Amazon Cloud infrastructure capacity largely sold out through 2027, CEO Jassy Reveals
Amazon Chief Executive Officer Andy Jassy announced that Amazon Web Services (AWS) has completely sold out its cloud computing capacity for 2026 and largely sold out available allocations through 2027. Speaking on an Andreessen Horowitz podcast alongside Ben Horowitz, Jassy revealed that multi-year enterprise demand for artificial intelligence infrastructure has also claimed a significant portion of planned capacity for 2028, despite Amazon's record capital expenditure budget.
The supply constraint underscores an unprecedented bottleneck across global data center capacity as cloud providers race to deploy specialized servers to support generative AI training and inference at scale.
Massive Capital Outlays and Multi-Year Customer Commitments
To meet the surging demand, Amazon is deploying approximately $220 billion in capital expenditures in 2026 to scale its global data center footprint and custom server hardware. Despite these massive outlays, long-term multi-year commitments from frontier AI research labs and corporate enterprises have fully absorbed available data center power and rack capacity years in advance.
Key capacity allocation details outlined by Jassy include:
2026 and 2027 Lockout: Large enterprise blocks for 2026 are completely sold out, with the vast majority of planned 2027 capacity fully booked by major AI research and enterprise clients.
Early 2028 Commitments: A substantial chunk of projected 2028 cloud capacity has already been claimed under forward-looking enterprise procurement agreements.
On-Demand Buffer: AWS continues to reserve a targeted pool of computing resources to support flexible, on-demand customer workloads through services like Amazon Elastic Compute Cloud (EC2) and the Amazon Bedrock generative AI platform.
Addressing High Inference Costs with Custom Silicon
Jassy emphasized that high operational costs remain a central challenge for widespread AI adoption, noting that model inference, generating real-time outputs from trained models, is still roughly ten times too expensive for pervasive enterprise deployment.
To reduce compute expenses, Amazon is heavily promoting its custom-designed Trainium AI accelerators and ARM-based Graviton processors, which deliver 30% to 40% better price-performance compared to market alternatives.
Amazon disclosed over $225 billion in future custom silicon commitments, anchored by key enterprise deals with leading AI developers including Anthropic and OpenAI.
Industry Implications Ahead of Q3 Financial Results
The capacity constraints highlight how structural demand for AI compute is outpacing global power grids, custom silicon manufacturing, and data center construction timelines. With AWS generating $42.2 billion in second-quarter revenue, markets are closely watching Amazon’s third-quarter earnings announcement on October 29 for further details regarding capital spending, capacity expansion timelines, and custom silicon deployments.
