Amazon’s Jassy says AI revenue run rate is over $15 Bln, hints at future chip sale

Amazon disclosed AI services at AWS generating >$15 billion annualized revenue in Q1 2026 and said its custom chip business (Graviton/Trainium/Nitro) doubled to >$20 billion annualized from the $10 billion run rate disclosed previously. Management reiterated a large AI-focused capex program (~$200 billion this year) and flagged potential future third‑party chip sales (management suggested a hypothetical ~$50 billion standalone run rate), while shares rose ~2% pre-market and Citron set a $300 price target. These disclosures materially reduce investor concern about AI spending by showing sizable near-term revenue, with potential upside if chip sales to third parties materialize.

Analysis

Amazon’s move toward an integrated AI stack is a structural supply‑side shock: by internalizing key silicon and rack-level systems, the company can convert what are today variable GPU procurement costs into internalized capital and software margins. Over a 2–5 year horizon this favors AWS margin expansion (we estimate 3–7 percentage points potential gross‑margin tailwind), and creates pricing optionality that can undercut standalone GPU ASPs on cost/performance — not by substituting CUDA overnight, but by offering better total cost of ownership to hyperscaler customers.

The immediate second‑order winners are firms exposed to hyperscaler capex (TSMC, EDA and packaging vendors) and orchestration software that reduces friction for third‑party rack buyers; losers are incumbent GPU vendors whose addressable market for hyperscaler sales could compress and whose ASPs face renewed downward pressure. There is also potential channel conflict and demand reallocation risk for OEM systems integrators and board/ODM suppliers as Amazon shifts volume into vertically integrated designs, concentrating wafer and interconnect demand.

Key reversals to watch: silicon performance parity (or lack thereof) against the incumbent stack, enterprise adoption of non‑CUDA runtimes, and regulatory scrutiny if Amazon begins selling rack compute to former customers. Near‑term price moves will be event driven (earnings, guidance, trade shows) while the structural thesis plays out over multiple hardware refresh cycles (12–36 months); any clear benchmark showing Trainium/Graviton trailing by >20% on key workloads would materially slow adoption and reprice expectations.

Key Decisions for Investors

• Pair trade (12 months): Long AMZN equity vs Short NVDA equity, dollar‑neutral. Rationale: capture AWS margin re‑rate and expected ASP pressure on discrete GPUs. Position sizing: start 2% net long portfolio exposure, hedge to dollar neutrality; stop‑loss 20% on either leg and trim 50% at +25% relative outperformance.
• Options collar (9–18 months): Buy AMZN LEAP calls ~12–18 months out (10–20% OTM) financed by selling near‑term NVDA calls (3–6 months OTM). Rationale: levered upside to Amazon’s multi‑year margin optionality while monetizing short Vega/Theta in NVDA. Risk: NVDA short leg can be assigned in a short GPU cycle — cap position size to 1–2% of NAV.
• Event short (3–6 months): Short NVDA into positive headlines and reopen into any drawdown, funded by reducing tech beta elsewhere. Rationale: the market tends to front‑run structural threats; a disciplined short into spikes with a protective call is higher probability than a naked long on commoditization. Use a 15–25% protective stop and consider buying cheap longer‑dated puts if volatility spikes.
• Long MSFT or GOOGL (6–12 months) selective exposure: overweight Azure/Cloud/Vertex players to capture secular AI software monetization while hedging hardware commoditization risk. Rationale: these vendors benefit if AWS’s verticalization increases total market spend on higher‑level services even as hardware suppliers face pricing pressure. Target modest 1–3% overweight with profit‑taking at +30%.