Intel EMIB-T Packaging Could Play a Bigger Role in NVIDIA’s Rubin AI Chip Plans
Intel may be gaining momentum in the advanced chip packaging race as new industry reports suggest the company is actively promoting its EMIB-T technology to major AI chip designers. One of the most notable names reportedly connected to this push is NVIDIA, whose next-generation Rubin AI platform is expected to become a key driver of growth in the artificial intelligence hardware market.
According to analysis from UBS, NVIDIA could potentially use Intel’s EMIB-T packaging technology for a four-chip version of its Rubin Ultra GPU. If that happens, it would mark a significant win for Intel as it works to position its advanced packaging business as a serious alternative to TSMC’s CoWoS technology, which has become widely used in high-performance AI accelerators.
The growing demand for AI chips has made advanced packaging one of the most important parts of the semiconductor supply chain. Modern AI processors are no longer just about smaller transistor nodes. They also rely heavily on how multiple chip components, memory stacks, and compute dies are connected together. This is where packaging technologies such as Intel EMIB-T and TSMC CoWoS become critical.
Intel’s EMIB technology, short for Embedded Multi-die Interconnect Bridge, is designed to connect multiple chiplets efficiently without relying on a large silicon interposer. Instead, it uses small embedded bridges within the substrate to link different dies. This approach can offer advantages in cost, scalability, and design flexibility, especially as AI chips become larger and more complex.
One of the biggest benefits of EMIB-T is that it can reduce some of the size limitations associated with interposer-based designs. Traditional interposer packaging can create constraints when chipmakers need to build very large processors or combine several large dies. By using substrates and embedded bridges, EMIB-T may allow companies to design bigger and more powerful AI chips without requiring the same level of complex interposer stitching.
That flexibility appears to be one reason NVIDIA may be evaluating the technology for future AI platforms. UBS believes NVIDIA’s gross margins could remain around the 75 percent range through calendar year 2027, supported by the Rubin generation. However, the firm also suggests that margins beyond that point may depend partly on how NVIDIA structures its Rubin Ultra product lineup.
In particular, UBS noted that NVIDIA could offer both two-chip and four-chip versions of Rubin Ultra. The four-chip version is the one believed to be a possible candidate for Intel’s EMIB-T packaging. Such a design would likely demand advanced interconnect technology capable of handling massive data movement between chiplets and high-bandwidth memory.
NVIDIA Rubin is expected to be one of the company’s most advanced AI platforms yet. It is widely anticipated to bring major improvements in AI training and inference performance compared with the current Blackwell generation. The platform is also expected to use HBM4 memory, which should provide significantly higher bandwidth for large AI models, data center workloads, and next-generation machine learning applications.
The Rubin platform is expected to be paired with NVIDIA’s Vera CPU, which is reportedly designed with 88 Olympus cores. Together, Rubin and Vera are expected to form a powerful AI computing platform aimed at hyperscale data centers, cloud providers, and enterprise AI infrastructure. NVIDIA has suggested that Rubin could deliver a major performance uplift over Blackwell while also improving inference efficiency and reducing operating costs.
For Intel, winning any role in NVIDIA’s Rubin supply chain would be highly valuable. NVIDIA is currently one of the most influential companies in the global semiconductor industry due to the explosive demand for its AI accelerators. A packaging partnership, even for select Rubin variants, would help Intel demonstrate that its advanced packaging strategy can compete at the highest level of AI hardware manufacturing.
However, nothing is confirmed yet. UBS framed NVIDIA’s potential use of Intel EMIB-T as a possibility rather than a finalized decision. The success of such a move would likely depend on several factors, including yield, substrate supply, production capacity, cost efficiency, and whether Intel can scale EMIB-T to meet the needs of massive AI chip production.
Scaling advanced packaging is one of the biggest challenges facing the semiconductor industry. AI chips require enormous amounts of bandwidth and power delivery, and they often combine multiple compute dies with stacks of high-bandwidth memory. Any packaging solution must deliver strong performance while also achieving acceptable production yields. If yields are too low, costs rise quickly, making large-scale deployment difficult.
NVIDIA is not the only major technology company reportedly interested in Intel’s EMIB-T packaging. Google has also been mentioned as a potential customer for future AI chip designs, particularly for its TPU processors. These custom AI accelerators are used for machine learning workloads and could benefit from advanced packaging if Intel’s technology proves reliable at scale.
Industry analyst Ming-Chi Kuo has also suggested that Google’s interest could depend on whether EMIB-T can achieve strong enough yields for commercial production. This highlights the broader reality of the AI chip race: performance matters, but manufacturability matters just as much.
TSMC’s CoWoS remains a dominant force in advanced AI chip packaging, and demand for it has surged as NVIDIA, AMD, and other chipmakers increase production of high-performance accelerators. Intel’s challenge is to prove that EMIB-T can offer a competitive alternative at the scale required by the world’s largest AI hardware companies.
If Intel succeeds, it could strengthen its position in the semiconductor market beyond traditional CPU manufacturing. Advanced packaging is becoming a strategic growth area, and Intel has been investing heavily in foundry services, chiplet integration, and next-generation packaging technologies. Landing customers such as NVIDIA or Google would provide credibility and could open the door to additional AI chip packaging deals.
The timing is also important. UBS reportedly believes Rubin chip and compute board production is on track to begin soon, although rack-level cooling refinements may push mass production of full systems into the September or October timeframe. Cooling has become a major challenge for AI data centers, as next-generation accelerators consume huge amounts of power and generate significant heat.
As the AI infrastructure market continues to expand, packaging technology will play an increasingly central role in determining which companies can deliver faster, more efficient, and more scalable processors. NVIDIA’s Rubin platform is expected to be a major milestone in that evolution, and Intel’s EMIB-T could become part of the story if it meets the demanding requirements of high-end AI manufacturing.
For now, the possibility of NVIDIA using Intel EMIB-T remains an industry watchpoint rather than a confirmed partnership. Still, the discussion shows how important advanced packaging has become in the AI chip supply chain. The battle is no longer only about who can design the fastest GPU. It is also about who can connect massive chiplets, memory, and compute engines in the most efficient way possible.
If Intel can turn EMIB-T into a trusted solution for leading AI chipmakers, it may gain a stronger foothold in one of the fastest-growing segments of the semiconductor industry. And if NVIDIA adopts the technology for a future Rubin Ultra variant, it could signal a major shift in the competitive landscape for AI chip packaging.
