NVIDIA Moving Toward Supplying Entire AI Systems to Partners, Aims to Unify Rack Designs and Speed Up Deployment
NVIDIA appears to be preparing a major shift in its AI infrastructure business model, moving from supplying only GPUs and key board components to directly delivering complete AI systems to partners. This information surfaced during Wistron’s Q3 earnings call, where a JPMorgan analyst stated that NVIDIA is now working toward “directly supplying” full systems.
Today, NVIDIA relies heavily on Taiwanese manufacturers such as Foxconn, Quanta, and Wistron for AI server rack assembly. Traditionally, NVIDIA has supplied critical components such as AI GPUs, the server boards that house them, and hardware like the Bianca Port UPB. Suppliers would then design and build the remainder of the rack architecture. This model has worked in the past, but AI server deployments have grown in size and complexity, pressuring NVIDIA to reduce configuration variability and accelerate delivery timelines.
A very important question by a JPM analyst in Wistron's Q3 call yesterday, which has been circulated in the supply chain since October.
— Ray Wang (@rwang07) November 12, 2025
"NVIDIA appears to be planning a major shift in its business model—moving away from selling the board (Bianca Port UPB) and instead directly… pic.twitter.com/ssxkI1bH5M
According to JPMorgan’s commentary, NVIDIA plans to provide complete Level 10 system designs directly to suppliers, who will then follow these unified specifications rather than creating their own rack architectures. This effectively means NVIDIA will deliver system level “blueprints,” while partners handle manufacturing and assembly. The analyst notes that this shift is intended to streamline deployment and significantly reduce time to market.
Industry observers note that NVIDIA hinted at this transition when it introduced its MGX architecture. MGX defines the physical and electrical layout of entire servers and was designed to scale from single nodes to full rack level AI factories. The company now appears positioned to extend this philosophy to complete system production. By predefining roughly 80 percent of the design, NVIDIA can reduce deployment timelines from nine to twelve months down to as little as ninety days. This would apply to next generation systems such as Rubin and Rubin CPX racks, allowing customers to adopt the latest hardware much more quickly.
This shift also expands NVIDIA’s total addressable market by allowing the company to sell full systems rather than only boards and GPUs, increasing margins while strengthening control over the final product. Suppliers stand to benefit as well, since the work of assembling and delivering systems remains the same. During the earnings call, Wistron confirmed that the business model does not negatively impact its operations and may even provide advantages due to clearer system integration guidelines.
NVIDIA’s strategy indicates an effort to evolve from being primarily a supplier of AI chips into becoming a full infrastructure provider. This aligns with the rising demand for turnkey AI solutions and large scale data center deployments. Although NVIDIA has not yet issued an official confirmation, the information appears to reflect ongoing internal coordination between NVIDIA and its manufacturing partners as the company prepares for a new generation of AI hardware ecosystems.
Whether this model becomes the standard for upcoming NVIDIA AI factories will depend on how smoothly the new system integration approach rolls out across partners. For now, the industry is watching closely to see how this shift affects manufacturing timelines, pricing structures, and competitive dynamics in the rapidly expanding AI infrastructure market.
Do you think NVIDIA’s move toward full system integration will reshape the AI server industry, or will suppliers push back on tighter hardware standardization? Share your thoughts below.
