Oracle Broadens Its AI Cloud Infrastructure

Oracle announced plans to expand its artificial intelligence cloud infrastructure agreements with Nvidia while introducing additional chip options from Google and Cerebras Systems. The move reflects Oracle’s strategy of supporting a wider range of AI workloads through a more diversified hardware offering.

Oracle and Nvidia said the cloud provider would expand its deployment of Nvidia’s next-generation Vera Rubin AI servers to a “multi-gigawatt scale.” The companies did not disclose the financial value of the agreement. Nvidia Chief Executive Jensen Huang has previously said that Vera Rubin systems could generate up to $40 billion in revenue per gigawatt for the chipmaker.

The expanded agreement positions Nvidia technology to remain a central component of Oracle’s AI infrastructure. However, Oracle is also seeking greater flexibility in how it delivers computing resources to customers.

Google and Alternative Chipmakers Enter the Offering

Oracle will begin offering Google’s tensor processing units (TPUs) for the first time. It also plans to introduce chips from Cerebras Systems and SambaNova Systems, which specialize in hardware designed for AI inference.

Inference refers to the process through which an AI system generates responses to user requests. While Nvidia has a dominant position in AI training infrastructure, the inference market is more diverse, creating opportunities for alternative chipmakers.

Oracle said customers have different computing requirements depending on their AI workloads. Its expanded hardware portfolio is intended to give businesses more flexibility to select infrastructure based on performance, throughput and cost.

Cerebras Chief Executive Andrew Feldman said the agreement would significantly expand the company’s access to major customers and allow its systems to be deployed at scale in other companies’ data centers.

Oracle Adds Flexibility to Nvidia Infrastructure

Oracle’s diversification extends beyond the addition of competing processors. The company said some Nvidia-based systems would use Oracle’s internally developed Acceleron networking chips instead of certain Nvidia networking components.

The decision indicates that Oracle is pursuing a more customized infrastructure model rather than relying exclusively on a single supplier’s complete hardware stack. Combining different processors and networking technologies could allow Oracle to tailor its cloud offerings to specific computing demands.

For Nvidia, the expanded agreement represents continued demand for its latest AI servers, although Oracle’s adoption of alternative chips and in-house networking technology illustrates how major cloud providers are seeking more control over infrastructure design.

Capital Expenditure and Financing Remain Important

Oracle did not disclose how it intends to finance the additional chip purchases. During its earnings call in September, the company said customers were increasingly paying for chips and maintained its capital expenditure target of approximately $90 billion to $95 billion for fiscal 2027.

The scale of that spending highlights the financial demands associated with expanding AI cloud capacity. Access to customer funding may help support Oracle’s infrastructure commitments, but the company’s ability to deploy capacity efficiently and generate sustainable returns will remain important considerations.

Oracle is scheduled to hold an investor day this week, providing an opportunity for investors to assess its infrastructure strategy, spending plans and approach to financing future expansion.

What Investors Should Watch Next

Investors will be monitoring the scale and timing of Oracle’s Vera Rubin deployment, the commercial adoption of Google TPUs and alternative inference chips, and any additional details on financing. The effectiveness of Oracle’s diversified hardware strategy will depend on whether it can meet customers’ performance requirements while managing infrastructure costs.

The broader AI cloud market is also evolving beyond the dominance of a single chip supplier. As demand expands across training and inference workloads, the ability to offer different computing architectures may become increasingly important for cloud providers competing for enterprise AI spending.