SINGAPORE: As artificial intelligence (AI) breaks out of computer and mobile phone screens and takes physical form, chipmakers are increasing their focus on the hardware and technologies that will run robots and the infrastructure needed to support them.
Humanoid robots, widely considered the most revolutionary form of physical AI, are closer to mass production than many people realise.
Such robots will need purpose-built semiconductor platforms to respond in real time and perform real-world operations.
At the same time, hyperscale data centres (DC) will have to contend with unprecedented traffic, which could spike bandwidth and power requirements to levels they are not yet capable of handling.
US chipmaker GlobalFoundries (GF), which has a large manufacturing footprint in Singapore and recently bought a local semiconductor firm, sees physical AI as the next big wave of innovation and a massive opportunity for the company to deliver energy-efficient, highly integrated chips and devices.
Gregg Bartlett, the company’s chief technology officer, said GF has been working on physical AI and related semiconductor technologies for several years, with the goal of changing the company from being just a chips contract manufacturer to a more holistic technology provider.
Along the way, it has bought many smaller companies for their technology and manufacturing capacity, including Singapore’s Advanced Micro Foundry (AMF).
The 2025 acquisition has expanded GF’s silicon photonics technology portfolio, production capacity and research and development in the republic.
This complements the company’s existing capabilities in the United States, unlocking new market opportunities and meeting the growing optical communication demands of AI DCs.
Bartlett told The Straits Times: “This is the perfect time for GF to have made these anticipatory investments in intellectual property and capabilities that will fuel this massive ramp-up that’s going to be happening in the next five years.”
Acquiring AMF has made GF the world’s largest pure-play foundry by revenue for silicon photonics that integrates optical and electronic components onto a single silicon chip, using light – or photons – instead of electrical signals to achieve ultra-fast and energy-efficient data transfer.
The move has strengthened Singapore’s position as a trusted semiconductor manufacturing and innovation hub, and a critical node in the global chip-supply chain.
Bartlett said some early manifestations of physical AI are already commercially available in the form of autonomous vehicles, drones and robots.
Companies such as US electric vehicle maker Tesla, US robotics and engineering company Boston Dynamics and China’s Unitree Robotics also offer humanoid robots, which are so far limited in number and functionality.
But Bartlett said fully functional humanoid robots that will work at factories, construction sites and in homes are not too far away.
Bank of America (BoA) believes the population of humanoid robots – which can learn directly from the world, translate language into action and adapt to people in real time – is poised for a take-off.
BoA Global Research projects annual humanoid robot shipments will surge from 20,000 units in 2025 to 10 million by 2035.
In 2026 alone, sales are expected to reach 90,000 as leading companies ramp up production capacity and scale deployments.
By 2060, the number of humanoid units in operation could reach three billion, surpassing cars on a per-capita basis, analysts Lynelle Huskey and Vanessa Cook said.
While industrial and service applications are expected to lead adoption in the near term, the number of household humanoids that can help out with laundry, vacuuming, the dishes and other tasks is projected to accelerate quickly and ultimately account for the largest share, or 62%, by 2060, they said.
Bartlett said sensing, thinking, acting and communicating are the core capabilities that make up any physical AI application, including humanoid robots.
That means a real-time operating model behind a physical AI system should be able to capture data from the physical environment using multimodal sensors such as audio, video, temperature, motion and environmental sensing.
It should also execute precise, timely actions through motors or actuators with precision feedback loops, as well as exchange data reliably and securely across distributed systems – from edge to cloud, and across devices.
To deliver these capabilities, GF has developed a host of technology solutions, such as an FDX platform for power management, and a differentiated FinFET platform for efficient sensing, real-time processing and seamless communication.
GF can also provide memory solutions including magnetoresistive random access memory (RAM) and resistive RAM, silicon photonics and radio frequency innovations that drive high-speed connectivity.
The chipmaker also has the capability for advanced packaging to bring together diverse technologies into compact, efficient systems.
Its Singapore operations have played an important role in the development of some of these technologies, said Bartlett.
“Singapore has a vital role in physical AI, photonics, power, mobile, automotive and data centres,” he said.
“All of these capabilities are centred here. It plays a really important foundational role for everything we do.” — The Straits Times/ANN
