亚马逊：AWS在人工智能领域的地位如何？[第二部分]

U.S. Internet Amazon.Com Inc RatingOutperform Price Target AMZN300.00 USD Amazon: What place is AWS in for AI now? [Part II] Months in Internet are measured in hamster years. AWS sentiment was decidedly negative Close Date2 Feb 2026AMZN Close Price (USD)242.96Price Target (USD)300.00Upside/(Downside)23%52-Week Range258.60/161.38SPX6,976.44FYEDecDiv YieldNAMarket Cap (USD) (B)2,597.29EV (USD) (B)2,655.83 when we published Part I - under-allocated GPUs, no frontier model, who uses Trainium, and no splashy partners. A strong 3Q beat-and-raise + an OpenAI deal changed sentimentfor a minute, but the stock has since round tripped withAnthropic’s diversification deals andGoogle’s TPU traction once again questioning AWS’ place in the AI Cloud ecosystem. In Part II wedimensionalize AWS growth scenarios, and land in a compelling place.AWS to bring on more capacity than any hyperscaler over the next 2 years.We believe AWS can bring ~10 GW of capacity online through 2027, hitting the company’s goal of doubling capacity by then. AWS may well have a sustainable long-term build out advantagewith deep long-standing relationship across the supply chain and continued best-in-class non-AI compute infrastructure to complement. new cloud spending share, and we expect this lead to persist in 2026+. We’re in a pricingpower environment as hinted at in reported Anthropic cloud cost disclosures and AWS GPU price increases, which should also translate to AWS margin upside. A strong core with AI-adjacent upside.Our SSO work and CIO survey point to strong non-AI demand, and we expect accelerating AI adoption to drive adjacent storage and CPUdemand. We land atmid-20s AWS growth through 2027 with high-20s upside. Into the print:We view Azure’s revenue deceleration as idiosyncratic and tied to timing of new capacity and strategic allocation decisions.We remain convicted in accelerating 4QAWS revenue growththat should further accelerate through 2026. A sturdy AWS.Not in last place, not yet in first. Outperform, PT = $300. DETAILS PM SUMMARY The longer-term AWS’ place in the cloud market question remains, but steady beats & raises offer investors the best antidote to debating the unknowable.Answers to questions around the market for Trainium, the need for a leading frontier model, NVIDIA GPU allocation, gravity of cloud spend to follow AI workloads, and how to digest all the reported stakes in frontiermodels remain decidedly unknown. But here’s what we do know: We are in a supply-constrained environment for AI computeand AWS are world-class at bringing more compute capacity online -we expect AWS to bring on more incremental capacitythan any other hyperscaler over the next two years. If we’re right that AWS capacity adds translate directly into cloud revenues buoyed by pricing power, thenAWS should add more incremental net new cloud dollars than every other hyperscaler in 2026.Our proprietary SSO work alongsidepositive CIO survey results pointed to continued strength in core cloud compute spend, alongside significant AI revenue acceleration inclusive of Anthropic spend ramping with Project Rainier availability. In this note we detail the various vectors ofAWS revenue growth, andwe take up our AWS estimates to ~23% Y/Y growth in 4Q25 accelerating meaningfully to~25% growth in 2026 and ~26% growth in 2027. In an ultra-bull case scenario that pushes pricing, AI-software uptake, andGPU allocation to max levels, we can see a pathway to 30%+ Y/Y growth. A strong and durable AWS may just be what investorsneed to see to size up AMZN building upon a solid cloud beat and raise in 3Q25. concerningly retail media ad dollar attach rates may cap AMZN’s multiple, but that’s a debate for a different day. For what it’sworth, we continue to like Amazon’s odds in winning on-platform AI agentic shopping given early Rufus strength, the power ofthe 5-star review, and ingrained consumer behavior.We remain Outperform with a $300 price target on AMZN and viewthe stock as one of our top picks in Internet this year. We value Amazon using a 50/50 combination of SOTP method whichvalues Amazon’s largest segment (Retail) at an unchanged 2026E EV/Sales multiple of 3.0x, and a DCF using a WACC of 9%and a terminal growth rate of 3.5%. We revise up our 2026 Sales estimate from $808B to $812B, and our 2027 Sales estimate CAPACITY CONSTRAINTS, CAPACITY BUILD OUT, AND INELASTIC SUPPLY… FOR NOW Before 3Q25 earnings, we held passionate debates over bps of AWS revenue growth with the bogey arbitrarily set at 20% Y/Ygrowth. And even then, we often heard something along the lines of “why do I need to bother debating basis points when Azure and GCP can offer me double the growth rate?” Indeed, a market leader ceding share is never an easy stock to get behind. But aquarter later, and the debate appears quite different - armed with a better understanding of current supply-demand dynamics,we’re now asking just how high AWS growth can go, and whether they c