Oracle (ORCL) deepened its Nvidia partnership Tuesday while simultaneously adding Google TPUs, Cerebras, and SambaNova chips to its cloud catalogue, a supply-diversification move that signals the intensifying battle for AI inference workloads.
The shift matters to investors because it suggests Oracle is positioning its cloud infrastructure business as hardware-agnostic at a critical moment when the inference market – where AI systems respond to live user requests – is fragmenting rapidly and margins could diverge sharply from the Nvidia-dominated training segment.
Key Takeaways
- Oracle will scale Nvidia Vera Rubin servers to “multi-gigawatt” capacity.
- Google TPUs, Cerebras, and SambaNova chips added for inference workloads.
- Oracle’s proprietary Acceleron networking chips will replace some Nvidia interconnects.
Market Context & Competitive Positioning
Nvidia (NVDA) retains an estimated 70-80% share of the AI accelerator market, but rivals are gaining ground specifically in inference, where latency economics differ from the compute-heavy training phase that built Nvidia’s dominance. Oracle’s decision to simultaneously expand its Vera Rubin commitment and onboard three alternative silicon providers illustrates how hyperscalers and cloud operators are hedging against single-vendor concentration risk – a dynamic also playing out at Microsoft, Amazon, and Meta. The broader AI chip sector has been navigating this transition between training and inference cycles, with valuations increasingly sensitive to which vendors secure inference contracts.
Nvidia CEO Jensen Huang has previously said Vera Rubin systems can generate up to $40 billion in revenue per gigawatt for the chipmaker, implying that a multi-gigawatt Oracle buildout could represent tens of billions in potential Nvidia revenue over the contract’s life – though neither company disclosed a deal value Tuesday 1.
Detailed Analysis: Three Layers of Diversification
Oracle’s strategy operates on three distinct layers. First, it is scaling its Nvidia relationship to multi-gigawatt levels using the next-generation Vera Rubin architecture, ensuring it remains competitive on raw training and large-model compute. Second, it is substituting Nvidia’s networking chips with its own in-house Acceleron interconnects – a cost and margin lever that reduces per-rack spend even within an Nvidia-heavy stack 1.
Third, and most strategically significant for the inference market, Oracle said it will offer Google’s tensor processing units (TPUs) for the first time, alongside chips from Intel-backed SambaNova Systems and Cerebras Systems. The latter two companies specifically target inference workloads, where Nvidia’s dominance is structurally weaker. Oracle’s cloud buildout has already attracted large-scale tenants – TikTok’s U.S. operations represent one high-profile example of AI compute demand flowing into Oracle Cloud Infrastructure.
Oracle did not disclose how it intends to finance the expanded chip purchases. During its most recent earnings call, the company said customers are increasingly funding chip procurement directly, and it maintained its capital expenditure guidance of approximately $90 billion to $95 billion for fiscal year 2027 1.
Management & Partner Quotes
“Different AI workloads have different compute needs. That choice gives [customers] the flexibility to choose the right infrastructure for the performance, throughput and economics of each AI workload.”
– Mahesh Thiagarajan, President, Oracle Cloud Infrastructure 1
Cerebras CEO Andrew Feldman underscored the commercial significance of the tie-up for his company. “That’s a massive expansion of our customer base of tier-one customers,” Feldman said in an interview. “We’re being deployed in other people’s data centers at large scale.” 1
Investor Outlook
Oracle is scheduled to hold an investor day this week, where analysts will likely press management on the capital structure behind the chip expansion and whether customer-funded procurement can sustain the company’s capex trajectory without diluting free cash flow. The multi-vendor approach reduces Oracle’s chip-supply concentration risk but also complicates margin modeling, since different silicon vendors carry different cost profiles and support requirements.
For Cerebras and SambaNova, the Oracle deal provides a meaningful distribution channel into enterprise customers who might otherwise default to Nvidia-based infrastructure. The inference market’s competitive openness, relative to training, could accelerate revenue diversification across the chip sector – a theme that has already begun influencing semiconductor valuations. Broader AI infrastructure financing trends, including the role of large capital allocators, remain a watch item as detailed further in coverage of Broadcom’s capital cycle positioning.
Conclusion
Oracle’s expanded chip strategy represents a calculated hedge: doubling down on Nvidia for training-era workloads while seeding alternative silicon for the inference era. Whether the multi-vendor model translates into measurable margin improvement or customer wins will likely be the central question at this week’s investor day.
Not investment advice. For informational purposes only.
References
1Stephen Nellis (2026-10-27). “Oracle expands chip supply deal with Nvidia, adds Google and Cerebras offerings”. Idaho Statesman / Reuters. Retrieved 2026-10-27.
2Stephen Nellis (2026-10-27). “Oracle expands chip supply deal with Nvidia, adds Google and Cerebras offerings”. Reuters. Retrieved 2026-10-27.