Oracle AI Chip Shakeup Widens Beyond Nvidia

Oracle has unveiled an Oracle AI Chip Shakeup that widens its cloud far beyond a single supplier, expanding a major deal with Nvidia while adding rival chipmakers Google, Cerebras and SambaNova to its lineup. The cloud firm said it will expand to multi-gigawatt scale with Nvidia's newest generation of Vera Rubin servers, while also offering Google's tensor processing units for the first time, according to Reuters. Shares of Cerebras closed higher on the day, gaining by a double-digit percentage, according to market data published with the coverage. For companies building artificial intelligence products, the message is that the cloud's menu of hardware is about to get much broader, even if some of the new options will not be available until next year.

The announcement matters because it reframes a market that has long looked like a one-company story. Nvidia's hardware dominates the training of large AI systems, but the business of running those systems for everyday users, known as inference, is more open, and Oracle is now betting that customers want to pick the right tool for each job rather than accept a single default. That choice could affect prices, speed and availability for startups and large enterprises alike, and it puts Oracle in the middle of a contest between the world's biggest chipmakers.

A wider menu for AI builders

At the centre of the announcement is scale. Oracle and Nvidia said the cloud firm will expand to very large scale with the newest servers from Nvidia, although the companies did not disclose the value of that part of the deal, Reuters reported. Even those Nvidia-based systems will skip some of Nvidia's networking chips in favour of Oracle's own in-house networking hardware, a detail that shows how far the cloud firm is willing to go to shape its own stack. Alongside the Nvidia expansion, Oracle said it will begin offering Google's specialised AI chips, the hardware behind many of Google's own AI services, giving cloud customers access to technology that was previously far harder to rent.

The diversification does not stop there. Oracle plans to offer chips from Intel-backed SambaNova and from Cerebras, both of which build hardware aimed at inference, the moment when an AI system answers a user's question rather than learning from data. Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said in a statement that different AI workloads have different compute needs, and that choice gives customers the flexibility to pick the right infrastructure for performance, throughput and economics. That framing is the heart of the Oracle AI Chip Shakeup: a pitch that flexibility, not loyalty to one vendor, is what enterprise buyers actually want.

Why answering questions is the new battleground

To understand the stakes, it helps to separate the two halves of AI computing. Training a model is a one-off, enormously expensive effort that rewards the most powerful hardware available, and Nvidia has built a commanding lead there. Inference, by contrast, happens millions of times a day, every time a chatbot answers, a photo is captioned or a document is summarised, and it rewards speed, efficiency and cost as much as raw power. Because inference is repeated at such volume, even small savings per answer add up quickly, which is why a more diverse field of challengers has been able to gain ground in that segment while training remains concentrated.

Cerebras chief executive Andrew Feldman told Reuters that the Oracle deal represents a massive expansion of his company's base of top-tier customers, adding that the technology is already being deployed at large scale inside other companies' data centres. The reaction from investors was swift: the company's shares rose sharply on the news, a sign that the market sees distribution through a major cloud provider as a turning point for a smaller challenger. A similar distribution story is playing out in physical AI, where autonomous drone fleets are bringing AI processing to the edge rather than the data centre. In both cases, the prize is the same: getting specialised AI hardware into the hands of customers who would never build it themselves.

What it means, and what is still unproven

For readers building or buying AI tools, the practical upside is choice. A team running a customer-service chatbot, a coding assistant or a video tool could soon compare offerings on speed, cost and availability rather than taking whatever a single supplier provides, and competition of that kind has historically pushed prices down. The same week has brought other signs of enterprise technology in flux, including a large buyout offer reaching tens of thousands of employees at a software giant. Taken together, these moves suggest an industry redrawing its alliances as AI workloads grow from experiments into everyday infrastructure.

There are, however, reasons to keep expectations in check. The planned Cerebras service remains in the planning stage, with Oracle saying only that it is scheduled to become available during the next calendar year, without giving a more precise launch date, according to the coverage. No deal values were disclosed for the Nvidia expansion either, and the announcement offered no quantitative comparison of the new service's speed against rival systems. Nvidia, for its part, remains central to Oracle's plans, not displaced by them, which means the Oracle AI Chip Shakeup is better read as a widening of the tent than a changing of the guard. Watch for firm launch dates, pricing and early customer results: those will show whether a broader menu becomes a genuinely better deal, or simply a longer list.