ASMPT Group CEO Robin Ng said in a Bloomberg interview on July 29 that chip architecture will continue to evolve as newer AI systems use more chips. For a company focused on semiconductor assembly and packaging, that view is less a forecast than a business outlook.

The packaging segment drawing less attention

AI hardware coverage often centers on chip designers such as Nvidia or AMD, while the companies that physically assemble those chips receive far less attention. ASMPT, headquartered in Singapore, is one of the largest players in back-end semiconductor packaging, the stage in which individual chips are assembled into the multi-chip modules that power data centers.

That part of the supply chain becomes more important as AI systems move toward more complex multi-chip designs, because the performance of the finished package depends not only on the silicon itself but also on how precisely the chips are connected. A shift in architecture therefore has implications beyond chip design, extending to the tools and processes used in assembly.

A key product for the company during the AI boom is its thermo-compression bonding tools, or TCB tools. These machines are essential for connecting chips in the tight and precise configurations required for AI and high-performance computing applications.

In December 2025, ASMPT secured orders for 19 chip-to-substrate TCB tools specifically designed for AI and HPC applications.

Earlier, in July 2025, the company reported better-than-expected bookings across both its advanced packaging and mainstream business segments. ASMPT said the stronger results were driven by “AI tailwinds” tied to data center expansion.

Why multi-chip architectures matter

ASMPT’s product portfolio is aligned with this shift. The company offers chip-on-wafer and chip-on-substrate solutions, both of which are important for assembling multi-chip packages.

It is also expanding further into next-generation AI chip development through a collaboration with IBM Research.

The demand case highlighted by Ng

Ng also said in the interview that the introduction of cheaper and more accessible AI models could increase long-term demand for AI chips rather than reduce it.

The logic is straightforward: as AI becomes cheaper to deploy, more companies will use it. More deployments would mean more inference workloads, which would in turn require more chips and more packaging.

For suppliers like ASMPT, that makes the packaging layer part of the broader AI buildout rather than a separate niche, since every step toward denser, more capable systems increases the need for advanced assembly equipment. The December 2025 TCB tool orders and the strong first-half 2025 bookings indicate that ASMPT may be in the early phase of a sustained demand cycle.