Oracle (ORCL.N) will integrate the AI computers from CBRS.O into its cloud infrastructure, further expanding the range of chips and accelerators available in Oracle's cloud infrastructure (OCI). Both parties stated on Tuesday that they plan to begin deploying Cerebras devices in the first half of 2027, but did not disclose the scale of the purchase or the transaction amount.

This collaboration has enabled Cerebras to acquire another large cloud computing customer. Cerebras adopts wafer-level chip design, utilizing an entire wafer that is typically used to fabricate multiple processors for a single computing system, with a focus on workloads such as AI inference that require high response speeds and throughput.

Cerebras CEO Andrew Feldman ( Andrew Feldman ) said in an interview with Reuters that the collaboration with Oracle will "significantly expand" the company's base of primary customers, and Cerebras devices are being deployed on a large scale in data centers of other companies.

Cerebras has currently established partnerships with OpenAI as well as Amazon Web Services AWS. According to average forecasts compiled by Bloomberg, the company's revenue is expected to increase to nearly $3 billion next year, which is about three times that of this year. Cerebras went public in May this year, and the market expects its annual revenue for this year to still be below $1 billion.

The introduction of Cerebras is part of Oracle's new round of AI infrastructure expansion. On Tuesday, Oracle also announced that it will offer Google's tensor processors (TPU) for the first time in OCI, and plans to incorporate Intel-supported SambaNova Systems chips. OpenAI's own-designed Jalape no chips will also enter OCI.

Mahesh Tiagarajan, President of Oracle Cloud Infrastructure, stated that different workloads have varying computational requirements, and increasing the range of hardware options allows customers to choose the appropriate infrastructure based on performance, throughput, and cost.

The products of Cerebras and SambaNova are mainly aimed at AI, which is the stage where the model processes user requests and generates responses after training is complete. NVIDIA holds a dominant position in the AI training market, but there are currently more chip architectures and suppliers in the inference market, which has also become an important direction for cloud vendors to expand their choices of computing platforms.

Tiangalajan stated that the shortage of chip supply still exists, and large customers, including OpenAI, can improve the flexibility of their infrastructure by using more types of accelerators. He also mentioned that Oracle's addition of other chips does not mean a reduction in dependence on NVIDIA, as NVIDIA GPU still has strong versatility.

While adding alternative chips, Oracle is also further expanding its cooperation with NVIDIA (NVDA.O). Both parties stated that OCI will scale up the deployment of AI servers using NVIDIA's new generation of Vera Rubin architecture to the "terawatt level", but no specific transaction amount was disclosed.

NVIDIA CEO Jensen Huang previously stated that each gigawatt of Vera Rubin systems could generate approximately $40 billion in revenue for NVIDIA. Based on the deployment scales disclosed by both parties, Oracle will still be one of NVIDIA's major AI infrastructure customers in the future.

Even in NVIDIA systems, Oracle has begun to incorporate its own technologies. According to Reuters, some new systems will not use NVIDIA's networking chips but instead opt for Oracle's internally developed Acceleron networking chips. This indicates that Oracle is simultaneously expanding the sources of computing chip supply as well as its internal technology stack for data centers.

In recent years, Oracle has been continuously expanding its AI data centers in an effort to increase its share in the large-scale AI computing power market. Last month, the company maintained its capital expenditure target for the fiscal year 2027 of approximately $90 billion to $95 billion and stated that an increasing number of customers are beginning to bear part of the chip procurement costs directly. Currently, Oracle has not disclosed the financing method for this Cerebras equipment procurement.

AI Infrastructure expansion has also become a focal point for Oracle's investors. The company's stock price has fallen by more than 30% this year, and the market continues to assess its financing needs for large-scale data center construction, capital costs, and the profit margins of its cloud business. Oracle will hold an Investor Day this week, and progress on data center projects, capital expenditures, as well as the profitability of its AI cloud business, are expected to remain the main topics of discussion.