Amazon AWS Cloud Capacity Sold Out Through 2027 as AI Demand Surges

Amazon Web Services (AWS) is facing a growing shortage of available cloud computing capacity as demand for artificial intelligence infrastructure accelerates. CEO Andy Jassy said on October 23, 2026, that AWS’s capacity is largely sold out through 2027, with a substantial portion of its 2028 capacity already committed to customers. The comments highlight the scale of demand from AI developers and businesses building and deploying increasingly resource-intensive applications.

The development comes as Amazon plans to spend approximately $220 billion on capital expenditure in 2026. Despite this investment, the company is struggling to expand infrastructure quickly enough to meet customer requirements. The situation underscores a central challenge for the AI industry: building advanced models and applications is not just a software problem but also depends on data centers, specialized chips, electricity, networking equipment and the ability to bring new computing capacity online.

Key Takeaways

• AWS capacity is largely booked through 2027: Amazon CEO Andy Jassy said significant customer commitments extend into 2028.
• AI is driving demand: Research laboratories and enterprise customers need more computing power to train and run AI models.
• Amazon plans major investment: The company has indicated approximately $220 billion in capital expenditure for 2026.
• On-demand capacity remains available: AWS continues to reserve some resources for on-demand services, so the announcement does not mean every instance or service is unavailable.
• AI infrastructure is becoming a competitive factor: Cloud providers must expand computing capacity while managing construction, equipment, power and investment costs.
• Amazon’s next earnings update: The company was scheduled to report its third-quarter 2026 results on October 29.

What Did Amazon CEO Andy Jassy Announce?

During a video podcast with Andreessen Horowitz co-founder Ben Horowitz and general partner Erik Torenberg, Jassy said AWS was struggling to keep pace with demand from AI laboratories and enterprise customers.

He explained that capacity for 2026 was already sold out and that 2027 was largely booked. A significant amount of capacity for 2028 had also been committed.

Jassy added an important qualification: Amazon continues to reserve some capacity for on-demand services, including Amazon Elastic Compute Cloud (EC2) and Amazon Bedrock. Therefore, the announcement concerns the availability of large blocks of computing capacity rather than a complete shutdown of new customer access.

The comments suggest that demand is extending beyond short-term cloud usage. Large AI customers often need substantial computing resources for extended periods, making long-term capacity commitments increasingly important to cloud providers.

For AWS, the challenge is to fulfill existing commitments while bringing enough additional infrastructure online to accommodate new customers and growing workloads.

Why Is AI Demand Consuming So Much Cloud Capacity?

AI applications require different amounts of computing power depending on their purpose. Training a large model can involve enormous quantities of specialized processors working together, while serving an AI model to millions of users requires infrastructure capable of handling continuous inference requests.

Demand is also expanding beyond the companies that build foundation models. Banks, retailers, software companies and other enterprises are experimenting with AI assistants, automated workflows, customer-service tools and internal productivity applications.

These workloads can require computing resources even after a model has been trained. Every user request, generated response or automated agent action may consume additional processing capacity.

Three factors are particularly important:

• Model training: Developing and improving large AI models requires clusters of accelerators, high-speed networking and substantial power.
• Inference at scale: Running models for users and businesses creates ongoing demand for computing resources.
• Agentic AI: AI agents that use tools, execute code and complete multi-step tasks may generate multiple model requests for a single user objective.

As businesses move AI projects from experiments into production, cloud demand can become more sustained and predictable. This helps explain why major customers may reserve capacity years in advance.

Amazon’s $220 Billion Capital Spending Plan

Amazon’s planned capital expenditure of approximately $220 billion for 2026 illustrates the scale of infrastructure investment required to meet the AI boom.

The spending supports a broad infrastructure strategy, including data centers, servers, networking equipment and other assets needed to operate cloud services. Not all of this expenditure should be interpreted as spending exclusively on AI; Amazon also operates a large retail, logistics and technology business.

Jassy indicated that additional investment could make sense while demand remains strong. His comments reflect the opportunity AWS sees in selling computing services to customers that increasingly rely on cloud infrastructure for AI development and deployment.

However, spending more money does not instantly create usable computing capacity. Data centers require suitable sites, electrical connections, cooling systems, equipment procurement and construction. Specialized chips and networking components must also be delivered and integrated before new capacity can serve customers.

The result is a timing challenge: customer demand can rise faster than the physical infrastructure required to support it can be built.

What Does “Sold Out Through 2027” Actually Mean?

The phrase may sound as if customers can no longer use AWS, but that is not what Jassy described.

AWS offers a wide range of services, regions, instance types and purchasing arrangements. Availability can differ according to the customer’s location, required hardware, workload and contract.

Term	Meaning
Capacity largely sold out	Much of the relevant computing capacity has already been committed to customers
Long-term capacity commitments	Customers have reserved resources for future periods
On-demand capacity	Resources customers can provision when needed, subject to availability
Amazon EC2	AWS service that provides virtual computing resources
Amazon Bedrock	AWS service for building generative AI applications using supported foundation models

A business seeking a particular accelerator, region or large block of capacity could face different constraints from a company running a smaller application on conventional cloud instances.

The practical impact will therefore vary by customer. The announcement signals pressure on the overall infrastructure pipeline, not universal unavailability across every AWS product.

How AWS Capacity Constraints Could Affect AI Companies

Limited capacity can influence how AI developers plan their products, negotiate cloud contracts and choose infrastructure providers.

For large AI laboratories, access to sufficient computing resources is essential to training models and serving growing numbers of users. Capacity constraints may require earlier planning, longer commitments or the use of multiple infrastructure partners.

Enterprise customers could also face challenges when deploying AI applications at scale. A successful pilot may require considerably more computing power once thousands of employees or customers begin using it.

Possible consequences include:

• Greater competition for specialized computing resources.
• More emphasis on long-term infrastructure agreements.
• Increased interest in alternative cloud providers and dedicated infrastructure.
• Greater attention to model efficiency and reducing the cost of each AI request.
• Longer planning horizons for projects that depend on specific hardware.

These are potential effects of capacity pressure, not confirmation that AWS customers are universally experiencing delays or service disruptions.

What This Means for the Cloud Computing Industry

AWS is one of the largest cloud infrastructure providers, but it is not the only company investing heavily to serve AI workloads. Microsoft Azure, Google Cloud and specialized AI infrastructure providers are also competing for customers that need large quantities of computing power.

When demand outpaces the supply of available infrastructure, providers face a difficult balance. They must invest enough to capture future growth without building so much capacity that expensive equipment sits unused if demand later slows.

Jassy has previously emphasized the importance of managing this risk. Reporting by The Information noted his comments that Amazon can adjust some purchases of servers and networking equipment closer to when the capacity is needed, offering a way to manage uncertainty after data centers have been prepared.

The broader industry is also confronting constraints beyond financial capital. Access to electricity, grid connections, cooling, construction capacity and advanced semiconductor supplies can determine how quickly new computing facilities become operational.

Consequently, cloud competition is increasingly shaped by the ability to deliver reliable infrastructure at scale, rather than simply offering more AI models or software features.

The Bigger Picture

Amazon’s capacity announcement illustrates the infrastructure demands created by the transition from experimental AI projects to large-scale deployment. Training advanced models attracts attention, but serving those models to users and businesses creates a continuing requirement for processors, networking, electricity and physical data-center capacity.

For AWS, strong demand creates a significant commercial opportunity, provided the company can deliver the resources customers have reserved. For the wider AI industry, the capacity crunch highlights the importance of efficient models, optimized inference and infrastructure planning. Better software efficiency can reduce the resources needed for individual tasks, but it cannot eliminate the need for substantial physical computing infrastructure.

FAQs

1. Is Amazon AWS sold out through 2027?

CEO Andy Jassy said AWS’s cloud computing capacity is largely sold out through 2027, with a significant portion of 2028 capacity already committed. He also said AWS continues to reserve some capacity for on-demand services.

2. Why is AWS running short of capacity?

Demand from AI research laboratories and enterprise customers is increasing the need for computing infrastructure used to train and run AI models. Expanding capacity takes time because it requires data centers, specialized chips, power and networking equipment.

3. How much does Amazon plan to invest in 2026?

Amazon indicated capital expenditure of approximately $220 billion for 2026. This supports its broader infrastructure and business needs and should not be interpreted as spending exclusively on AI.

4. Does this mean businesses cannot use Amazon EC2 or Bedrock?

No. Jassy said Amazon continues to reserve some capacity for on-demand services, including EC2 and Bedrock. Actual availability depends on the service, region, hardware requirements and customer workload.

5. Could the capacity shortage benefit other cloud providers?

It could create opportunities for competing providers if customers cannot obtain their preferred capacity from AWS. However, the effect will depend on available resources, pricing, regional requirements and how quickly competing providers can deliver suitable infrastructure.

Looking Ahead

AWS’s ability to convert customer commitments into operational capacity will be an important measure of its execution over the coming years. Amazon’s upcoming earnings report may provide additional context on capital spending, infrastructure expansion and cloud growth, although the precise timing of new capacity becoming available will depend on construction and equipment deployment.

For the AI industry, the next phase of competition will depend on both infrastructure expansion and more efficient use of computing resources. Companies that can secure reliable capacity, reduce inference costs and deliver useful AI products at scale will be better positioned to meet demand as businesses continue adopting artificial intelligence.