Wednesday, November 18, 2026
AI 인프라 · 뉴스 & 분석
논평

AWS Blackstone Sale-Leaseback, November 2026

Blackstone Digital Infrastructure Trust paid $5 billion for a majority stake covering 13 AWS data-center shells — leased back to Amazon — a move Omdia analysts say may be partly economic in motivation as AI construction weighs on hyperscaler balance sheets; Amazon's own capex reached $131.82 billion in fiscal 2025.

설비투자 · SEC 공시 기준전체 이력 →
최근 회계연도 설비투자
$131.82B (FY2025)
전년 대비
58.8% · $83.00B → $131.82B
투자 / 매출
18% (FY2025, $716.92B)
최고 기록 기간
$131.82B · 2025-12-31

The late-2026 disclosure that a Blackstone joint venture had acquired AWS data-center campuses and leased them back to Amazon crystallized a tension building inside every major hyperscaler's finances: the relentless capital demands of the AI era may be prompting even the world's largest cloud operator to rethink which assets belong on its own balance sheet. Blackstone Digital Infrastructure Trust's $5 billion acquisition of a majority stake in the AWS joint venture — covering 13 newly built, power-ready data-center shells — is the most concrete transaction yet suggesting that Amazon is willing to separate real-estate ownership from infrastructure operations as it pursues a build-out of unprecedented scale.

The financial pressure behind the move is quantifiable. Amazon's capital expenditure reached $131.82 billion in fiscal year 2025, a 58.8% jump from $83.00 billion the prior year, consuming 18% of the company's $716.92 billion in revenue — the highest absolute capex figure among hyperscalers tracked by Slicast's SEC XBRL compilation. That rate of spending, sustained while demand still far outpaces supply, creates natural pressure to optimize the form of capital deployed. CEO Andy Jassy stated in October 2026 that AWS capacity is "largely" sold out through 2027, with much of 2028 already pre-committed — excluding EC2. Omdia analysts, as reported, suggested the Blackstone arrangement may be partly economic in motivation, with AI construction weighing on hyperscaler balance sheets. By placing shell real estate in a sale-leaseback structure, Amazon can in principle direct its own capital toward compute, networking, and power assets that more directly generate cloud revenue without slowing the physical build-out.

That build-out is vast and accelerating on multiple fronts. AWS has announced a $4-billion-plus investment to establish its first cloud region in Chile, committed $5.3 billion toward a 50-megawatt deployment at Saudi Arabia's AI Zone targeting a 2028 power-on date, and has reportedly acquired a campus with direct connection to the 2.5-gigawatt Susquehanna nuclear facility as the company hunts for large, stable power supplies. Plans for a 36-building data-center campus in Indiana County, Pennsylvania have been disclosed alongside a 235-megawatt development approval in Sydney, Australia. A 420-terabit-per-second trans-Pacific fiber cable from Japan to Washington state is under construction to handle growing AI traffic volumes. On the silicon side, a reported multi-generational agreement with Qualcomm — valued at up to $60 billion — would supply custom inference accelerators to AWS data centers; AWS has simultaneously added Nvidia GPU partitioning to SageMaker inference endpoints. Enterprise commitments are accumulating in parallel: Verizon signed a $2 billion AWS AI infrastructure agreement in October 2026, and a planning notice suggests the UK Home Office's cloud-services contract renewal could be worth roughly £1 billion to AWS.

Yet the picture is not without complications. AWS CEO Matt Garman noted in November 2026 that AI inference still represents less than 1% of its eventual scale — a growth argument, but also an acknowledgment that monetization remains early-stage relative to the capital already committed. Amazon settled a legal dispute in October 2026 against AWS managers accused of accepting kickbacks in data-center procurement — a reminder that large-scale, high-velocity procurement carries governance risk. The company also moved to drop non-disclosure agreements on data-center development projects and pledged $1 billion in community investment across five years, reducing local political friction at the cost of greater competitive visibility. Separately, AWS reportedly instructed clients to abandon Middle East cloud assets following drone strikes on facilities in Bahrain and the UAE, with no recovery timeline provided — an illustration that geopolitical risk is now a live variable in infrastructure planning at this scale.

The balance of signals for the coming quarters remains unresolved. Demand visibility is as strong as any hyperscaler has publicly described: pre-sold capacity through 2027, a deepening enterprise AI pipeline, and Garman's framing of inference as a runway still in its opening meters. The Blackstone sale-leaseback, if it becomes a template, offers a mechanism to accelerate physical deployment without proportionally expanding Amazon's own asset base — a structurally important option as capex intensity remains elevated. Three signals merit close monitoring: whether Blackstone's model is extended to additional AWS campuses or replicated by peer hyperscalers, which would indicate sale-leaseback has become a structural financing tool for the AI build-out broadly; the cadence at which Qualcomm's custom inference silicon moves from contract to production deployment, since AWS's silicon diversification thesis depends on execution at scale; and how AWS navigates the power constraint — the reportedly acquired Susquehanna campus and the appointment of a dedicated energy policy manager suggest that energy supply, not compute alone, may be the binding constraint on the next phase of growth.

Based on 117 archived reports · AWS / Amazon →
AWS Blackstone Sale-Leaseback, November 2026 · Slicast