Oracle is expanding its AI cloud infrastructure by integrating chips from Google, Cerebras, and SambaNova alongside Nvidia. This move aims to cut costs and supply chain risks, even as the company plans to spend $90 billion to $95 billion on infrastructure by fiscal 2027. Investors will focus on how this heavy spending translates into long-term cloud revenue and profitability.
Oracle is changing its approach to AI data centers, moving away from relying only on Nvidia for hardware. The company has announced plans to mix different types of AI chips into its cloud infrastructure, including Google’s Tensor Processing Units (TPUs) and specialized processors from Cerebras Systems and SambaNova. This strategy is designed to give enterprise customers more flexibility, allowing them to choose the best hardware for their specific AI tasks, whether for training models or running them.
By diversifying its chip suppliers, Oracle is trying to solve two big problems: the high cost of Nvidia hardware and the risk of being dependent on a single supplier. As demand for AI cloud services grows, supply constraints for top-tier chips have been a major bottleneck across the industry. By incorporating alternatives like Google’s TPUs and Cerebras chips, Oracle aims to ensure it has enough computing power to meet customer needs without being squeezed by one vendor’s pricing or production limits.
The company is also taking more control over its networking hardware. Oracle plans to deploy its own 'Acceleron' networking chips in its data centers instead of using certain Nvidia components. This is similar to the strategy used by major cloud providers like Amazon and Google, who build their own hardware to improve efficiency and lower costs. By controlling the networking layer, Oracle hopes to get better performance out of its data centers and reduce its dependence on third-party suppliers for critical infrastructure.
For investors, the biggest focus remains the company's massive spending plan. Oracle expects capital expenditure to reach between $90 billion and $95 billion by the 2027 fiscal year. This is a significant amount of money to spend on expanding data centers. However, Oracle is trying to manage this risk by shifting to a more collaborative model. The company indicated that customers are increasingly sharing the financial burden of these AI deployments, likely through pre-payments or committed capacity, which could help protect Oracle's cash flow.
The success of this strategy will depend on whether Oracle can execute its expansion without hurting its profit margins. Investors will need to track how much of this new capacity is actually being used by customers, and whether the move to diverse, in-house hardware actually leads to lower operating costs over time. The next few quarterly updates will be important to see if customer demand for this specialized cloud infrastructure justifies the heavy investment spending.