Investing.com -- Amazon Web Services (AWS) is recording its fastest growth rate in nearly five years, powered by a massive surge in artificial intelligence workloads across both frontier research labs and traditional enterprise sectors, AWS CEO Matt Garman revealed in an interview on Bloomberg TV’s Bloomberg Tech.

Speaking with Bloomberg host Ed, Garman outlined how the cloud computing giant is navigating unprecedented demand, expanding its custom silicon offerings, and making multi-billion-dollar infrastructure investments to keep pace with customer needs.

Below is a breakdown of the core details and developments outlined during the interview.

1. Broad-Based Growth Beyond Frontier Labs

While headline-grabbing AI labs like OpenAI and Anthropic are driving significant model training on AWS, Garman emphasized that the growth is widespread. Startups and enterprise clients in financial services, healthcare, retail, and media are adopting AI to streamline operations and create new customer experiences. Unlike competitors who may rely heavily on a small handful of massive clients, AWS’s expansion is distributed across a broad customer base.

2. A $25 Billion AI Business Shift Toward "Inference"

Garman noted that AWS’s AI business currently operates at a $25 billion revenue run rate. This figure encompasses model training from tech giants, but an increasingly large portion comes from inference—running existing models via services like Amazon Bedrock to power real-world applications and agentic workloads.

Garman highlighted that customer spend is steadily shifting toward inference, where direct business value is generated for end customers.

3. Capacity Shortages and Massive $220 Billion Capex

Addressing Amazon’s overall capital expenditure—which stands at $220 billion this year, up by $20 billion—Garman confirmed that AWS will continue heavy capital spending next year. Demand continues to significantly outstrip supply, forcing AWS to secure long-term agreements:

• Capacity Commitments: Much of AWS’s capacity is already spoken for through the end of 2027 and well into 2028.




• Long-Term Contracts: Customers are signing five-year commitments to secure their compute needs.

4. Custom Silicon Strategy: Trainium and Graviton

The $25 billion run rate for AWS’s chip business stems from renting out capacity powered by its in-house processors, rather than selling chips outright.

• Trainium Demand: AWS’s Trainium capacity is largely sold out through the end of next year.




• Cost Efficiency: By controlling the full hardware and software stack, AWS enables customers to save 20% to 30% on inference costs using Trainium compared to standard market options.




• Coexistence with Nvidia: AWS remains one of Nvidia’s largest customers and continues to offer Nvidia GPUs alongside Trainium to give clients maximum hardware choice.




• Future Direct Sales: While AWS currently rents out capacity, Garman noted the company might consider selling chips outright to third parties in the future.

5. Open-Weight AI Models and Regulation

Explaining AWS’s decision to sign a recent open-weights letter, Garman advocated for a balanced approach to government oversight:

• Level Playing Field: Regulatory frameworks should apply consistently across both closed frontier models and open-weight models to avoid over-legislating innovation.




• Future Monetization: Garman predicted that open-weight model creators will increasingly move toward licensing structures when their models are deployed in commercial cloud environments, as developers look to monetize their intellectual property.

When & Outlook

The interview captures AWS at a pivotal operational moment—experiencing record-setting demand while racing to scale its physical data centers and custom chips. With compute capacity booked years in advance, AWS plans to sustain elevated capital expenditure levels into next year to construct the infrastructure necessary to satisfy global enterprise demand.