Insights on Physical AI
Powerful new forces are reshaping how industries design and build products. As demand grows for AI everywhere, power efficiency at scale and stronger, more reliable connectivity, the need for essential, high‑performance chip technology has never been greater. Through this shift, GF’s role remains clear: we make technology real. We turn our customers’ ideas into products through our essential technologies—spanning power, RF, real time compute and differentiated platforms—delivered through dependable, globally distributed manufacturing and deep, collaborative partnerships.
Where the future takes shape is an ongoing series focused on these changes and the work happening across GF to meet them. Each edition will explore the technology, market trends and customer programs that matter most, and highlight how our global teams bring new ideas from design to manufacturing to real-world deployment.
We’re starting with Physical AI—intelligence moving out of the cloud and into the devices, vehicles and systems people rely on every day. These applications demand real-time performance, power efficiency and long-lived, reliable platforms so they can sense, think, act and communicate. Together with our customers and partners, we are building the technologies that make this possible.
Physical AI is one part of the broader opportunity ahead, and we will continue to share how GF is scaling and delivering where it matters most.
- Tim Breen , CEO at GF
A deeper look at Physical AI with Ed Kaste
Today, AI lives mostly in the cloud, but the next wave is bringing intelligence into the physical world. From vehicles and robots to industrial systems and wearables, Physical AI is enabling machines to sense, think and act in real time, under strict power and latency constraints. That shift toward real‑time secure compute is redefining what semiconductor technologies must deliver at the edge.
Ed Kaste , SVP of Ultra‑Low Power business at GF, leads the company’s product lines that help customers build these next‑generation systems across smart mobile devices, home IoT and automotive. Read the Q&A below for Ed’s perspective on why Physical AI is accelerating, how it’s reshaping semiconductor requirements, and how GF is investing to power this next wave of computing.
Read the Q&A below for his thoughts on the future of Physical AI, then head over to the GF blog for Ed’s full perspective on how GF is powering this next wave of computing.
Physical AI is beginning to show up everywhere. How do you define Physical AI, and why is it accelerating now?
Ed: Physical AI is AI moving off the cloud and into the real world—into machines that can sense, think, act and communicate in real time. What’s accelerating this shift is the demand for machines that can interpret their environments instantly and operate safely and autonomously. That requires real‑time performance, low latency, reliable connectivity and extremely efficient power usage. These are things traditional data‑center‑optimized silicon wasn’t built for providing an exciting opportunity for innovation at the edge.
You say Physical AI is reshaping semiconductor requirements. What’s changing?
Ed: Physical AI workloads are incredibly broad. Devices must incorporate multimodal sensing, local intelligence, actuation and communication—all under tight thermal and power constraints. At GF, platforms like FDX, FinFET, embedded MRAM/RRAM, RF and silicon photonics and advanced packaging are designed to deliver high performance at the lowest possible power. They allow customers to build systems that are adaptive and reliable over long lifecycles.
Power and latency seem to be recurring themes. Why are they so critical to Physical AI?
Ed: These systems run continuously, often with limited thermal headroom or battery capacity. If power consumption is too high or latency is too long, the system can’t make timely decisions—or worse, it fails to operate safely. Optimizing at the semiconductor level lets us deliver ultra‑low power, ultra‑low latency performance so Physical AI devices can do more under strict constraints. Without breakthroughs here, Physical AI simply won’t scale beyond early pilots.
Recommended by LinkedIn
What role does distributed intelligence play in the future of Physical AI?
Ed: We’re shifting away from centralized compute models. Instead of sending everything up to the cloud, more intelligence is moving closer to where the action is—right alongside the sensors and actuators in the real world. And as systems become more software‑defined, we can decouple hardware from software more cleanly. That means you don’t have to replace entire platforms to add new capabilities, extending system lifetimes by updating features over time. That’s especially important for long‑lived platforms like vehicles, robotics and industrial systems.
Vehicles and humanoid robots seem to be major examples. What makes them such strong indicators of where Physical AI is headed?
Ed: Modern vehicles offer a great preview of what Physical AI looks like in scale. These already use hundreds of chips for ADAS, connectivity, infotainment and battery management—and semiconductor content per vehicle keeps rising. Physical AI is core to autonomy, electrification and safety. For humanoid robots, the demands are even greater. They need multimodal sensing, ultra‑low latency distributed intelligence and precise motor control across dozens of degrees of freedom. A high‑end humanoid can require up to four times more semiconductor content than a software‑defined vehicle. They show how critical real‑time, power‑efficient platforms will be.
With GF’s recent acquisition of MIPS, what new opportunities does that unlock for Physical AI?
Ed: Combining MIPS’ compute architecture and IP with GF’s optimized process technologies gives us a stronger foundation for the real‑time, deterministic workloads of Physical AI. It lets us deliver platforms that integrate sensing, decision‑making, actuation and communication more efficiently and with more predictable performance. It’s another step toward scaling Physical AI across markets—from mobile and IoT to automotive, industrial and beyond.
Ed’s blog explores in more detail how GF is powering Physical AI innovation by enabling today’s platforms to think, sense and act. Check out the full piece:
You can also listen to GF Chief Business Officer Mike Hogan 's perspective on how Physical AI is shifting semiconductor design:
Did you know?
By 2030, next‑gen AI processors will require more than 5 kW each.
As AI workloads surge, power is becoming the new bottleneck. Engineers are rethinking chip architectures from the earliest design stages to squeeze out every possible watt, enabling next‑generation processors to keep scaling sustainably even as their power demands climb past 5 kW by 2030.
From Tim Nutt , Data center power business lead at GF's recent article, “How HVDC and GaN are transforming hyperscale data centers.” Read the entire piece at EDN: Voice of the Engineer :
Subscribe to be notified when the next issue of GF's Where the future takes shape is published.
#####################################################################################################################
It's great to read and hear GF's leadership thought when it comes to #PhysicalAI. Ed Kaste 's points really resonate. Exciting!
--
7mo
Very interesting development. Do you see increasing demand for precision metal components to support cooling and reliability in physical AI hardware?
Absolutely, #AI is the future of the Corporate World! ✨ *08.03.2026 - Happy #Women's Day! Thanks for sharing an insightful article, & Best wishes, GlobalFoundries. Syed Awees, ACCA. Syed Suheb. Aspiring PhD Scholar. ✨ I hope, you can find this useful! https://www.linkedin.com/posts/s-saidhamiyan-mie-mba_happy-womensday-ugcPost-7436318821705097216-FaE5?utm_source=share&utm_medium=member_android&rcm=ACoAAAW6WS8BeP6Ni7c8l_yGdbiV4nJNg5_PwZ8 #Thanks!
As AI workloads move closer to the edge, manufacturing scalability and reliability will become just as important as architectural innovation. Delivering high-performance chips at volume requires tight alignment between design, process capability, and global manufacturing execution.