Microsoft Ends AI Cloud Capacity Deal With Oracle as It Rethinks Infrastructure Partnerships

Microsoft is winding down its agreement to rent artificial intelligence computing capacity from Oracle after deciding not to renew the contract, according to a Bloomberg report published on November 6, 2026. The move ends a significant infrastructure arrangement announced in 2023, when Microsoft turned to Oracle Cloud Infrastructure (OCI) to secure additional computing resources for its AI-powered Bing services.

The decision reportedly reflects changing hardware requirements rather than a complete breakdown in the relationship between the two technology companies. Microsoft found that the computing capacity it rented from Oracle had become less useful because the servers relied on older generations of Nvidia graphics processing units (GPUs). The development comes as Microsoft reassesses billions of dollars in cloud infrastructure commitments amid rising costs, delivery delays and rapidly evolving AI hardware requirements.

Microsoft Declines to Renew Oracle Cloud Agreement

Microsoft has been switching off computing capacity rented from Oracle in recent months after deciding against renewing the agreement, Bloomberg reported, citing people familiar with the matter.

The arrangement was initially announced in 2023 as a multi-year agreement under which Microsoft would use Oracle’s cloud infrastructure to support AI-powered features in its Bing search engine.

At the time, Microsoft was facing growing demand for computing power as it expanded its generative AI products. Its own data centre capacity was not sufficient to meet all its requirements, prompting the company to secure additional resources from external cloud providers.

Oracle offered Microsoft access to its cloud infrastructure, allowing the software company to increase its available computing capacity without waiting for all its own facilities to become operational.

However, the value of the rented capacity reportedly declined as Microsoft’s AI requirements evolved. The Oracle servers used older generations of Nvidia chips that no longer matched the company’s changing needs.

The financial terms and full size of the original agreement were not publicly disclosed.

Older Nvidia GPUs No Longer Meet Microsoft’s Requirements

One of the central reasons behind Microsoft’s decision is reportedly the generation of hardware used in Oracle’s rented computing infrastructure.

AI models require powerful processors to handle training and inference workloads. GPUs are particularly important because they can perform large numbers of calculations simultaneously, making them suitable for processing the enormous datasets used by modern AI systems.

However, AI hardware is advancing quickly. Newer generations of Nvidia accelerators offer improvements in computing performance, memory bandwidth and energy efficiency.

As Microsoft’s AI workloads become more demanding, older-generation GPUs may offer less value compared with newer systems. This can affect the economics of long-term cloud rental agreements, particularly when companies are paying for large amounts of computing capacity.

The development also highlights a challenge for cloud providers: infrastructure that is commercially attractive when a contract begins may become less competitive as chip technology advances.

Oracle, however, defended the continuing demand for its infrastructure. An Oracle spokesperson told Reuters that demand for OCI’s AI infrastructure remains strong, including for earlier-generation GPUs.

The company also pointed to renewed contracts for older GPU capacity at higher prices, arguing that demand for these systems continues.

Why Microsoft Is Reassessing Its Cloud Partnerships

The Oracle decision is part of a broader reassessment of Microsoft’s external cloud computing commitments.

The company has invested heavily in AI infrastructure to support products such as Microsoft Copilot, Azure AI services and its partnership with OpenAI. Building and operating data centres requires substantial capital, while securing enough GPUs and related infrastructure has become a major competitive challenge.

Microsoft has therefore relied on external providers to supplement its own data centre capacity. These arrangements have helped it access computing resources more quickly, but they also create dependencies on third-party infrastructure, delivery schedules and pricing.

Recent reports indicate that Microsoft is becoming more selective about these commitments.

The company has reportedly tightened oversight of data centre projects involving providers such as IREN, Nebius and Nscale. Delays and rising costs have increased the importance of ensuring that contracted capacity becomes operational on schedule.

This does not necessarily mean Microsoft is abandoning external cloud providers. Rather, it suggests that the company is placing greater emphasis on hardware specifications, project execution, costs and the timing of capacity delivery.

IREN Faces Delays After Microsoft Requests Network Changes

Another development illustrating Microsoft’s closer scrutiny of infrastructure suppliers involves AI cloud provider IREN.

The company disclosed that Microsoft had requested changes to network design, delaying certain infrastructure deliveries. The changes pushed back IREN’s target for bringing more than $4 billion in annualised revenue capacity online.

The target, initially expected by the end of 2026, was shifted to January 2027, with a possible further delay until late February.

The delay is important because AI data centres require more than GPUs. They also depend on high-speed networking, power infrastructure, cooling systems and other components that allow thousands of processors to operate together.

If network architecture does not meet a customer’s requirements, even a facility with substantial computing capacity may not be ready for commercial deployment.

For Microsoft, these requirements are particularly important because it needs infrastructure that can support large-scale AI workloads reliably.

The IREN situation and the Oracle decision involve different issues: one concerns infrastructure delivery and design changes, while the other involves the usefulness of existing rented hardware. Together, they illustrate Microsoft’s growing focus on the quality and suitability of its external computing resources.

What the Decision Means for Oracle

Oracle has been expanding its cloud infrastructure business as demand for AI computing increases.

Its Oracle Cloud Infrastructure division has secured major contracts and attracted customers looking for access to GPUs and data centre capacity. The company has positioned OCI as an important alternative to established cloud providers such as Microsoft Azure, Amazon Web Services and Google Cloud.

The end of Microsoft’s rental arrangement removes one source of demand, but it does not establish that Oracle’s broader AI infrastructure business is weakening.

Oracle has maintained that demand for its computing infrastructure remains strong, including for older-generation GPUs.

The development nevertheless raises questions about how cloud providers manage the long-term value of expensive AI hardware. If customers increasingly prioritise the latest processors, providers may face pressure to upgrade equipment more frequently or offer more flexible pricing arrangements.

This could increase capital requirements and make infrastructure planning more complicated.

AI Infrastructure Is Becoming a More Selective Market

The AI infrastructure market initially rewarded companies capable of securing GPUs, building data centres and offering computing capacity quickly.

However, the market is becoming more demanding as customers evaluate not just availability but also performance, energy efficiency, networking and the total cost of operating AI systems.

For cloud providers, owning large amounts of computing capacity may not be enough. They must ensure that the infrastructure matches customer requirements throughout the life of a contract.

For technology companies such as Microsoft, the challenge is balancing long-term capacity commitments against the rapid pace of hardware innovation.

A contract that guarantees computing resources for several years can provide supply certainty, but it can also create the risk of paying for equipment that becomes less competitive over time.

This is particularly relevant in AI, where new generations of processors and memory technologies are arriving at a rapid pace.

The Bigger Picture

Microsoft’s decision to wind down its Oracle agreement illustrates how quickly the economics of AI infrastructure can change. The same computing capacity that helped address a shortage in 2023 may no longer be suitable for the company’s current workloads in 2026.

The development also highlights the risks facing cloud infrastructure providers. They must invest billions of dollars in facilities and hardware while anticipating what customers will need several years into the future. Faster hardware innovation can make those investment decisions more difficult.

For Microsoft, the priority appears to be securing computing capacity that meets its current technical and commercial requirements. For Oracle, the challenge is demonstrating that its infrastructure remains competitive across different generations of AI hardware and customer workloads.

Looking Ahead

The next phase of Microsoft’s infrastructure strategy will depend on how effectively it balances its own data centre expansion with capacity rented from external providers. Its decisions involving Oracle, IREN, Nebius and Nscale will be closely watched for indications of how the company evaluates costs, hardware performance and delivery reliability.

For Oracle, the end of this particular arrangement does not signal the collapse of its AI cloud business, but it underscores the importance of keeping infrastructure aligned with rapidly changing customer needs. As AI computing demand continues to grow, the companies that can deliver the right hardware, at the right cost and on schedule, are likely to have an advantage in the increasingly competitive cloud market.