Oracle expands cloud chip options with NVIDIA and Google partners

Diversifying the AI hardware stack

Oracle recently announced a significant expansion of its cloud AI chip cooperation. While the company continues to work closely with NVIDIA, it is also bringing in hardware from several competitors to provide a broader range of options for its cloud customers.

The new strategy includes the introduction of Google's TPU chips, which are specialized processors designed by Alphabet. Oracle also plans to offer computing power from SambaNova Systems and Cerebras Systems, both of which focus on AI inference. Inference is the specific part of the process where an AI system generates a response to a user's request.

Mahesh Thiagarajan, President of Oracle Cloud Infrastructure (OCI), explained that different AI workloads have different requirements. He noted that this selection allows customers to flexibly pick the most suitable infrastructure based on performance, throughput, and cost-effectiveness for their specific needs.

Deepening the NVIDIA partnership

Despite the move toward diversification, Oracle is still making major investments in NVIDIA technology. The company plans to leverage NVIDIA's latest-generation Vera Rubin AI server to scale its cloud deployment to a "gigawatt-scale" level.

However, the partnership is evolving in technical ways as Oracle seeks more control over its infrastructure. The company announced it will phase out some of NVIDIA's network chips in favor of its own self-developed Acceleron network chips. This allows Oracle to integrate its proprietary technology into systems that still utilize NVIDIA's primary processing power.

By balancing these two approaches, Oracle aims to maintain a high-performance environment for heavy AI training while creating a more flexible ecosystem for various other tasks. This dual strategy addresses the fact that while NVIDIA dominates the training market, the inference market is becoming more diverse and competitive.

Market dynamics and competition

The expansion of Oracle's chip portfolio highlights a growing trend in the AI industry where multiple manufacturers are vying for market share. Companies like Cerebras Systems are looking to expand their first-tier customer base by getting their chips deployed in large-scale enterprise data centers.

For Oracle, providing these diverse options is a way to manage the high costs associated with AI hardware. The company noted that a growing number of its customers are now covering the costs of the chips themselves, which may alleviate some of the financial pressure on the provider.

The move also positions Oracle as a versatile host for different types of AI development. By offering chips from various sources, they can cater to different budget levels and technical specifications, preventing them from being locked into a single hardware roadmap.

Financial outlook and expectations

Oracle is currently facing significant scrutiny regarding its high capital expenditure as it builds out this infrastructure. The company has set a capital expenditure target of about $90 billion to $95 billion for fiscal 2027 to support its growth goals.

Investors are closely watching the company's upcoming Investor Day in Las Vegas. There is a general expectation among Wall Street institutions that Oracle may raise its long-term financial targets, following previous increases in revenue and OCI revenue expectations.

Analysts are looking at the company's remaining performance obligations, which are the amounts contracted but not yet fulfilled. These figures are often used to measure the business reserve of subscription-based enterprises. With these obligations reaching $664 billion, some analysts believe there is sufficient support for the company's future share price.

What comes next for cloud AI

The next phase for Oracle involves the physical rollout of these diverse chip sets across its global cloud infrastructure. The company will need to ensure that these different hardware types can work seamlessly together within a single cloud environment for its customers.

As the market for AI inference continues to grow, the competition between chip makers like Google, Cerebras, and NVIDIA will likely intensify. Oracle's role as a distributor of these technologies gives it a strategic advantage in capturing a wide variety of enterprise workloads.

The success of this expansion will depend on how well Oracle can balance the high costs of infrastructure with the demand for cost-effective AI solutions. The upcoming Investor Day will be a critical window for management to address concerns about these investments and the path to positive free cash flow.