Oracle Adds Cerebras to Cloud AI Lineup
Oracle will offer cloud services built on Cerebras Systems’ AI computers in its data centers beginning in 2027, the two companies said. The arrangement gives Cerebras another prominent customer and places its wafer-scale hardware inside Oracle Cloud Infrastructure, one of the largest enterprise cloud platforms. Oracle did not disclose how much Cerebras hardware it will deploy or the financial terms of the agreement.
The service is slated to become available in 2027, according to the companies. Oracle Cloud Infrastructure president Mahesh Thiagarajan said in a statement that “AI inference demands choice,” a comment that frames the deal as part of a broader strategy to give customers multiple options for running trained models. The announcement was made at Oracle’s annual AI World conference, where the company also outlined deals with Nvidia and Intel and noted that it buys Arm’s own chips.
Key Facts at a Glance
- Oracle will offer cloud services running on Cerebras AI computers in its data centers from 2027.
- Cerebras went public in May and is expected to nearly triple revenue next year to almost $3B, from less than $1B this year.
- Oracle did not say how much Cerebras hardware it will use.
- Cerebras already counts OpenAI and Amazon Web Services as customers.
- Cerebras shares rose about 9.5% on Tuesday after the announcement.
- Oracle announced the Cerebras service at its AI World conference alongside deals with Nvidia and Intel.
- Oracle plans to spend $70B on capital projects in the year to May 2027, after spending $55.7B the year before.
What Cerebras Brings to the Table
Cerebras builds each chip from a whole silicon wafer, the disk that is normally cut into many smaller processors. By keeping the wafer intact, Cerebras says its systems can move data across a much larger surface area and answer AI requests much faster than rivals. The company argues that this design is especially well suited to inference, the process of running trained AI models and generating responses for users. Inference has become a critical battleground in the AI hardware market because more companies are moving from experimentation to production-scale deployments.
The technical pitch matters because AI inference workloads are different from training workloads. Training often requires massive clusters working in parallel for weeks or months, while inference must respond to individual user requests with low latency and predictable throughput. Cerebras has positioned its wafer-scale engines as a way to reduce the number of chips and the amount of networking required to serve AI models. If Oracle can integrate those systems into its cloud, enterprise customers may gain another option for running demanding inference workloads without relying exclusively on graphics processors from Nvidia.
Why Oracle Wants More AI Options
Oracle has been building out its cloud infrastructure business with an emphasis on AI capacity. The company has signed a series of agreements to secure hardware and services from multiple chip suppliers, including Nvidia, Intel and Arm. That multi-vendor approach is intended to avoid lock-in and to give customers access to different performance profiles and price points. Adding Cerebras to the mix extends that strategy into wafer-scale computing, a niche but technically distinctive segment of the AI accelerator market.
Oracle’s capital spending plans underscore how aggressively it is expanding data center capacity. The company plans to spend $70B on capital projects in the year to May 2027, after spending $55.7B the year before. That level of investment reflects the demands of AI workloads, which require specialized power, cooling and networking infrastructure. The Cerebras service will therefore be one piece of a much larger buildout, and its success will depend on whether Oracle can translate the hardware’s technical advantages into reliable, cost-effective cloud offerings.
Revenue Growth and Market Skepticism
Cerebras is expected to nearly triple its revenue next year to almost $3B, from less than $1B this year. The growth projection suggests that demand for its systems is accelerating, even as the company remains much smaller than Nvidia. Nvidia’s products bring in close to $100B in sales each quarter, a scale that illustrates the gap between the market leader and its challengers. Investors are therefore weighing Cerebras’s rapid growth against the risk that it may remain confined to a small niche.
Cerebras went public in May, and its shares rose about 9.5% on Tuesday after the Oracle announcement. The stock reaction indicates that investors see the deal as a meaningful validation, though not necessarily a decisive shift in the competitive landscape. The company already counts OpenAI and Amazon Web Services as customers, which gives it credibility among large AI buyers. Still, the Oracle agreement does not disclose deployment volumes, so it is difficult to estimate how much revenue it will generate or how quickly the service will scale.
The Inference Battlefield
Much of the AI hardware discussion has focused on training large models, but inference is where many businesses expect to spend increasingly large sums over time. Every query to a chatbot, every recommendation, and every automated decision requires inference compute. As models become more capable and more widely deployed, the volume of inference requests is expected to rise sharply. That creates an opening for hardware vendors that can offer lower latency, better energy efficiency or lower total cost of ownership.
Cerebras argues that its wafer-scale design gives it an edge in answering AI requests quickly. The company’s systems are built to keep more data on chip and reduce the need for external memory and networking, which can be a bottleneck in conventional accelerator clusters. Whether those advantages translate into broad enterprise adoption depends on software support, reliability and integration with cloud platforms. Oracle’s involvement could help address some of those concerns by providing a managed service through an established cloud provider.
Oracle’s AI World Announcements
Oracle made the Cerebras announcement at its annual AI World conference, an event that has become a showcase for the company’s AI infrastructure ambitions. Alongside the Cerebras deal, Oracle announced agreements with Nvidia and Intel and noted that it buys Arm’s own chips. The cluster of announcements signals that Oracle wants to be seen as a neutral platform that can host a range of AI hardware, rather than as a company tied to a single vendor.
That positioning could appeal to enterprises that want flexibility. Many large organizations are wary of depending too heavily on one supplier for AI compute, especially as demand outstrips supply and prices remain high. By offering access to Nvidia, Intel, Arm and now Cerebras, Oracle can pitch itself as a one-stop shop for AI infrastructure. The challenge is that supporting multiple architectures can increase operational complexity and may make it harder to deliver consistent performance across workloads.
Customer Base and Credibility
Cerebras already counts OpenAI and Amazon Web Services as customers. Those relationships are important because they show that the company’s hardware has passed scrutiny from some of the most demanding AI organizations. OpenAI’s use of Cerebras systems, if limited in scope, gives the chipmaker a marquee reference. Amazon Web Services, meanwhile, represents a major cloud channel that could expose Cerebras technology to a broad base of enterprise customers.
The Oracle deal adds another large customer and another cloud distribution point. For Cerebras, cloud partnerships are a way to reach customers that may not want to buy and operate specialized hardware themselves. For Oracle, the arrangement adds a differentiated offering that could help attract AI-native companies and enterprises running heavy inference workloads. The financial terms remain undisclosed, so the immediate revenue impact is uncertain, but the strategic value lies in the validation and the potential for future expansion.
Timing and Execution
The service is not expected until 2027, which gives both companies time to integrate hardware and software. Cloud deployments of specialized AI chips require substantial engineering work, including scheduling, virtualization, monitoring and customer-facing APIs. Oracle will need to ensure that Cerebras systems can be provisioned and managed alongside its existing GPU and CPU offerings. Cerebras will need to demonstrate that its hardware can operate reliably at cloud scale and that its software stack is mature enough for enterprise customers.
The 2027 timeline also means the competitive landscape could look different by the time the service launches. Nvidia is expected to keep advancing its roadmap, and other chipmakers are targeting inference with specialized designs. Oracle’s decision to add Cerebras now is a bet that wafer-scale computing will have a durable role in the market. Whether that bet pays off will depend on customer adoption, performance benchmarks and the economics of running AI workloads in the cloud.
Capital Spending and Data Center Expansion
Oracle’s plan to spend $70B on capital projects in the year to May 2027 follows $55.7B in the prior year. The increase reflects the enormous cost of building AI-ready data centers, which require high-density power delivery, advanced cooling and fast networking. Those investments are necessary to support the growing set of AI services that Oracle and its customers want to run. The Cerebras service will consume data center capacity, but Oracle has not said how much space, power or hardware it will dedicate to the offering.
The broader buildout also highlights the stakes for Cerebras. Cloud providers are choosing which accelerators to deploy and how to allocate scarce data center resources. A partnership with Oracle gives Cerebras a place in those decisions, but it does not guarantee volume. The company’s ability to convert the agreement into meaningful revenue will depend on demand from Oracle customers and on Cerebras’s capacity to deliver hardware at scale. For now, the deal stands as a notable vote of confidence in a challenger that is trying to carve out a role in a market dominated by Nvidia.