What happened
Microsoft has decided not to renew its AI‑computing rental contract with Oracle and is gradually shutting down the capacity it had been using on Oracle Cloud Infrastructure (OCI). The decision stems from the fact that the older‑generation NVIDIA GPUs deployed in Oracle’s data centers no longer satisfy Microsoft’s performance specifications for its AI services, such as Bing’s conversational search. The original agreement, announced in November 2023, allowed Microsoft to supplement Azure with OCI GPU resources when accelerator supply was tight. No details on the specific GPU models or the financial terms of the deal were disclosed. Oracle, however, says demand for its GPU capacity remains strong, noting that renewed or resold contracts in its Q1 FY2027 period fetched a 20 % premium despite the hardware being at least four years old.
According to a Bloomberg report dated November 6, 2026, Microsoft declined to renew its AI‑computing rental agreement with Oracle and has been gradually turning off the rented capacity over recent months. The report cites unnamed sources familiar with the matter.
The original partnership, announced in November 2023, allowed Microsoft to use Oracle Cloud Infrastructure alongside Azure to provide AI for Bing’s conversational search, addressing a shortage of AI accelerators at the time.
Microsoft’s internal requirements have evolved, and the older‑generation NVIDIA GPUs (specific models not disclosed) in Oracle’s data centers are now considered insufficient for the company’s AI workloads. No financial terms of the contract were made public.
Oracle responded via a Reuters‑cited statement, emphasizing that demand for its GPU capacity remains robust. The company reported that GPU contracts up for renewal in its Q1 FY2027 were renewed or resold at a 20 % premium, with utilization near 98 % despite the hardware being at least four years old.
Why it matters
The move highlights a growing tension in the AI industry between rapid hardware innovation and the economics of large‑scale AI infrastructure. As newer GPU architectures (e.g., NVIDIA’s Vera Rubin platform) deliver substantially higher performance, older accelerators can become a bottleneck for leading AI workloads, forcing cloud providers and their customers to reassess long‑term contracts. For Microsoft, relying on outdated hardware could limit the quality and speed of AI features in products like Bing, potentially affecting user experience and competitive positioning. For Oracle, the episode underscores the risk that a portion of its AI‑focused cloud revenue may be vulnerable if customers shift to newer hardware, even though the company reports high utilization and premium pricing for existing GPU inventory. The broader implication is that AI‑intensive enterprises may need to build more flexible procurement strategies or invest in their own next‑generation hardware to avoid similar disruptions.
The AI hardware landscape is advancing quickly; newer GPUs deliver higher FLOPS, better memory bandwidth, and lower , which are critical for large language models and real‑time . When a major player like Microsoft finds older GPUs inadequate, it signals that the performance gap is widening.
For cloud providers, the episode raises questions about the longevity of AI‑focused infrastructure investments. If customers can outpace the hardware lifecycle, providers may need to accelerate refresh cycles or offer more modular, upgrade‑friendly contracts.
Microsoft’s AI services, particularly Bing’s conversational search, rely on cutting‑edge performance. Using sub‑par hardware could degrade response times or limit model size, potentially eroding competitive advantage against rivals such as Google.
Oracle’s claim of strong demand and premium pricing for older GPUs suggests a market niche for legacy accelerators, but the segment may be limited to workloads with lower performance requirements. This could affect Oracle’s revenue mix as AI customers migrate to newer hardware.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Watch for Microsoft’s next steps in securing newer AI accelerators, either by expanding Azure’s own GPU fleet or by negotiating fresh agreements with other cloud providers. Oracle’s upcoming hardware roadmap, including planned deployments of next‑generation NVIDIA, AMD, and other accelerators, will indicate whether it can retain AI customers seeking the latest performance. Industry observers should also monitor pricing trends for older GPUs in the secondary market, as demand from other workloads may keep those assets viable despite being unsuitable for cutting‑edge AI services.
Whether Microsoft will announce new GPU procurement plans, either expanding Azure’s own hardware inventory or forming fresh partnerships with other cloud vendors.
Oracle’s upcoming hardware rollout, especially any announcements about deploying NVIDIA’s Vera Rubin or other next‑generation accelerators, which could help retain AI customers.
Pricing trends for older NVIDIA GPUs (e.g., A100, V100) in the secondary market, as continued demand could sustain a niche but may not satisfy top‑tier AI workloads.
Potential shifts in the broader cloud market, as other providers (e.g., Google Cloud, AWS) may capitalize on Microsoft’s move by offering newer AI accelerator options.