Andy Jassy Says This Could Be a $50 Billion Business for Amazon

Amazon CEO Andy Jassy says demand for its AI chips is so strong that the company could sell “racks” to third parties in the future. The Trainium chip is positioned as an efficiency-focused option, with a potential chip-business cash run-rate estimated around $50B annually, implying ~7% incremental growth versus Amazon’s $717B revenue base. The article also highlights valuation support, noting AMZN trades at ~30x P/E vs ~38x for the tech sector ETF, despite the stock lagging the S&P 500 over the past year.

Analysis

This is more meaningful as a cloud-economics story than a chip-revenue story. If Amazon can substitute cheaper internal silicon for merchant GPUs in inference-heavy workloads, the first-order benefit is AWS margin expansion and better customer retention, which can justify multiple expansion even before any external chip sales show up in reported revenue.

The market may be underestimating the second-order effect on pricing power: a credible in-house accelerator lets Amazon negotiate harder with enterprise AI buyers and reduce the risk that AI capex migrates entirely to the GPU ecosystem. That creates a relative headwind for NVDA at the margin, but the bigger competitive threat is to smaller cloud and platform vendors that cannot finance custom silicon at scale. The flip side is execution risk: third-party chip sales sound large, but they can easily become a low-margin, support-heavy business if Amazon overestimates ecosystem pull or underestimates packaging/supply constraints.

The near-term catalyst is the next AWS commentary cycle; over 1-3 months, investors will care less about the theoretical run-rate and more about whether Trainium starts showing up in utilization, pricing, and capex efficiency. Over 6-18 months, the thesis only works if Amazon proves it can convert silicon into durable cloud share gains. Falsifiers are simple: if AWS margin does not inflect, if AI capex keeps rising faster than revenue, or if NVDA maintains share without pricing pressure, the stock should stop repricing on this narrative.

Key Decisions for Investors

• Initiate a tactical long AMZN / short NVDA pair over the next 1-3 months. The bet is that custom silicon improves AWS unit economics faster than it dents Nvidia’s moat, with better risk/reward if entered on any AI-led pullback in AMZN.
• If already long AMZN, prefer upside via 1-3 month call spreads rather than outright stock. The catalyst is incremental AWS margin commentary, and the options structure limits downside if the chip story stays purely optionality.
• Use NVDA as a hedge, not a conviction short. Short interest should be sized small unless the next hyperscaler capex cycle shows explicit accelerator substitution; otherwise the trade risks getting run over by broader AI spend.
• Watch AWS gross margin and capex intensity at the next quarterly print as the key falsifier. If margins do not improve while AI-related capex keeps climbing, fade the chip narrative and rotate back to quality software/infra leaders.
• No aggressive trade on the article alone if AMZN has already re-rated into the print. The market may be extrapolating a long-dated optionality into near-term earnings, which is usually where these stories overextend.