Amazon.com Inc.
Rating
Strong Buy
High Conviction — Core Position
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Dominant scale, switching costs (Prime), and cost advantage (Logistics). The Q2 2026 report (July 30) moved the single most important moat metric decisively: AWS's contracted backlog reached $496B, from $364B at the end of Q1 2026 and $244B at the end of 2025 — it has roughly doubled in two quarters, and it is contracted revenue, not pipeline. AWS itself grew 37% to $42.2B (a $169B annualised run rate) at a 39.4% operating margin, its fastest growth in 18 quarters, with the AI business and the custom-silicon business each past a $25B run rate growing triple digits. The backlog now rests on both frontier labs rather than one: Amazon invested $50B in OpenAI on February 27, 2026 as part of a $110B round, making AWS the exclusive third-party cloud distributor for OpenAI Frontier and expanding the November 2025 $38B compute agreement by a further $100B over eight years, including ~2 GW of Trainium3/Trainium4 capacity. Amazon has separately committed $33B total to Anthropic ($8B prior + $25B new deal at $380B valuation, announced April 20, 2026) — and Anthropic in turn committed $100B+ in AWS spend over the next 10 years, with 5 GW of capacity additions and access to Trainium2 through Trainium4. On May 4, 2026 Amazon launched Amazon Supply Chain Services (ASCS) — opening its full freight, distribution, fulfillment, and parcel network (≈80,000 trailers; 25% lower transport cost vs alternatives) to any business, regardless of whether they sell on Amazon. P&G, 3M, Lands' End, and American Eagle signed as anchor customers. BofA estimates a $1.3T 3PL TAM; each 1% share equals ~$13B annual revenue and the launch is being framed as the next AWS — externalising an internal capability into a high-margin platform. On June 10, 2026 Amazon expanded ASCS's LTL (less-than-truckload) freight service beyond inbound-to-Amazon shipments to serve any destination — third-party warehouses, distribution centers, retail partners, and distributors — building on a base that has already moved millions of pallets annually for tens of thousands of sellers and vendors since 2019, an early proof point that the platform is scaling as designed. ASCS was not quantified in the Q2 release or on the call — three months after launch there is still no disclosed revenue, customer count, or volume, so it remains optionality rather than an observed contributor. The competitive pressure from Azure and GCP that dominated the last two updates has narrowed rather than widened: at 37% AWS is no longer the slowest-growing hyperscaler, and Jassy told investors AWS is capacity-constrained through 2027 and could eventually be a trillion-dollar annual revenue business. What replaces it as the live risk is the bill — 2026 capex was raised to ~$220B from $200B on higher memory costs, TTM property-and-equipment spend is $169B (+64% YoY), and TTM free cash flow has swung to −$7.6B from +$18.2B. Governance tail risk also persists: a June 2026 report that Amazon's CEO raised security concerns with U.S. officials contributed to an export-control ban on Anthropic's Fable 5 and Mythos 5 models — a reminder that the frontier-lab relationships carry regulatory exposure alongside their commercial upside.
Amazon possesses a Wide Economic Moat driven by four primary pillars:
- Cost Advantage: Its massive fulfillment infrastructure creates unit costs that no competitor can match, allowing for faster delivery and lower prices.
- Switching Costs: The Prime ecosystem locks in consumers. Once a household is integrated into Prime, the convenience makes shopping elsewhere a "costly" friction.
- Network Effect: The 3rd party marketplace creates a flywheel where more sellers attract more buyers, which attracts more sellers.
- AWS as the Primary AI Training Cloud: Amazon has now invested $33B total in Anthropic ($8B prior + $25B new, April 20, 2026) — locking in AWS as the exclusive primary cloud for training and deploying Claude models. Anthropic committed to spending $100B+ on AWS technologies over the next 10 years, adding 5 GW of capacity and using Trainium2 through Trainium4 (including future chip generations). Since February 27, 2026 the same structure exists with the other frontier lab: Amazon put $50B into OpenAI's $110B round, became the exclusive third-party cloud distributor for OpenAI Frontier, and expanded the November 2025 $38B compute agreement by $100B over eight years covering ~2 GW of Trainium3/Trainium4 capacity. Both are bilateral lock-ins — Amazon funds the lab, the lab is structurally dependent on AWS for training and inference — and together they are the visible cause of the backlog moving from $244B at end-2025 to $364B in Q1 2026 to $496B in Q2. That $496B is contracted, multi-year, and concentrated in the two labs least able to move: a demand floor no competing cloud can replicate, and the reason Jassy can describe AWS as capacity-constrained through 2027 rather than demand-constrained. The concentration cuts both ways — a material share of the backlog now sits with two private counterparties whose own funding is not yet self-sustaining.
- ASCS — Logistics-as-a-Service: On May 4, 2026 Amazon launched Amazon Supply Chain Services (ASCS), opening its end-to-end logistics stack — cross-border and domestic freight, bulk warehousing, fulfillment, and parcel — to any business, in any industry, on or off the Amazon marketplace. Anchor customers include P&G, 3M, Lands' End, and American Eagle, with the service explicitly targeting healthcare, automotive, manufacturing, and retail. Amazon cites up to 25% lower transport cost vs alternatives and a 20% conversion lift for fully-managed sellers, leveraging proprietary AI forecasting and a ~80,000-trailer fleet. BofA models a $1.3T 3PL TAM where each 1% share equals ~$13B in annual revenue (~+2% to retail revenue at 1% share by 2027). The strategic parallel is AWS in 2006 — externalising an internal capability into a high-margin platform business. UPS (-8.9%) and FedEx (-7.4%) sold off on the announcement, signalling that the market views this as a structural rather than incremental event.
Ten Moats Verdict
Amazon's moats are overwhelmingly AI-resilient, and Q2 2026 deepened the strongest one rather than the newest: the AWS backlog doubled to $496B in two quarters, custom silicon and the AI line each passed a $25B run rate growing triple digits, and both frontier labs are now contractually and financially bound to AWS. No moat status changes this quarter — the evidence under proprietaryData and systemOfRecord is materially stronger, but statuses were already strong. The ASCS thesis is unchanged and unconfirmed: three months after launch Amazon quantified none of it in its first opportunity to do so, so the logistics platform stays in the moat description as a durable structural argument and out of the growth drivers as a measured one. The live question is no longer whether the moats hold but what they cost — ~$220B of 2026 capex against TTM free cash flow of −$7.6B is the price being paid to keep them.
Alexa voice interface and AWS console are table stakes easily replicated; not a durable differentiator.
AI is commoditizing logistics routing and retail recommendation logic that once required years to build.
Amazon's 200M+ product reviews and buyer behaviour signals still provide meaningful aggregation advantages, but AI scraping and competitor datasets have eroded the edge. The moat persists in scale, not exclusivity.
AWS cloud operations require fewer skilled human operators thanks to AI-powered automation.
Prime bundle (shipping + video + music + Alexa + Pharmacy + Gaming) is deeply differentiated and drives 200M+ loyalty.
Purchase intent data, AWS usage telemetry, and last-mile logistics operational data are genuinely irreplaceable. The AWS backlog reached $496B in Q2 2026 — double the $244B carried at end-2025 — which is contracted commitment, not pipeline. Both frontier labs are locked in bilaterally: $33B invested in Anthropic against $100B+ of committed AWS spend over 10 years, and $50B invested in OpenAI against an expanded agreement worth $38B plus $100B over eight years. With the May 2026 ASCS launch, Amazon now also ingests freight, warehousing, and parcel telemetry from non-Amazon shippers (P&G, 3M, Lands' End, AE) — feeding the same AI forecasting models that already power FBA, compounding the data advantage UPS/FedEx cannot match.
AWS GovCloud, DoD JEDI, HIPAA, FedRAMP, and financial services compliance create enormous switching friction.
Marketplace two-sided flywheel: more buyers → more sellers → better selection → more buyers. Self-reinforcing. ASCS extends the flywheel to non-marketplace shippers — every additional cubic foot routed through Amazon's network lowers unit cost for everyone, reinforcing the cited 25% transport cost advantage.
1-click purchasing habits, Prime subscription, and AWS embedded in the infrastructure of the global internet. ASCS now embeds Amazon directly into the supply chain operations of P&G, 3M, Lands' End, and American Eagle — moving the company from being a sales channel to being core logistics infrastructure for non-Amazon revenue streams. Switching cost rises with every integrated SKU.
AWS is the system of record for global cloud infrastructure; a $169B annualised run rate as of Q2 2026, growing 37%, and S3 stores more data than any competitor by a wide margin.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Dominant scale, switching costs (Prime), and cost advantage (Logistics). The Q2 2026 report (July 30) moved the single most important moat metric decisively: AWS's contracted backlog reached $496B, from $364B at the end of Q1 2026 and $244B at the end of 2025 — it has roughly doubled in two quarters, and it is contracted revenue, not pipeline. AWS itself grew 37% to $42.2B (a $169B annualised run rate) at a 39.4% operating margin, its fastest growth in 18 quarters, with the AI business and the custom-silicon business each past a $25B run rate growing triple digits. The backlog now rests on both frontier labs rather than one: Amazon invested $50B in OpenAI on February 27, 2026 as part of a $110B round, making AWS the exclusive third-party cloud distributor for OpenAI Frontier and expanding the November 2025 $38B compute agreement by a further $100B over eight years, including ~2 GW of Trainium3/Trainium4 capacity. Amazon has separately committed $33B total to Anthropic ($8B prior + $25B new deal at $380B valuation, announced April 20, 2026) — and Anthropic in turn committed $100B+ in AWS spend over the next 10 years, with 5 GW of capacity additions and access to Trainium2 through Trainium4. On May 4, 2026 Amazon launched Amazon Supply Chain Services (ASCS) — opening its full freight, distribution, fulfillment, and parcel network (≈80,000 trailers; 25% lower transport cost vs alternatives) to any business, regardless of whether they sell on Amazon. P&G, 3M, Lands' End, and American Eagle signed as anchor customers. BofA estimates a $1.3T 3PL TAM; each 1% share equals ~$13B annual revenue and the launch is being framed as the next AWS — externalising an internal capability into a high-margin platform. On June 10, 2026 Amazon expanded ASCS's LTL (less-than-truckload) freight service beyond inbound-to-Amazon shipments to serve any destination — third-party warehouses, distribution centers, retail partners, and distributors — building on a base that has already moved millions of pallets annually for tens of thousands of sellers and vendors since 2019, an early proof point that the platform is scaling as designed. ASCS was not quantified in the Q2 release or on the call — three months after launch there is still no disclosed revenue, customer count, or volume, so it remains optionality rather than an observed contributor. The competitive pressure from Azure and GCP that dominated the last two updates has narrowed rather than widened: at 37% AWS is no longer the slowest-growing hyperscaler, and Jassy told investors AWS is capacity-constrained through 2027 and could eventually be a trillion-dollar annual revenue business. What replaces it as the live risk is the bill — 2026 capex was raised to ~$220B from $200B on higher memory costs, TTM property-and-equipment spend is $169B (+64% YoY), and TTM free cash flow has swung to −$7.6B from +$18.2B. Governance tail risk also persists: a June 2026 report that Amazon's CEO raised security concerns with U.S. officials contributed to an export-control ban on Anthropic's Fable 5 and Mythos 5 models — a reminder that the frontier-lab relationships carry regulatory exposure alongside their commercial upside.
Growth Score
Q2 2026 (reported July 30) turned the Q1 re-acceleration into a trend. Revenue of $200.6B grew 20% YoY against $196.4B consensus, operating income rose 43% to $27.5B, and AWS grew 37% to $42.2B — nine points faster than Q1's 28%, six points ahead of the 31% expected, and the fastest AWS has grown in 18 quarters. AWS operating margin reached 39.4% (from 33.1% a year ago) and the contracted backlog hit $496B, roughly double the $244B carried at end-2025. The AI business and the chips business each passed a $25B run rate growing triple digits. Advertising accelerated to 26% ($19.8B). North America grew 16% ($116.2B) and International 15% ($42.2B). The offset is the capital bill: Q2 capex was $53.1B, full-year 2026 capex was raised to ~$220B from $200B on higher memory costs, and TTM free cash flow swung to −$7.6B from +$18.2B. Q3 guidance of $197–202B prints as only 9–12% YoY, but management attributes ~400bps of that step-down to Prime Day falling in Q2 this year and ~80bps to FX, leaving roughly 13–17% underlying. Net income of $62.6B ($5.75 diluted) is not a run rate — it includes $53.4B of non-operating pre-tax gains, primarily marking the Anthropic stake to the ~$965B June 2026 round.
Valuation Score
Q2 2026 (July 30) beat on every operating line — revenue $200.6B (+20%), operating income $27.5B (+43%), AWS $42.2B (+37%) at a 39.4% margin — and the contracted backlog roughly doubled in two quarters to $496B. The stock closed at $235.50 on July 30 and traded to ~$258 after the print. The ladder is raised for the first time since May: bear $195 → $205, base $310 → $330, bull $395 → $420, reflecting an AWS run rate nine points faster than the last revision assumed and a backlog that now underwrites several years of that growth. The raise is deliberately smaller than the beat, because the same quarter took TTM free cash flow to −$7.6B from +$18.2B and pushed 2026 capex to ~$220B: the earnings power is being bought, not compounded, until the depreciation cycle turns. At ~$258 the stock sits about a fifth below base, in the lower-middle of the corridor. Two things in the price are not operating income — the Anthropic stake, marked at roughly $145–200B (15–21% of a ~$965B June round) against a $2.80T market cap and heading for an IPO as early as October, and the reported $5.75 EPS, of which $53.4B of pre-tax gains is a mark, not a run rate.
The Economic Moat
Amazon possesses a Wide Economic Moat driven by four primary pillars:
- Cost Advantage: Its massive fulfillment infrastructure creates unit costs that no competitor can match, allowing for faster delivery and lower prices.
- Switching Costs: The Prime ecosystem locks in consumers. Once a household is integrated into Prime, the convenience makes shopping elsewhere a "costly" friction.
- Network Effect: The 3rd party marketplace creates a flywheel where more sellers attract more buyers, which attracts more sellers.
- AWS as the Primary AI Training Cloud: Amazon has now invested $33B total in Anthropic ($8B prior + $25B new, April 20, 2026) — locking in AWS as the exclusive primary cloud for training and deploying Claude models. Anthropic committed to spending $100B+ on AWS technologies over the next 10 years, adding 5 GW of capacity and using Trainium2 through Trainium4 (including future chip generations). Since February 27, 2026 the same structure exists with the other frontier lab: Amazon put $50B into OpenAI's $110B round, became the exclusive third-party cloud distributor for OpenAI Frontier, and expanded the November 2025 $38B compute agreement by $100B over eight years covering ~2 GW of Trainium3/Trainium4 capacity. Both are bilateral lock-ins — Amazon funds the lab, the lab is structurally dependent on AWS for training and inference — and together they are the visible cause of the backlog moving from $244B at end-2025 to $364B in Q1 2026 to $496B in Q2. That $496B is contracted, multi-year, and concentrated in the two labs least able to move: a demand floor no competing cloud can replicate, and the reason Jassy can describe AWS as capacity-constrained through 2027 rather than demand-constrained. The concentration cuts both ways — a material share of the backlog now sits with two private counterparties whose own funding is not yet self-sustaining.
- ASCS — Logistics-as-a-Service: On May 4, 2026 Amazon launched Amazon Supply Chain Services (ASCS), opening its end-to-end logistics stack — cross-border and domestic freight, bulk warehousing, fulfillment, and parcel — to any business, in any industry, on or off the Amazon marketplace. Anchor customers include P&G, 3M, Lands' End, and American Eagle, with the service explicitly targeting healthcare, automotive, manufacturing, and retail. Amazon cites up to 25% lower transport cost vs alternatives and a 20% conversion lift for fully-managed sellers, leveraging proprietary AI forecasting and a ~80,000-trailer fleet. BofA models a $1.3T 3PL TAM where each 1% share equals ~$13B in annual revenue (~+2% to retail revenue at 1% share by 2027). The strategic parallel is AWS in 2006 — externalising an internal capability into a high-margin platform business. UPS (-8.9%) and FedEx (-7.4%) sold off on the announcement, signalling that the market views this as a structural rather than incremental event.
Ten Moats Verdict
Amazon's moats are overwhelmingly AI-resilient, and Q2 2026 deepened the strongest one rather than the newest: the AWS backlog doubled to $496B in two quarters, custom silicon and the AI line each passed a $25B run rate growing triple digits, and both frontier labs are now contractually and financially bound to AWS. No moat status changes this quarter — the evidence under proprietaryData and systemOfRecord is materially stronger, but statuses were already strong. The ASCS thesis is unchanged and unconfirmed: three months after launch Amazon quantified none of it in its first opportunity to do so, so the logistics platform stays in the moat description as a durable structural argument and out of the growth drivers as a measured one. The live question is no longer whether the moats hold but what they cost — ~$220B of 2026 capex against TTM free cash flow of −$7.6B is the price being paid to keep them.
Alexa voice interface and AWS console are table stakes easily replicated; not a durable differentiator.
AI is commoditizing logistics routing and retail recommendation logic that once required years to build.
Amazon's 200M+ product reviews and buyer behaviour signals still provide meaningful aggregation advantages, but AI scraping and competitor datasets have eroded the edge. The moat persists in scale, not exclusivity.
AWS cloud operations require fewer skilled human operators thanks to AI-powered automation.
Prime bundle (shipping + video + music + Alexa + Pharmacy + Gaming) is deeply differentiated and drives 200M+ loyalty.
Purchase intent data, AWS usage telemetry, and last-mile logistics operational data are genuinely irreplaceable. The AWS backlog reached $496B in Q2 2026 — double the $244B carried at end-2025 — which is contracted commitment, not pipeline. Both frontier labs are locked in bilaterally: $33B invested in Anthropic against $100B+ of committed AWS spend over 10 years, and $50B invested in OpenAI against an expanded agreement worth $38B plus $100B over eight years. With the May 2026 ASCS launch, Amazon now also ingests freight, warehousing, and parcel telemetry from non-Amazon shippers (P&G, 3M, Lands' End, AE) — feeding the same AI forecasting models that already power FBA, compounding the data advantage UPS/FedEx cannot match.
AWS GovCloud, DoD JEDI, HIPAA, FedRAMP, and financial services compliance create enormous switching friction.
Marketplace two-sided flywheel: more buyers → more sellers → better selection → more buyers. Self-reinforcing. ASCS extends the flywheel to non-marketplace shippers — every additional cubic foot routed through Amazon's network lowers unit cost for everyone, reinforcing the cited 25% transport cost advantage.
1-click purchasing habits, Prime subscription, and AWS embedded in the infrastructure of the global internet. ASCS now embeds Amazon directly into the supply chain operations of P&G, 3M, Lands' End, and American Eagle — moving the company from being a sales channel to being core logistics infrastructure for non-Amazon revenue streams. Switching cost rises with every integrated SKU.
AWS is the system of record for global cloud infrastructure; a $169B annualised run rate as of Q2 2026, growing 37%, and S3 stores more data than any competitor by a wide margin.
Growth Analysis
Growth Drivers
Key Risk
Amazon raised 2026 capex to ~$220B from $200B on higher memory costs; TTM property-and-equipment spend is $169B (+64% YoY) and TTM free cash flow has swung to −$7.6B from +$18.2B. Those are observed facts and are charged in the base and in margin, not here. What is unmaterialised is the return: the $496B backlog and the >$25B AI and chips run rates must convert to revenue fast enough to absorb the depreciation from two consecutive $200B+ capex years as it reaches the P&L in 2027–2028. Because AWS is supply-constrained through 2027 on management's own account, nothing before then tests whether the demand is elastic at these prices or simply front-loaded — and a material share of the backlog sits with two private labs whose funding is not yet self-sustaining. If AWS settles back to the mid-20s while depreciation steps up, margins compress while revenue still grows.
Score Derivation
82.3 base + 2.7 trajectory + 4 margin − 5 risk = 84
Base 82.3 (16–21% blended CAGR) + 2.7 trajectory (AWS and advertising accelerating; ASCS stable) + 4 margin expanding (consolidated operating margin 13.7% vs 11.4% LY; AWS 39.4% vs 33.1%) − 5 moderate key risk (capex return, unmaterialised) = 84. Two changes offset each other. The CAGR midpoint rises a point for AWS going 28% → 37% with the backlog doubling, which is a measured series and belongs in the base. Against it, ASCS drops from accelerating to stable: three months after launch Amazon disclosed no ASCS revenue, customer count or volume, so there is no series to call it accelerating on. The severity term does not move, but it now carries a different risk — the Azure/GCP gap it was charged for has narrowed rather than widened, and the live unmaterialised risk is whether ~$220B of 2026 capex earns its depreciation.
Financial Outlook
Research Covering This Name
Meta's Capex Has One Receipt, and It Is Price Per Ad
Meta spent $31.1B in a quarter and printed $784M of free cash flow, and the tape traded the capex line. The line that tests the build is the other one: price per ad up 12% for a second straight quarter while impressions decelerated. Only the outcome owner can charge for a better match.
The Odds on Alphabet's $200B AI Build
Alphabet beat on revenue, on Cloud and on margin, then fell about 5% because capex went up. The market is sorting hyperscalers by how much they spend. The variable that decides whether a build pays off is how much of it was sold before it was poured.
Price Scenarios (12–24 Months)
Valuation Analysis
Using a 10-year DCF with a 9% WACC and a 3.5% terminal growth rate, incorporating Q2 2026 actuals (AWS +37% at a 39.4% margin, operating income $27.5B, backlog $496B), Q3 guidance ($197–202B, ~13–17% ex Prime Day timing and FX), ~$220B of 2026 capex with the depreciation drag landing in 2027–2028, and the Anthropic stake carried separately at a discount to its private mark, our fair value estimate for AMZN is $330/share.
Valuation Multiples
| Forward P/E (2026E) | ~33× |
| Forward P/E (2027E) | ~27× |
| PEG (2027E) | ~1.9× |
| Price / Sales (NTM) | ~3.3× |
| Anthropic stake | ~$145–200B |
Trailing GAAP P/E is not usable this quarter: the $53.4B non-operating gain that produced $5.75 of Q2 EPS is a mark on the Anthropic stake, not earnings. On consensus struck before the Q2 beat, AMZN trades at ~33× 2026E and ~27× 2027E — a real premium to the S&P 500 (~21×), and the PEG of ~1.9× is worse than it looked a quarter ago because 2026 EPS carries the depreciation from a capex base that grew 64%. The offsetting point is that both estimates predate a quarter in which AWS grew 37% against 31% expected and the backlog reached $496B, so the denominator is more likely to be revised up than the multiple is to compress. The Anthropic stake at ~5–7% of market cap is real but is not free — it is the same balance sheet funding $220B of capex, and an IPO would convert a mark into a price the market sets rather than a round does.
Approximate figures as of July 2026.
Where We Are vs Targets
Loading live price…
The $496B backlog converts more slowly and less profitably than contracted, AWS settles back toward the mid-20s as capacity finally arrives, and the depreciation from two consecutive $200B+ capex years lands on the P&L before the revenue does — margins compress while revenue still grows, and free cash flow stays negative through 2028.
- AWS decelerates toward the mid-20s once the 2026–2027 capacity build completes and the supply constraint that flatters current growth is removed; the 39.4% Q2 operating margin proves to be a peak struck before the new assets began depreciating
- The backlog's concentration turns into its weakness: a material share sits with two private labs whose funding is not yet self-sustaining, and contracted commitments are renegotiated, deferred, or restructured rather than defaulted
- Capex overshoots ~$220B on continued memory-cost inflation while TTM free cash flow stays below zero into 2028; the market stops paying a growth multiple for revenue bought with negative free cash flow and compresses AMZN toward 20–22× forward earnings
- ASCS never gets quantified — no revenue disclosure in Q3 or Q4 either — and the $1.3T TAM stays a slide; retail growth normalises to low double digits as the Prime Day timing benefit reverses
- The Anthropic mark reverses on a weak IPO or a down round, removing 5–7% of market cap that has nothing to do with operations
AWS holds the low-30s through 2027 as the $496B backlog converts on schedule, advertising sustains the mid-20s, the AI and custom-silicon lines compound past $25B each, and free cash flow turns positive again in 2027 as capex growth decelerates and the assets begin earning.
- AWS sustains 30–35% into 2027 on the contracted backlog, with the ~2 GW of Trainium capacity committed by OpenAI and Anthropic's 5 GW converting to revenue as it is delivered; AWS run rate passes $220B
- Advertising holds mid-20s growth from the $19.8B Q2 base toward a $90B+ annualised run rate — the margin engine that funds the capex cycle without diluting consolidated operating margin
- Consolidated revenue grows in the high teens as retail's mid-teens pace and AWS's low-30s blend; Q3's headline 9–12% is understood as a Prime Day timing artefact and growth reverts above 15% from Q4
- Capex plateaus near $220–240B in 2027 rather than compounding; free cash flow crosses back above zero during 2027 and approaches $40–50B in 2028 as the depreciation is matched by delivered AI revenue
- ASCS produces its first disclosed metrics and captures ~0.5–1% of the $1.3T 3PL TAM by 2028 — meaningful, but a rounding error next to AWS
AWS holds the high-30s past 2027 and the backlog keeps compounding faster than it converts, custom silicon becomes a genuine cost moat rather than a hedge, ASCS finally reports and re-rates as the third platform, and the Anthropic stake is monetised at or above its private mark.
- AWS sustains 35%+ and the backlog moves past $700B as sovereign AI and enterprise inference join the two frontier labs; Jassy's trillion-dollar-revenue framing starts being modelled rather than quoted
- Trainium3/Trainium4 reach cost-performance parity with the leading merchant accelerators at scale, converting the AI business from a capex sink into a structural margin advantage and lifting AWS operating margin above 40% through the depreciation step-up
- ASCS discloses revenue and scales toward 2–3% of the $1.3T 3PL TAM ($25–40B run rate) by 2028–2029; the market re-rates it on AWS-style multiples rather than 3PL multiples
- The Anthropic IPO (filed June 1, 2026, possible as early as October) prices at or above the ~$965B private mark, converting a balance-sheet estimate into a liquid holding worth $200B+ and drawing attention to a sum-of-the-parts the operating multiple ignores
- Free cash flow inflects hard in 2028 as capex growth stops and delivered capacity earns; a buyback at that point re-rates the whole equity toward the mid-30s on forward earnings