Silicon Breaks Apart as Chiplets Take Center Stage

Growing AI demand is pushing Intel, AMD and Google toward modular chiplet chip designs for faster and more flexible data-centre processors

5 Mar 2026

A shift in semiconductor design is gaining pace as large cloud providers seek new ways to power expanding artificial intelligence workloads. Demand from companies running vast data centres is accelerating interest in processors built from modular components known as “chiplets”, an approach that could reshape how advanced AI chips are designed and produced.

For decades most processors were manufactured as a single piece of silicon. The model delivered predictable performance but is becoming harder to sustain as manufacturing costs rise and AI systems require larger and more complex chips.

Chiplets offer an alternative architecture. Instead of building a single monolithic processor, engineers assemble several smaller silicon components within one package. Each chiplet performs a specific role, allowing designers to combine computing, memory and connectivity modules in different configurations.

The approach has gained momentum as hyperscale cloud providers look for hardware that better matches the needs of their AI infrastructure. Companies running large AI platforms increasingly want processors tailored to their workloads rather than relying solely on standard designs. Chiplet architectures allow specialised components to be developed separately and integrated into a unified processor.

Large semiconductor groups have begun to organise their strategies around the model. Intel has promoted open standards intended to allow chiplets from different suppliers to operate together. The effort is being coordinated through the Universal Chiplet Interconnect Express consortium, chaired by Intel executive Debendra Das Sharma.

Other companies have already applied the concept in commercial products. AMD has used chiplet-based architectures in its server processors for several years, a move that has helped improve manufacturing efficiency while scaling performance for data-centre customers.

Cloud providers themselves are also becoming more active chip designers. Google continues to invest in custom processors designed for its AI workloads, reflecting a broader trend among large technology companies seeking tighter control over the hardware used in their infrastructure.

Several obstacles remain before a fully open chiplet ecosystem emerges. Industry groups are still working on interoperability standards, while intellectual property concerns could slow collaboration between suppliers.

Even so, the direction of travel is increasingly clear. As AI demand drives larger and more specialised computing systems, modular chiplet architectures are emerging as a central feature of the semiconductor industry’s next phase.