TLDR
- Broadcom expects to sell at least 1 million 3D stacked chips by 2027, a target that could generate billions in revenue.
- The chips stack two pieces of silicon on top of each other, improving data transfer speed and cutting energy use by up to 10x.
- Fujitsu is Broadcom’s first customer, making engineering samples now and planning production later in 2026.
- TSMC is fabricating the Fujitsu chip using a 2-nanometer process fused with a 5-nanometer chip.
- Broadcom plans to ship two more stacking-based products in H2 2026 and sample three more in 2027.
Broadcom ($AVGO) has set a target to sell at least 1 million chips based on its 3D stacking technology by 2027, according to an exclusive report from Reuters published February 26, 2026.
The forecast was shared by Harish Bharadwaj, Broadcom’s vice president of product marketing, who spoke directly with Reuters.
The projection marks a concrete new sales target for a product line that Broadcom spent roughly five years developing.
The technology works by stacking two chips directly on top of one another. This tight connection allows data to move faster between the chips while using less energy.
Bharadwaj said the approach delivers around 10 times better energy performance compared to traditional designs — a key selling point as AI workloads keep growing.
“Now, pretty much all of our customers are adopting this technology,” Bharadwaj told Reuters.
Fujitsu is the first customer to use the new design. The Japanese company is currently producing engineering samples and plans to move into full production later this year.
TSMC is handling fabrication for the Fujitsu chip, using its 2-nanometer process and fusing it with a 5-nanometer chip. Customers can mix and match TSMC’s manufacturing nodes depending on their needs.
More Products in the Pipeline
Broadcom is not stopping with Fujitsu. The company expects to ship two additional products built on the stacking technology in the second half of 2026.
Three more designs are planned for sampling in 2027. Engineers are also working toward chips with as many as eight stacked pairs — so up to 16 layers of silicon in a single package.
The 1 million unit target covers all of these designs combined, not just the Fujitsu chip.
Broadcom’s model in the AI chip space is to work with large tech companies on custom silicon. Google uses Broadcom to design its Tensor Processing Units (TPUs). OpenAI also partners with Broadcom for its in-house processors.
AI Revenue Already Climbing
Broadcom projected its AI chip revenue would reach $8.2 billion in its first fiscal quarter — roughly double the same period a year earlier.
That growth has come largely from custom chip deals with hyperscalers, where Broadcom translates early designs into physical chip layouts that TSMC then fabricates.
The 3D stacking tech adds a new layer to that business, literally and commercially.
Broadcom’s AI chip revenue was projected at $8.2 billion for Q1 fiscal 2026, doubling year-over-year.





