Exclusive-Broadcom expects to sell 1 million 3D stacked chips by 2027
Published by Global Banking & Finance Review®
Posted on February 26, 2026
3 min readLast updated: February 26, 2026
Published by Global Banking & Finance Review®
Posted on February 26, 2026
3 min readLast updated: February 26, 2026
Broadcom expects to sell 1M+ 3D stacked chips by 2027. The two-die design improves speed and efficiency; Fujitsu is sampling and TSMC will build on 2nm fused with 5nm. Additional products ship in H2 2026.
By Max A. Cherney
SAN JOSE, California, Feb 26 (Reuters) - Artificial intelligence chip designer Broadcom said that it expects to sell at least 1 million chips by 2027 based on its stacked design tech, an executive told Reuters on Wednesday.
The forecast, which Reuters is the first to report, marks a new product and sales target for Broadcom that could represent a revenue stream potentially worth billions of dollars.
Harish Bharadwaj, vice president of product marketing, said the 1 million chips the company projects it will sell are based on an approach Broadcom developed that stacks two chips on top of one another, allowing the distinct pieces of silicon to be tightly bound to improve the speed at which data can flow from one chip to another.
The company has refined the technology over five years to the point where its first customer, Fujitsu, is making engineering samples to test the design. Fujitsu plans to produce the stacked, or 3D, chips later this year.
The million-chip figure includes several additional designs beyond the Fujitsu chip.
The company's stacking approach gives its customers the ability to build chips that have more horsepower and use less energy to tackle the rapidly growing computing requirements AI software presents, Bharadwaj said.
"Now, pretty much all of our customers are adopting this technology," he said.
Broadcom does not typically design entire AI chips itself. It works with companies such as Google for its tensor processing units (TPUs) and ChatGPT maker OpenAI for its in-house custom processors. The Broadcom engineers help translate an early design into the physical layout of a chip that can be fabricated by manufacturers such as TSMC.
The company's chip business has grown significantly because of the custom deals with companies such as Google. Broadcom projected that its AI chip revenue would double year-over-year to $8.2 billion in its first fiscal quarter.
As a result, Broadcom has emerged as one of the most significant competitors to Nvidia and Advanced Micro Devices, as it races to produce silicon that competes with the chip giants.
Fujitsu is using the new tech for a data center chip. Taiwan Semiconductor Manufacturing Co is fabricating the chip using its cutting-edge 2-nanometer process and fusing it with a 5-nanometer chip.
Companies can mix and match which manufacturing process TSMC uses with the Broadcom technology. TSMC fuses the top and bottom chip during the fabrication process.
Broadcom has several more designs in the works and expects to ship two more products based on the stacking tech in the second half of this year and to sample an additional three in 2027.
The company spent roughly five years developing the groundwork for the stacked chip tech and testing various designs to come up with a commercial product. Engineers are working to make chips that have as many as eight stacks of two chips each.
(Reporting by Max A. Cherney in San Jose; Editing by Lisa Shumaker)
Broadcom projects sales of at least 1 million 3D stacked chips by 2027, describing a two-die stacking approach that boosts bandwidth and lowers power for AI data center workloads.
Fujitsu is the first customer building engineering samples, and TSMC is fabricating the chips by fusing a 2nm die with a 5nm die. Broadcom also collaborates with firms like Google and OpenAI on custom AI silicon.
Broadcom plans to ship two additional stacked-chip products in the second half of this year and sample three more in 2027, with engineers exploring designs using up to eight stacked pairs.
Explore more articles in the Finance category



