AWS Accelerates Again to +28% — OpenAI Signs for 2GW of Trainium and the Demand Side of the $200B Capex Story Finally Lines Up
Initial Read: Across-the-board beat with the AWS reacceleration thesis fully validated — AWS +28% (fastest in 15 quarters, +400bps QoQ from +24%), revenue $181.5B vs. Street $177.3B, Q2 guide midpoint $196.5B vs. Street ~$179B (9% above), OpenAI signs for 2GW of Trainium and Anthropic expands to 5GW. Capex stepped to $44.2B in the quarter — but the demand prints make that math work. AH tape mixed (~flat to +4%) as the market digests capex; the print is decisively bullish.
Key Takeaways
- AWS at +28% Y/Y to $37.6B — the fastest growth in 15 quarters and a fourth consecutive quarter of acceleration (+17% Q1'25 → +17.5% Q2 → +20.2% Q3 → +24% Q4 → +28% Q1'26). AWS operating income $14.2B (+23% Y/Y) at a 37.7% margin, up 80bps Y/Y. The "supply-side capacity wall" thesis from Q1-Q2 2025 is now fully reversed — capacity is converting to revenue at the highest rate AWS has ever printed.
- Demand validates the $200B capex commitment: OpenAI signs for ~2GW of Trainium capacity (ramping 2027), Anthropic expands to up to 5GW of current and future Trainium generations, Meta signs for tens of millions of Graviton cores. The chips business crossed a $20B annual revenue run rate growing triple-digit Y/Y. The Q4 selloff ($200B FY26 capex spook) was specifically about whether demand would arrive — Q1 prints are the strongest possible answer.
- Retail and ads are also accelerating: WW paid units +15% (highest since covid lockdowns), Advertising $17.2B (+24% Y/Y, +22% ex-FX, fifth straight quarter at +22% ex-FX), NA segment $104.1B (+12%) at 7.9% operating margin, International $39.8B (+19%, +11% ex-FX) at 3.6% margin (record International margin). The retail engine that was supposed to slow on tariffs and weakening consumer is doing the opposite.
- Q2 guide is the tell: revenue $194.0–199.0B (+16-19% Y/Y, midpoint ~$196.5B) vs. Street ~$179B — 9-10% above consensus at the midpoint. Operating income $20.0-24.0B vs. Street ~$19B and Q2'25's $19.2B. Guide assumes Prime Day in Q2 (it's been pulled forward to June, a month earlier).
- FCF TTM compressed to $1.2B from $25.9B Y/Y as TTM capex stepped to $151B (vs. $93B prior TTM). This is the visible cost of the $200B FY26 commitment — but operating cash flow grew 30% to $148.5B, so the cash engine itself is fine. FCF is structurally suppressed for the duration of the buildout, not impaired.
- Rating: Maintaining Outperform. Q3 2025's upgrade thesis ("AWS reacceleration off the capacity wall is real and the operational engine is excellent") just got its third consecutive validation, now amplified to its strongest data point. The OpenAI Trainium contract specifically retires the bear case that custom silicon was a single-customer (Anthropic) story. Nothing in this print warrants a further upgrade only because we are already at Outperform.
Results vs. Consensus
| Metric | Actual (Q1 2026) | Consensus | Beat/Miss | Magnitude |
|---|---|---|---|---|
| Revenue | $181.52B | $177.30B | Beat | +2.4% |
| EPS (GAAP, diluted) | $2.78 | $1.64 | Beat | +69.5% |
| Operating Income | $23.9B | ~$18.0B (implied) | Beat | +~33% |
| AWS Revenue | $37.59B (+28% Y/Y) | ~$36.0B (implied +22-23%) | Beat | +~$1.6B |
| AWS Operating Margin | 37.7% | ~35% (implied) | Beat | +~270bps |
| Advertising Revenue | $17.24B (+24% Y/Y) | ~$16.9B (+21%) | Beat | +~2% |
| FCF (TTM) | $1.2B | n/a (capex-suppressed) | In line | -$24.7B Y/Y |
Quality of Beat
- Revenue: Beat is broad-based across all three reporting segments. Y/Y growth was +17% reported and +15% ex-FX (the $2.9B FX tailwind drove ~200bps of the headline). Importantly, the beat is not consumer-fragility-related — WW paid unit growth of +15% is the strongest reading since the tail end of covid (Q4 2020-Q1 2021), and 3P seller mix at 60% means the unit economics are broad. Tariff and consumer-weakness fears that defined the late-2025 narrative did not materialize in the quarter.
- AWS: +28% reported, +28% ex-FX (no meaningful FX in AWS) — pure organic acceleration. AWS operating income of $14.2B (37.7% margin) was up 80bps Y/Y despite a $200B/yr capex run rate, because depreciation step-up is being more than offset by mix and absorption. Margin will likely compress over the back half as new capacity comes online and depreciates against partial-quarter revenue, but the beat here suggests AWS GMs are holding at structurally higher levels than the bear case modeled.
- EPS: The $2.78 GAAP EPS is heavily flattered by a $16.8B pre-tax non-operating gain from Anthropic investments — strip the Anthropic mark and EPS would be roughly $1.55-1.65 (a clean 0-5% beat), still solid but not the 70% headline. The operating beat is $23.9B operating income vs. ~$18B implied consensus, which is the durable read. Anthropic gains are non-cash and will continue to appear as long as Anthropic raises at higher valuations — this is now a recurring "below-the-line" benefit, but should not be relied on as core EPS.
- Margins: NA operating margin 7.9% (-110bps Y/Y from 9.0% in Q4'25 but flat with mid-2025; this is the seasonal pattern, not deterioration). International 3.6% — a record International segment margin print. WW operating margin 13.1% — up 180bps Y/Y. The operational margin engine continues to grind higher.
- FCF: The $1.2B TTM FCF is the lowest Amazon has printed in years and reflects capex stepping to $151B TTM vs. $93B prior TTM (+$59B). This is the single most-watched line in the report, but the numerator (operating cash flow) actually grew 30% to $148.5B — so this is purely a denominator (capex) issue. FCF will stay compressed through FY26 as the $200B annual commitment plays out, and the math only works if AWS revenue conversion catches up — which is exactly what this print delivered.
AWS Segment Performance
| Metric | Q1 2026 | Q4 2025 | Q1 2025 | Y/Y | QoQ |
|---|---|---|---|---|---|
| AWS Revenue | $37.59B | $35.58B | $29.27B | +28.4% | +5.6% |
| AWS Y/Y growth (reported) | 28% | 24% | 17% | +400bps QoQ | — |
| AWS Operating Income | $14.16B | $12.45B* | $11.55B | +22.6% | +13.7% |
| AWS Operating Margin | 37.7% | 35.0% | 39.5% | -180bps | +270bps |
| AWS Operating Margin (TTM) | 35.2% | 35.4% | 37.5% | — | — |
*Q4 2025 AWS op income implied from segment disclosures.
Assessment: The Inflection Compounds
The AWS growth arc is now +17% → +17.5% → +20.2% → +24% → +28% — five consecutive quarters of acceleration with the most recent two prints adding +400bps each. There is no public-cloud business in history that has reaccelerated this hard from a $115B+ run rate. The Q4 disclosure of $244B backlog (+40% Y/Y) framed the demand visibility; this print is the demand visibility converting to revenue at a faster rate than even the optimistic backlog math implied.
The 80bps Y/Y margin compression Q1 vs. Q1 (39.5% → 37.7%) is what was expected — depreciation from the capex ramp lands on the P&L well before the corresponding revenue. What's notable is the 270bps QoQ margin expansion (35.0% → 37.7%), which suggests the AI capacity coming online in Q4 is already running at near-mature absorption. This is the cleanest read yet that the Trainium2/3 capex is converting to high-margin revenue, not stranded.
The OpenAI 2GW commitment is genuinely thesis-changing. Through Q4 2025, the bear narrative on AWS custom silicon was that it was effectively an Anthropic story (Anthropic was the anchor customer for Trainium2 and Project Rainier). OpenAI signing for 2GW of Trainium — beginning to ramp in 2027 — proves that custom silicon has multi-customer demand pull at frontier-model scale. Combined with the Anthropic 5GW expansion and Meta signing for "tens of millions of Graviton cores," AWS now has three of the four largest AI hyperscalers as accelerator/CPU customers. The chip business at $20B+ run rate growing triple digits is no longer a "promising new line" — it is a top-5 standalone semiconductor business by revenue.
North America & International Segments
| Segment | Revenue | Y/Y (reported) | Y/Y (ex-FX) | Op Income | Op Margin | Y/Y Margin |
|---|---|---|---|---|---|---|
| North America | $104.1B | +11.7% | +11% | $8.27B | 7.9% | +160bps |
| International | $39.8B | +19.0% | +11% | $1.43B | 3.6% | +60bps |
| AWS | $37.59B | +28.4% | +28% | $14.16B | 37.7% | -180bps |
| Consolidated | $181.5B | +17% | +15% | $23.86B | 13.1% | +180bps |
Retail Underlying Detail
| Revenue Line | Q1 2026 | Y/Y (reported) | Y/Y (ex-FX) |
|---|---|---|---|
| Online stores | $64.25B | +12% | +9% |
| Physical stores | $5.79B | +5% | +4% |
| Third-party seller services | $41.58B | +14% | +12% |
| Advertising services | $17.24B | +24% | +22% |
| Subscription services | $13.43B | +15% | +12% |
| Other | $1.65B | +25% | +25% |
Assessment
North America at 7.9% margin (+160bps Y/Y) sustains the trajectory of the Q4 2025 record (9.0% — seasonally inflated). The 7.9% Q1 read is the highest non-Q4 NA margin Amazon has ever printed and reinforces that the supply-chain re-engineering work disclosed across 2025 (regionalization, inbound network productivity, robotics deployment) is now structural. Unit growth of +15% WW with shipping cost growth of +14% means roughly flat unit-shipping economics on a unit basis — Amazon is still leveraging volume against shipping, and the 1-hour and 3-hour same-day expansion is happening within that envelope.
International at 3.6% margin is a record. Two years ago this segment was loss-making; the +320bps print in Q2'25 was the first credible "International is structurally profitable now" data point, and Q1 2026 confirms that progress (with the prior trough behind us). The +11% ex-FX growth is among the better non-AWS prints and is helped by Amazon Now (30-minute delivery) launching in Tokyo and 8 Brazilian cities and same-day expansion globally.
Advertising at $17.24B (+24% reported, +22% ex-FX) is the fifth consecutive quarter at +22% ex-FX. TTM advertising revenue crossed $70B per Jassy's letter — making Amazon Ads now larger than Meta's Reality Labs revenue (admittedly a low bar) and trailing only Google Search and Meta Family of Apps in scaled digital ad businesses. The Netflix DSP tie-in disclosed in the release adds to the existing CTV inventory stack (Prime Video, Spotify, Netflix, Roku, Disney) and consolidates Amazon's #1 position in CTV ads.
3P seller mix dropped to 60% from 61-62% — the lowest in the disclosure window. This is worth flagging: 1P inventory growth is outpacing 3P, which is a small headwind to long-run gross margin (1P is lower-margin than take-rate-only 3P). It's likely tariff-related (1P merchants are absorbing more inventory risk), but worth tracking.
Notable Items in the Release
OpenAI Signs for 2GW of Trainium Capacity
This is the headline single disclosure of the release. OpenAI committing to ~2GW of Trainium capacity through AWS infrastructure, beginning to ramp in 2027, represents the first major non-Anthropic frontier-model commitment to AWS custom silicon. Until today, the bear case on Trainium was that it was an Anthropic-only chip — OpenAI's commitment retires that argument.
Assessment: Strategically the most important disclosure in any AMZN earnings release in 2025-2026. The 2027 ramp timing is consistent with Trainium3 supply commitments disclosed in Q4 2025; this likely represents some portion of that "nearly fully committed" inventory landing with OpenAI.
Anthropic Trainium Commitment Expanded to 5GW
Q4 2025 disclosed Anthropic on Trainium at multibillion-$ revenue with 1M chip target by year-end (Project Rainier). Today's release expands that to "up to 5 gigawatts" of current and future Trainium generations — a meaningful step-up in scale and a signal that Anthropic is increasing rather than reducing AWS exposure even as Anthropic's enterprise customers are diversifying.
Assessment: Confirms Anthropic's commitment to AWS infrastructure beyond initial Project Rainier scale. 5GW vs. 2GW (OpenAI) implies Anthropic remains the larger Trainium customer.
Chips Business Run Rate Crosses $20B
The custom silicon business (Graviton + Trainium + Nitro) crossed a $20B annual revenue run rate growing "triple digits Y/Y." Q4 2025 disclosed $10B+ run rate; this is roughly a doubling in two quarters. At $20B run rate, AWS's chip business is larger than Marvell's, larger than ARM Holdings' 2024 revenue, and roughly 1/3 of AMD's data-center segment — entirely organic, entirely captive to AWS workloads.
Assessment: The chip business is a thesis pillar in its own right now. It is the single highest-leverage line item AMZN has — captive demand, integrated economics, and no external sales-channel cost.
Anthropic Mark-to-Market Gain of $16.8B
Q1 2026 includes a $16.8B pre-tax gain in non-operating income from Amazon's Anthropic investments. This is a non-cash mark on the warrant/equity stake recognized as Anthropic's implied valuation rose during the quarter. Net income of $30.3B vs. $13.5B operating-attributable net income (after-tax operating income proxy) means roughly $13B of net income is the Anthropic non-cash mark.
Assessment: Important to model this out of clean EPS. The Anthropic stake will continue to mark up as Anthropic raises (and could mark down on a write-down event). Treat this as below-the-line investment income, not core operating earnings. Headline $2.78 EPS overstates underlying earnings power by ~40-45%.
Capex Step to $44.2B in the Quarter
Purchases of property and equipment hit $44.2B in Q1 2026 alone vs. $25.0B in Q1 2025 — a $19.2B Y/Y step. TTM capex of $151B (vs. $93B prior TTM) is now running at a $151B annualized pace, putting the company on track for the $200B FY26 commitment if Q2-Q4 averages ~$50B/quarter. The step from $44B Q1 to ~$50B average for Q2-Q4 implies acceleration through the year, consistent with Trainium3/4 supply landing and one million NVIDIA GPUs deploying starting in 2026.
Assessment: Capex is on plan for the disclosed $200B FY26 commitment. The market spook from Q4 2025 was about whether demand would arrive — the OpenAI/Anthropic/Meta announcements today validate that demand is in fact arriving. The ratio of multi-year contracted demand to annual capex is now structurally favorable.
Project Hail Mary, Prime Day, Amazon Leo
Three notable retail/entertainment items: (1) Project Hail Mary (Prime Video's tentpole release) at nearly $615M global box office to date — strong tentpole confirmation for the Studios slate; (2) Prime Day will occur in June, a month earlier than the typical July timeframe — pulls Prime Day fully into Q2 (consistent with the Q2 guide assumption); (3) Amazon Leo signed Delta Air Lines (in-flight Wi-Fi from 2028), Vodafone (mobile coverage), and the DP World Tour (golf tournament connectivity), and announced a planned acquisition of Globalstar to add direct-to-device services with Apple iPhone and Apple Watch as first deployment.
Assessment: The Globalstar acquisition is a meaningful Leo-thesis upgrade — it converts Leo from satellite-only to satellite + direct-to-device (D2D), with Apple as the anchor D2D partner. Leo commercializes in 2026 with anchor enterprise customers signed; this looks broadly on plan from the Q4 2025 framing.
Headcount Roughly Flat
Total full-time and part-time employees of 1,575,000 — down 1,000 from Q4 2025 and up 1% Y/Y. This is the lowest Y/Y headcount growth in the disclosure window. AWS revenue is growing +28% on roughly flat headcount, which is the operating-leverage definition of cloud unit economics scaling.
Assessment: Operating leverage in services is real and visible. Compare to AWS revenue growth of +28% and ad revenue growth of +24% — both at flat headcount. This is what scaling cloud infrastructure looks like at the most efficient operator.
Guidance & Outlook
| Metric | Q2 2026 Guide | Q2 2025 Actual | Implied Y/Y | Street Pre-Print | Beat vs. Street |
|---|---|---|---|---|---|
| Revenue (low) | $194.0B | $167.7B* | +15.7% | ~$179B | +8.4% |
| Revenue (mid) | $196.5B | $167.7B* | +17.2% | ~$179B | +9.8% |
| Revenue (high) | $199.0B | $167.7B* | +18.7% | ~$179B | +11.2% |
| Operating Income (low) | $20.0B | $19.2B | +4.2% | ~$19.0B | +5.3% |
| Operating Income (mid) | $22.0B | $19.2B | +14.6% | ~$19.0B | +15.8% |
| Operating Income (high) | $24.0B | $19.2B | +25.0% | ~$19.0B | +26.3% |
*Q2 2025 base of ~$167.7B implied from Q1 2026 commentary on Q2 growth rates.
Implied Growth Trajectory
The Q2 guide midpoint of +17.2% Y/Y is essentially a maintain of the Q1 +17% reported growth rate (and +15% ex-FX), indicating Amazon is not signaling deceleration in Q2. Critically, the Q2 base benefits from Prime Day pulling forward to June (which adds ~$1-2B of incremental revenue vs. a non-Prime-Day Q2). Strip Prime Day pull-forward and underlying Q2 growth is likely closer to +15-16% — still robust and consistent with the Q1 trajectory.
Operating income guide is the bigger tell: $20-24B (midpoint $22B) vs. $19.2B Q2'25 implies +15% mid-point operating income growth. This is materially better than the +5% the Q2'25 guide gave us a year ago at the equivalent point ($13-17.5B vs. $17.7B Q1'25). The op-income guide style has shifted from "just good enough to maintain" to "comfortably ahead of consensus" — which is consistent with the fundamental momentum.
Guidance Style Assessment
This is the most aggressive Q2 guide Amazon has issued in the disclosure window. The Q1'25 guide for Q2 was conservative on op-income and got punished. The Q1'26 guide for Q2 is comfortably ahead on both lines. Either management has higher visibility (most likely — backlog, AWS contracted commitments, ad pipeline) or they've calibrated guidance style upward (less likely, Amazon historically guides conservatively). Either way, the guide signals confidence — and the $4B operating-income guide range ($20-24B) is wider than usual, reflecting capex-dependent depreciation step-ups that are still uncertain in timing.
Free Cash Flow & Capital Structure
| Metric (TTM) | March 31, 2026 | March 31, 2025 | Y/Y Change |
|---|---|---|---|
| Operating Cash Flow | $148.5B | $113.9B | +30.4% |
| Capex (gross) | $151.0B | $93.1B | +62.2% |
| Free Cash Flow | $1.2B | $25.9B | -95.4% |
| Long-term debt issued (TTM) | $68.4B | $0.7B | +$67.6B |
| Cash & equivalents (end of period) | $104.7B | $69.9B | +$34.8B |
The headline FCF compression to $1.2B TTM is the single most challenging line in the print and the one that creates the most short-term tension with multiple-expansion arguments. But the components tell a more nuanced story:
- OCF +30% — the cash-generation engine is intact and accelerating.
- Gross capex +62% to $151B — driving the FCF compression. Of this, a meaningful portion is AWS infrastructure (data centers, custom silicon, NVIDIA GPUs) and the balance is fulfillment expansion + Amazon Leo + Project Kuiper buildout.
- $53B of long-term debt issued in Q1 alone vs. nothing in Q1'25 — Amazon is now using its balance sheet aggressively to fund the buildout. Cash on hand grew $35B Y/Y, so the company is accumulating liquidity ahead of capex deployment.
- $15.4B of acquisitions and non-marketable investments in Q1 — likely reflects the Globalstar planned acquisition + Anthropic round participation + non-marketable investment marks.
The FCF curve looks ugly until you map it against AWS revenue conversion. AWS revenue grew ~$8.3B Y/Y to $37.6B (annualized run rate ~$150B). Each $1B of AWS data-center capex converts to roughly $300-400M of incremental annualized revenue at ~35-37% margins. The $58B incremental capex drives roughly $17-23B of incremental annualized revenue when fully ramped — so over a 24-36 month conversion window, the math closes. FCF will rebuild in 2027-2028 as the capex curve normalizes against the larger AWS base.
Questions for the Call
- OpenAI 2GW Trainium contract — economics and exclusivity: What is the dollar value of the OpenAI Trainium commitment, and is OpenAI making concurrent commitments to other clouds (Microsoft, Google) or are they consolidating frontier inference workloads on AWS? Bullish answer: 2GW @ $5-10B/yr revenue at maturity, multi-year, with implied volume that consolidates a meaningful share of OpenAI inference. Bearish answer: Vague on dollar size, with explicit commentary that this is one of multiple cloud deployments at OpenAI.
- AWS margin trajectory through the back half: The 270bps QoQ margin expansion (35.0% → 37.7%) is striking. Is this absorption-driven (capacity coming online and running near-mature) or mix-driven (high-margin AI training revenue offsetting lower-margin general compute)? Bullish answer: Absorption is now fully matched to demand, and the margin compression bear case is largely behind us. Bearish answer: Q1 was a benign mix quarter and we should expect 100-200bps of compression as new capacity lands in Q2-Q3.
- $200B FY26 capex — split between AWS, retail/Leo, and acquisitions: What portion of the $200B is AWS data centers / custom silicon vs. fulfillment + Leo + other? Q1's $44B and the implied $52B/quarter for Q2-Q4 gives a track, but the segment split is unclear. Bullish answer: 80%+ AWS, with AWS capex efficiency (revenue per $ capex) continuing to improve. Bearish answer: Lower AWS share with more going to Leo, Kuiper, and other lower-return adjacencies.
- FCF inflection timing: When does TTM FCF return to the $25-30B range last seen in early 2025? Implicit in the capex commitment is a multi-year suppression. Bullish answer: Mid-to-late 2027, as AWS revenue conversion catches up. Bearish answer: 2028 or later if capex stays elevated.
- Tariff impact and 1P/3P mix shift: 3P seller mix dropped from 61-62% in 2025 to 60% in Q1'26. Is this tariff-related (3P sellers reducing inventory exposure) or 1P-driven (Amazon expanding direct selection)? Bullish answer: 1P-driven, reflecting Amazon's expanded selection and higher-margin private label. Bearish answer: 3P retreat under tariff pressure, which is a slow-bleed gross margin headwind.
- Anthropic/OpenAI dual-customer dynamics: Anthropic at 5GW + OpenAI at 2GW in Trainium gives AWS a unique multi-frontier-model position. How does AWS think about competitive dynamics between two customers operating frontier models on the same custom silicon platform — does it create pricing pressure or pricing power? Bullish answer: Pricing power, because Trainium economics are favorable to both customers and AWS controls the platform. Bearish answer: Hint of customer concentration risk if one of the two pulls back.
Market Reaction
Earnings released today after market close (~5:00 PM ET); call begins 5:30 PM ET. As of the time of writing, after-hours trading shows AMZN roughly flat to +4%, with reports of price action bouncing around in extended trading. The mixed AH read appears to reflect investors weighing the operational beat (+28% AWS, $9B above-consensus Q2 guide) against the persistent capex compression of FCF — the same tension that drove the -7-11% reaction to the Q4 2025 print where the $200B FY26 capex was first disclosed.
Our directional call is that the AH chop will resolve meaningfully positive after the call, particularly if management quantifies the OpenAI 2GW deal economics or provides any framework for FY26-27 AWS revenue conversion against the capex commitment. The Q4 2025 selloff happened because demand visibility was unclear at the moment of the $200B disclosure; this print delivers the demand visibility (OpenAI + Anthropic 5GW + Meta Graviton + +28% AWS print). We expect the multiple to re-expand modestly into the next 1-2 weeks as the Street works through the implications.
Model Implications
| Item | Prior View | Post-Earnings | Reason |
|---|---|---|---|
| FY26 AWS revenue growth | +22-24% | +25-27% | Q1 +28% sets a higher base; OpenAI/Anthropic commitments visible from H2 2026 |
| FY26 consolidated revenue | ~$770-790B | ~$790-810B | Q1 print + Q2 guide both running 5-9% above prior models |
| FY26 operating income | ~$93-100B | ~$100-108B | Q1 op income $23.9B + Q2 guide midpoint $22B implies H1 ~$46B; consistent with $100B+ FY |
| FY26 AWS operating margin | 34-35% | 35-36% | Q1 37.7% gives buffer against H2 capex depreciation step-up |
| FY26 capex | $200B (per Q4 disclosure) | $200B (maintained) | Q1 $44B + ~$52B/quarter Q2-Q4 lands at the disclosed commitment |
| FY26 FCF | $5-25B (suppressed) | $0-15B (more suppressed) | Q1 TTM $1.2B; FY26 likely the trough year for FCF |
| FY27 AWS revenue (preliminary) | ~$190B | ~$200-210B | +25% on a $160B FY26 base; OpenAI 2027 ramp adds upside |
Valuation: The print supports a higher fair-value range than our prior Q4 2025 framing. AWS at +28% growth and 35-36% margins on a $150B+ run rate, combined with retail at structurally higher margins (NA at 7-9%, International at record 3.6%) and advertising compounding at +22% ex-FX, justifies a multiple at the upper end of where AMZN has historically traded. The FCF compression is the constraint — but it is a temporary buildout cost against a forward earnings curve that just got materially stronger.
Thesis Scorecard
| Thesis Point | Status | Notes |
|---|---|---|
| Bull #1: AWS reacceleration off the capacity wall is real and sustained | Confirmed (strongly) | +28% Q1 vs. +24% Q4 vs. +20.2% Q3 — third consecutive quarter of acceleration |
| Bull #2: Trainium is a multi-customer, multi-billion-dollar standalone business | Confirmed (strongly) | OpenAI 2GW + Anthropic 5GW + Meta Graviton; chips business at $20B run rate triple-digit growth |
| Bull #3: Retail margins are structurally higher post-supply-chain re-engineering | Confirmed | NA 7.9% Q1 (a non-Q4 record); International 3.6% Q1 record; +15% WW unit growth |
| Bull #4: Advertising is structurally a top-3 digital ad platform globally | Confirmed | $70B+ TTM, +22% ex-FX 5th straight quarter; Netflix DSP integration adds CTV reach |
| Bull #5: Custom silicon vertical integration drives AWS unit economics advantage | Confirmed (strongly) | AWS margin expanded 270bps QoQ to 37.7% despite capex ramp; chips business +100%+ Y/Y |
| Bear #1: $200B FY26 capex commits to demand that may not arrive | Challenged (decisively) | OpenAI/Anthropic/Meta commitments + +28% AWS print + $20B chips run rate retire this concern |
| Bear #2: FCF compresses to multi-year low and stays there | Neutral (still suppressed) | FCF $1.2B TTM, likely trough; but OCF growing +30% means cash-gen engine is fine |
| Bear #3: Memory chip supply tightening pressures 2026 unit economics | Neutral (not visible yet) | Q1 results don't show memory pressure; risk remains real for H2 2026 |
| Bear #4: Tariffs / consumer weakness pressure NA retail | Challenged | WW unit growth +15% (highest since covid) and NA op income +43% Y/Y is the opposite of consumer weakness |
| Bear #5: Anthropic mark-to-market gains create earnings volatility | Neutral (real, manageable) | $16.8B Q1 mark inflates EPS by ~$0.95; needs to be modeled out of clean EPS |
Overall: Thesis materially strengthened. The AWS reacceleration arc has now compounded to its strongest data point. The single most important development is the OpenAI 2GW Trainium commitment, which retires the "Anthropic-only" bear case on custom silicon and converts AWS's chip business into a multi-frontier-model platform. The capex commitment that drove the Q4 selloff now has demand-side validation that makes the math credibly close.
Preliminary Action: Hold positions; the multiple should re-expand on this print. Adding on weakness is appropriate for risk-tolerant accounts, but we are not chasing into the AH tape pending the call. Pending call confirmation on (1) OpenAI dollar economics, (2) AWS margin sustainability, and (3) FCF inflection timing.