Nvidia and Intel announced a major collaboration to build new AI systems for data centers and personal computers. Nvidia will buy $5 billion of Intel common stock, and the two companies will co-develop multiple generations of products that pair Nvidia accelerators with Intel x86 CPUs. The agreement aims to cut the slow links between chips and speed up large model training and PC AI features.
Deal specifics and terms
Intel will design custom x86 CPUs that connect to Nvidia GPUs with Nvidia high-speed links. Nvidia said it will integrate those Intel-designed CPUs into its AI infrastructure platforms for sale to cloud operators and other customers. The companies described a multi-generation roadmap that covers both data center systems and a new class of PC system-on-chip that combines Intel cores and Nvidia GPU chiplets.
Nvidia will acquire the Intel shares at a fixed price of $23.28 per share. The investment represents about a four percent stake in Intel after the purchase. Company statements said the move is aimed at technical integration and commercial packaging rather than outright control. The firms also said they will coordinate with foundries and partners for manufacturing.
The plan calls for using Nvidia NVLink or similar interconnects so CPUs and GPUs can trade data faster than through conventional interfaces. That is meant to reduce bottlenecks when many accelerators work together on large models. Intel will continue its independent product roadmap while also delivering custom chips that Nvidia can bundle with its acceleration technology.
Market and impact
The market reacted quickly. Intel shares rose strongly on the news and broader indexes gained as investors saw the deal as a major strategic shift. Analysts noted that the partnership is unusual given the long rivalry between the two firms. Some observers see the move as pragmatic and timely given the surge in AI demand. Others warned it may raise regulatory and supply chain questions.
Technically, the deal aims to close the link between CPU and GPU in large-scale systems. That could improve performance for training and inference and reduce wasted time moving tensors across slow links. If the collaboration succeeds, it could reshape data center designs and give cloud operators a tightly coupled option for very large models. It could also influence PC makers that want to add advanced on-device AI features.
There are open questions. Nvidia still relies heavily on external foundries for GPU fabrication. Intel has its own fabs and also works with partners. Observers will watch whether Nvidia will move more of its manufacturing strategy or keep TSMC as its main supplier. There are also antitrust and national security considerations because the deal changes important parts of the chip supply chain, and investors will watch regulatory review closely.
For customers, the possible benefits are concrete. Cloud operators may get systems that run large models faster and with less overhead. Enterprises may see new packaged solutions that simplify deployment. PC users could gain more advanced local AI features without a large battery or latency penalties. For suppliers and rival chip makers, the move will force strategic choices about partnerships and capacity.
Risks and next steps remain. The firms will need to translate the roadmap into silicon and then into reliable systems. That work covers architecture design, firmware validation, cooling, and manufacturing scale. Executives on both sides said the collaboration will span several product cycles and that they will test and evolve the designs as they go. Market watchers will track the first co-designed products closely.