Aws Buildout Targets 2027-28, Cloud Still <20%

Context:

In a CNBC Fortt Knox interview on February 17, 2026, AWS (Amazon Web Services) CEO Matt Garman said Amazon is expanding data centers and power for 2027-28 as AI (artificial intelligence) demand keeps ramping. He noted that less than 20% of global workloads - apps, databases, analytics pipelines, and AI training and inference jobs - have moved to the cloud, leaving a long runway for migration.

Constraints:

Garman was blunt: "We will be capacity-constrained for the next couple years." That means compute accelerators - high-end GPUs (graphics processing units) and AWS-designed chips such as Trainium and Inferentia - plus rack space and power will be tight even as Amazon builds new capacity. See The Motley Fool for more.

Spending:

Garman did not give a new dollar figure during the segment. For context, Amazon guided to roughly $100 billion of capital expenditures in 2025, largely tied to AWS and AI, and has since signaled about $200 billion in 2026 for the company, with most of that going to cloud infrastructure. Rivals are also investing heavily: Microsoft and Alphabet flagged about $80 billion and $75 billion for 2025, respectively. See CNBC for the Amazon guidance.

Europe sovereign cloud:

AWS's European Sovereign Cloud offers a physically and logically separate cloud operated in the EU (European Union) for customers with strict regulatory or data-residency needs. Key features include:

• EU-only operations and governance, with an independent advisory board and a dedicated EU Security Operations Center.
• A separate control plane, EU-only DNS (Route 53), and an EU-only certificate authority.
• Full EU data residency, including metadata, billing, and IAM (Identity and Access Management).

In short, the service provides independent governance and operational separation, staffed by EU-resident personnel with no non-EU control. See the Amazon press release for details.

Advice for builders:

• Lock capacity early: use Reserved Instances or committed-use plans to secure GPUs and accelerators and to stabilize pricing. These options buy or reserve capacity ahead of time so you avoid being stuck in line at the last minute.
• Design for scarcity: build multi-region or hybrid architectures to avoid single-region bottlenecks. Pre-compute work where it's cheap, and run model inference - generating outputs from a trained model - closer to users when low latency matters.
• Stay EU-sale-ready: for regulated customers, map which workloads need the sovereign cloud's EU control plane, EU-only operations, and data residency guarantees.

Bottom line:

AWS is playing the long game on AI infrastructure. Capacity will stay tight, prices are unlikely to fall quickly, and the cloud migration wave is still early. Plan now, or expect to wait in line later.