosed that its custom AI chips are already at a $20 billion annual revenue run rate and could reach $50 billion if sold to external customers, implying a major new profit pool for AWS. The Trainium line is said to deliver 30% to 40% better price-performance than comparable GPUs and could save AWS tens of billions in annual capex while adding several hundred basis points of operating margin advantage. The article argues Amazon is becoming a credible long-term competitor to Nvidia, but the impact is primarily company-specific rather than market-wide.

Analysis

Amazon’s chip disclosure reframes AWS from a pure demand-side beneficiary of the AI boom into a vertically integrated supplier with a potentially meaningful share of the economics. The second-order effect is margin defense: every point of inference workload migrated from third-party GPUs to in-house silicon reduces unit cost, improves pricing flexibility, and widens the moat against hyperscale peers that still have to buy their way into the stack. The bigger implication is that Amazon can subsidize customer pricing and still expand operating leverage, which is a stronger competitive weapon than headline chip revenue alone.

The market may still be underestimating the optionality embedded in externalizing the silicon business. If Amazon sells racks or chip systems to third parties, the TAM is no longer just internal AWS savings; it becomes a direct assault on Nvidia’s inference layer and, over time, on full-stack AI infrastructure economics. That does not require Amazon to beat Nvidia on every benchmark—only to own enough of the lower-cost workload mix to pressure average selling prices and slow Nvidia’s pricing power at the margin.

The main risk is timing mismatch: capex is immediate, monetization is delayed, and investors could continue to penalize near-term free cash flow before the 2027–2028 payoff window arrives. The biggest catalyst to reverse skepticism would be evidence that Trainium adoption is not just internal but externally pull-based, because that would validate a broader ecosystem shift rather than a one-off AWS optimization. A less obvious downside is that stronger AWS economics may intensify competitive capex from peers, which could keep industry returns on capital suppressed even as Amazon itself improves.

Contrarian view: the consensus is likely overpricing Nvidia’s permanence in inference and underpricing Amazon’s willingness to commoditize part of the stack to win the cloud war. But the move may still be under-owned because the street tends to value Amazon as retail-plus-cloud, not as a semiconductor/platform hybrid. If Amazon executes, the multiple expansion comes from a more durable FCF profile, not from chip revenue alone.

Key Decisions for Investors

• Long AMZN vs short NVDA in a 6-12 month pair trade: express the view that Amazon’s vertically integrated AI economics can compress Nvidia’s pricing power on inference even if GPU demand stays strong; use a 1:1 notional starting point and rebalance on relative performance inflections.
• Buy AMZN call spreads expiring in 12-18 months to capture the market’s likely delayed recognition of chip-driven margin expansion; structure for upside tied to AWS margin surprises rather than absolute multiple expansion.
• Avoid shorting NVDA outright here; if expressing the bearish view, keep it inside a pair trade because Nvidia still benefits from the broader AI capex wave and could outperform on training workloads for several quarters.
• Overweight AMZN on any post-capex drawdown over the next 1-3 months: the risk/reward improves if investors sell the stock on near-term FCF optics while the 2027-2028 monetization window remains discounted.
• Monitor external Trainium shipments as the key catalyst over the next 2-4 quarters; if Amazon starts selling racks to third parties, add to AMZN and reduce NVDA hedge because that is the point where the competitive threat becomes structurally real.