Artificial intelligence (AI) is no longer confined to data centers. At Computex 2025 in Taipei, executives from Arm, Nvidia, and MediaTek laid out a shared vision for AI's next chapter—one where intelligence moves to the edge, power becomes the new bottleneck, and collaboration across the ecosystem is the only way forward.
In a high-profile executive panel hosted by Chris Bergey, SVP and GM of Arm's Client Line of Business, three companies with divergent strengths found common ground on what it will take to scale AI across billions of devices. Joining him were Kevin Deierling, Nvidia's EVP of Networking, and Adam King, VP and GM of MediaTek's Personal Computing Business Unit.
Nvidia: Inference is revenue, latency is king
Deierling made a forceful case for 2025 as the "year of inference"—the stage of AI where money is made, not just spent. "Training foundational models is expensive," he said, "but inference is where value is delivered." He emphasized the increasing complexity of inference, especially as "agentic workflows"—AI models talking to each other—become the norm. In this context, latency becomes a critical constraint, and computing must move closer to where data is generated.
Enter "physical AI": warehouse robots, smart vehicles, and factory assistants that rely on ultra-low latency and rapid decision-making. "The laws of physics matter again," Deierling quipped. Power and thermal budgets—not silicon—are now the limiting factors. Nvidia is responding with innovations in co-packaged optics, NVLink 72, and liquid cooling infrastructure for data centers, as well as Jetson and Thor platforms for low-power edge AI. The company's DGX GB200 NVL72 system, paired with Arm-based CPUs, was also cited as an example of bringing petaflop-scale AI to the desktop.
Deierling recalled his own "aha moment" as the arrival of ChatGPT, which reframed AI as a daily partner rather than abstract technology. "You have to think of AI not just as software," he said, "but as a colleague—ask it how, not just what."
MediaTek: Edge-first thinking and the human talent gap
MediaTek's Adam King offered a grounded view from the consumer device front. From smart home gear to Chromebooks, he said, AI "lifts all boats"—enhancing both legacy devices and enabling new form factors like wearables.
Yet the real breakthrough, King noted, is the migration of workloads from cloud to edge. He showcased MediaTek's Kompanio Ultra SoC, now the most powerful AI chip for Chromebooks, offering 50 TOPS (trillion operations per second) of on-device inference. Powered by Arm CPUs, the chip underscores the role of right-sized compute engines—CPU, GPU, or NPU—tailored to specific workloads.
King also warned that Moore's Law alone won't save the industry. "If I could remove one barrier, it wouldn't be a technical one—it would be the shortage of semiconductor talent," he said, calling for renewed investment in engineering education. He described himself as a "techno-optimist", championing the democratization of AI—a future where mobile devices provide real-time access to expertise previously locked in books or institutions.
King praised cross-industry collaboration—citing the DGX Spark system as a "perfect example" of what's only possible when leaders like Arm and Nvidia align.
Arm: Architecting AI's ubiquitous future
Chris Bergey tied the session together by emphasizing Arm's foundational role in the AI era. "AI needs ubiquity, power efficiency, and a strong developer ecosystem," he said. With over 310 billion Arm-based chips shipped, the company sees itself as the connective tissue for AI, from hyperscale clouds to wearables.
Bergey pointed to the growing adoption of Arm CPUs in data centers—most notably AWS's Graviton, which now powers over 50% of new EC2 instances. He argued that legacy x86 architectures weren't built for AI's demands, especially in inference scenarios where CPUs remain the throughput bottleneck. Arm's low-power designs, combined with specialized silicon co-developed with partners, are changing that.
He closed by framing AI as a once-in-a-generation opportunity, but one that demands shared ambition. "The future of AI won't be built by any one company. It's an ecosystem game."
Article edited by Jack Wu
