Key Takeaways
- Broadcom aims to deliver 1 million units of its 3D stacked chip technology by 2027, opening the door to substantial revenue growth.
- The technology layers two silicon pieces vertically, boosting data transfer speeds while reducing energy consumption by up to 10x.
- Fujitsu leads as the inaugural customer, currently developing engineering samples with production scheduled for late 2026.
- TSMC handles manufacturing for the Fujitsu design, combining its 2-nanometer process with 5-nanometer technology.
- Two additional stacking-based products will ship in the latter half of 2026, with three more entering sample phase during 2027.
Broadcom ($AVGO) has announced plans to deliver 1 million chips powered by its 3D stacking architecture by 2027, based on an exclusive Reuters report dated February 26, 2026.
Harish Bharadwaj, vice president of product marketing at Broadcom, provided these details during a direct conversation with Reuters.
This milestone represents a tangible commercial objective for technology that took approximately five years to bring to market.
The innovation involves placing two chips in a vertical arrangement, one directly above the other. This configuration enables faster data flow between components while consuming significantly less power.
According to Bharadwaj, the design achieves roughly 10 times superior energy efficiency versus conventional architectures — a compelling advantage given the expanding demands of artificial intelligence applications.
“Now, pretty much all of our customers are adopting this technology,” Bharadwaj told Reuters.
Fujitsu serves as the inaugural commercial partner for this design approach. The technology company from Japan is currently manufacturing engineering samples and intends to begin volume production before the end of this year.
TSMC serves as the fabrication partner for Fujitsu’s chip, applying its 2-nanometer manufacturing process while integrating it with 5-nanometer technology. Clients have flexibility to combine different TSMC process nodes based on their specific requirements.
Expanding Product Portfolio
Broadcom’s roadmap extends well beyond the Fujitsu partnership. The semiconductor company plans to release two more products utilizing the stacking architecture during the second half of 2026.
Another three designs are scheduled to enter the sampling phase throughout 2027. Development teams are pushing the boundaries further, working on configurations featuring up to eight stacked pairs — potentially creating packages with 16 silicon layers.
The million-unit objective encompasses all these product variations collectively, rather than focusing on a single design.
Broadcom’s approach to the AI semiconductor market centers on collaborative custom silicon development with major technology platforms. Google leverages Broadcom’s expertise for its Tensor Processing Units (TPUs). OpenAI has similarly engaged Broadcom for proprietary processor development.
Strong AI Revenue Trajectory
Broadcom forecasted its AI semiconductor revenue would hit $8.2 billion during its first fiscal quarter — representing approximately double the figure from the corresponding period one year prior.
This expansion has primarily stemmed from bespoke chip agreements with hyperscale cloud providers, where Broadcom transforms conceptual designs into manufacturable chip layouts that TSMC subsequently produces.
The 3D stacking capability introduces another dimension to this business model, both in terms of technology and revenue potential.
Broadcom’s AI chip revenue was projected at $8.2 billion for Q1 fiscal 2026, doubling year-over-year.
