Amazon Sets Global Record for Capital Expenditure
Amazon has set a new world record for capital expenditure, driven by massive investments in AWS infrastructure and new AI data centers. The spending reflects a broader hyperscaler race to scale capacity and secure compute supply for anticipated AI demand.
- Amazon's projected capital expenditure for 2026 is approximately $200 billion, a significant increase from the $131.8 billion spent in 2025 and a figure that surpassed analyst expectations by about $50 billion. - The spending is part of a larger hyperscaler investment race, with Google projecting $175-$185 billion and Meta planning $115-$135 billion in 2026 capital expenditures to build out AI infrastructure. - A key driver of the expenditure is AWS's vertical integration strategy, heavily investing in its custom silicon, including Trainium for AI training and Inferentia for inference, developed by its acquisition Annapurna Labs. - The business for AWS's in-house chips, including the general-purpose Graviton and AI-focused Trainium, has already surpassed a $10 billion annual revenue run rate. - AWS positions its custom chips as a more cost-effective alternative to third-party hardware, with Trainium2 instances priced at roughly half of comparable NVIDIA H100 instances. - CEO Andy Jassy stated the investment is primarily for AWS, driven by demand for both generative AI and surprisingly strong growth in core, non-AI enterprise cloud workloads. - Despite the record spending, AWS CEO Matt Garman anticipates being "capacity constrained for the next couple of years," indicating they will sell all available server capacity and still have unmet demand. - Major AI companies are leveraging this custom hardware; for example, Anthropic uses a cluster of nearly 500,000 Trainium2 chips to train its Claude foundation models.
