Mckinsey & Company at COMPUTEX 2026: Power, Cooling, Networking, and Compute: The New Four Horsemen of AI Scale
Publish Date :2026/07/22
▲In his keynote, Ryan Fletcher, Partner at Mckinsey & Company, examines the challenges and opportunities facing the AI industry as demand for intelligent systems continues to grow. Beyond models and algorithms, AI relies on a complex ecosystem of infrastructure, including power, cooling, networking, and compute.
At the COMPUTEX Pre-Show Forum, Ryan Fletcher, Partner at McKinsey & Company, delivered a keynote titled "Power, Cooling, Networking, and Compute: The New Four Horsemen of AI Scale." Fletcher examines the challenges and opportunities facing the AI industry as demand for intelligent systems continues to grow. He emphasizes that despite being perceived as an abstract collection of tokens and data, AI has a tangible, real-world impact.
As AI transitions from prototyping to mass production, the competitive landscape is also shifting from model iterations to foundational infrastructure systems. Fletcher states that the four critical constraints AI scaling enterprises face are Power, Cooling, Networking, and Compute. A failure in just one can compromises the entire infrastructure.
Power determines the speed at which enterprises scale AI. In North America, data center electricity demand is projected to rise from 4% of total generation in 2025 to 14% by 2030, yet planned power additions can only meet 40% of this demand.
Cooling presents another challenge, with data centers facing year-round, highly concentrated heat dissipation. Though thermal management has evolved from air and liquid cooling to direct-to-chip cooling, it resolves only 10% of this systemic issue.
Meanwhile, networking has failed to keep pace with raw processing power. Over the last two decades, network bandwidth grew by just 30% despite explosive growth in floating point operations per second (FLOPs). This brings enterprises face-to-face with a "Network Wall," forcing GPUs in a 100k-cluster to sit idle 30% to 60% of the time.
Finally, compute dictates how infrastructures are built, paving the way for next-generation "AI factories" that will bifurcate into two distinct markets, segmented by their cost and performance priorities.
Only by optimizing how underlying infrastructure is utilized can enterprises bypass physical bottlenecks and fully realize the commercial value of AI.
Watch the full keynote on @COMPUTEXTAIPEIshow YouTube: https://youtu.be/H2Y2tDhDoq0?si=MCK0A6IzHejCbkfP
Back