Amazon's Core Engine Roars Back

AWS reaccelerated to 20.2% YoY growth in Q3, producing $33.0B in revenue and $11.4B of operating income, while the AWS backlog expanded to $200B driven by rapid adoption of Trainium2 chips and multi‑billion AI compute commitments. Amazon’s consolidated EPS rose to $1.95, but massive CapEx pushed TTM free cash flow down to $14.8B as property and equipment spending hit $115.9B; CapEx is expected to approach ~$125B in FY25, creating liquidity strain and keeping valuation under pressure despite the cloud-led rebound.

Analysis

Market structure: AWS reacceleration (20.2% YoY, $33B revenue) and a $200B backlog make Amazon (AMZN) a primary beneficiary; Nvidia (NVDA) and other AI accelerator suppliers gain indirectly from rising AI compute demand while legacy on‑prem vendors and smaller cloud providers risk share loss. Pricing power for large-scale training is rising — expect upward pressure on spot GPU/accelerator rental pricing and longer-term committed‑capacity contracts, tightening supply/demand for AI compute through 2025–26. Fixed‑asset intensity (CapEx ~ $115.9B YTD, guidance toward ~$125B FY25) shifts capital to energy, copper and data‑center services, pressuring free cash flow growth near-term.

Risk assessment: Tail risks include regulatory AI/antitrust action (12–24 months), capital markets shock that forces Amazon to cut CapEx, or chip supply interruptions that delay backlog fulfillment. Immediate (days) risk is volatility around re‑rating; short term (weeks/months) is guidance‑driven liquidity scrutiny — TTM FCF fell to $14.8B and provesensitive if FY25 CapEx >$125B; long term (2–3 years) upside hinges on backlog conversion and sustained multi‑billion AI contracts. Hidden dependency: AWS growth depends on customer willingness to shift CAPEX to OPEX and on Trainium2/custom silicon supply — failure to scale would force price concessions.

Trade implications: Direct plays: selectively overweight AMZN for exposure to AI cloud (2–3% portfolio) while using downside protection; add NVDA exposure (1–2%) to capture GPU/AI secular tailwinds but size to volatility. Pair trade: go long AMZN and short ORCL (or legacy enterprise software exposure) to express cloud share rotation; use options collars to cap downside if AWS backlog conversion <20% over next four quarters. Rotate 3–6% from consumer discretionary into semiconductors (SOXX) and cloud infra names over 1–3 months; trim if AMZN FCF < $10B TTM or CapEx guidance rises >$130B.

Contrarian angles: Consensus underprices conversion risk of the $200B backlog — not all will convert within 12 months; markets may be over‑penalizing AMZN for CapEx even as AWS operating income ($11.4B) materially lifts EPS. Historical parallel: prior AWS reaccelerations led to multi‑year re‑ratings once backlog converted and margins expanded — watch conversion rates and large contract starts as lead indicators. Unintended consequence: aggressive CapEx builds moat (custom chips, scale) but create 12–24 month liquidity stress; exploit periods of panic selling to add positions if fundamentals (backlog conversion, FCF trends) remain intact.

Key Decisions for Investors

• Establish a 2–3% portfolio long in AMZN via stock or Jan 2027 LEAP calls (near‑ATM) to capture multi‑year AWS AI cycle; hedge with a 1% cost financed collar (sell 20–30% OTM calls, buy 15% OTM puts). Trim/add triggers: trim if FY25 CapEx guidance > $130B or TTM FCF falls below $10B; add if trailing 4‑quarter backlog conversion rate > 20%.
• Allocate 1–2% long NVDA via 9–12 month call spreads (buy calls, sell higher strike calls) to limit premium outlay while benefiting from sustained AI compute demand; take profits on a 30–50% move or if implied vol rises > 25 pts above historical realized vol.
• Initiate a relative value pair: long AMZN (1.5–2%) / short ORCL (1–1.5%) via stock or directional option positions to express cloud share shift; stop‑loss: cut the short if ORCL outperforms by 15% on a rolling 3‑month basis.
• Reduce consumer discretionary exposure (XLY or XRT) by 2–4% over the next 1–3 months and redeploy into cloud infra and semiconductor ETFs (SOXX, IGV) to reflect secular AI capex reallocation; re‑evaluate after next two quarterly AWS backlog conversion datapoints (next 2 quarters).