Amazon's custom silicon hits USD $25 billion run rate

Amazon's custom silicon business has exceeded a USD $25 billion annual revenue run rate. The figure covers chips developed for Amazon Web Services and marks a significant scale for the group's in-house hardware effort.

The disclosure provides a rare financial marker for one of the cloud market's most closely watched areas, as large technology companies invest heavily in processors designed for artificial intelligence and data centre workloads. It also shows how far Amazon has gone in building a business around chips it designs itself, rather than relying only on third-party suppliers.

Custom silicon has become a strategic battleground for cloud providers as demand for AI computing rises. By designing chips internally, companies aim to manage costs, improve efficiency for specific workloads and reduce dependence on a small group of external semiconductor vendors.

Amazon's chip work now spans several AWS product lines, including processors for general cloud computing and specialised chips for machine learning. The group has positioned those products as part of a broader effort to give customers more choice over the infrastructure they use to train and run AI models.

The USD $25 billion run-rate figure does not necessarily mean the business has generated that amount in booked annual revenue, but it does indicate the pace of sales the operation has reached over a given period. In financial terms, annual revenue run rate is commonly used to measure current scale rather than a full-year reported total.

Rivals are pursuing similar strategies. Microsoft, Google and Meta have all increased investment in internally designed chips as AI demand puts pressure on capacity, pricing and supply chains. Nvidia remains the dominant supplier of AI processors, but the largest cloud groups have sought alternatives for parts of their infrastructure and customer offerings.

For Amazon, the development is also notable because custom silicon has been a long-running project that predates the recent boom in generative AI. AWS has spent years building a portfolio of chips for different tasks, from handling standard cloud workloads to supporting training and inference for AI applications.

AI pressure

The surge in AI spending has intensified scrutiny of whether cloud operators can secure enough computing resources while keeping margins under control. Specialised chips are central to that equation because they shape both the cost of serving AI products and the economics of renting computing capacity to customers.

Demand for training large language models and running them in production has pushed data centre operators to commit vast sums to infrastructure. Investors and customers alike are watching to see whether that spending translates into differentiated services, lower operating costs or stronger bargaining power with suppliers.

Amazon's announcement suggests its silicon effort has moved well beyond an experimental or supporting role inside AWS. A business running at that scale would be a substantial operation in its own right, even within a company of Amazon's size.

Amazon has previously argued that designing chips around specific workloads can improve price and performance for customers using its cloud platform. That case has become more important as businesses weigh the high cost of adopting AI systems and look for ways to manage spending on compute-intensive applications.

Customer impact

Customers are likely to view the milestone through two main lenses: availability and cost. If Amazon can steer more workloads to its own processors, it may be better placed to offer alternatives when demand for third-party AI chips outstrips supply, while also exerting more control over pricing across parts of its cloud stack.

The move could also influence software development choices. As cloud providers deepen investment in proprietary hardware, customers often face a trade-off between using tools tailored to a specific provider's infrastructure and preserving the flexibility to move workloads elsewhere.

That tension has grown as AI services become more tightly integrated with the hardware on which they run. Companies buying cloud services want access to the newest processors, but many also want to avoid dependence on a single vendor's ecosystem.

Amazon's position in cloud computing gives it a large installed base through which it can sell those chips indirectly via AWS services. That distribution advantage differs from the model used by traditional semiconductor companies, which generally sell hardware directly to equipment makers, cloud groups or enterprises.

Long build-out

The milestone also reflects a broader shift in the economics of the cloud sector. In earlier phases, competition centred largely on data centre scale and basic computing services, but the rise of AI has increased the value of specialised infrastructure, including networking, storage and chips tuned for particular workloads.

Building a silicon business at this level requires long-term investment in design, engineering and manufacturing partnerships. It also requires enough customer demand to justify integrating those processors into a wide range of cloud services.

For AWS, that effort has become part of a wider contest over who captures the profit pool created by AI. If cloud companies can bring more of the hardware stack in-house, they may retain a larger share of spending that might otherwise flow to external chip makers.

At the same time, success in custom silicon does not eliminate reliance on outside suppliers. Even the biggest cloud operators continue to buy large volumes of chips from established semiconductor companies while developing and deploying their own designs.

The USD $25 billion annual revenue run rate nevertheless signals that Amazon's chip strategy is no longer a niche infrastructure project inside AWS, but a large commercial business tied directly to the shape of competition in cloud computing and artificial intelligence.