AI infrastructure: Jensen Huang’s five-layer framework at Davos

At the 2026 World Economic Forum in Davos, Jensen Huang, CEO and co-founder of NVIDIA, laid out a clear vision for scaling artificial intelligence. He presented a five-layer model for AI infrastructure, stressing that each layer must be built, operated, maintained, secured and expanded. The speech underscored the massive physical and operational demands behind AI’s next wave.

Jensen Huang and NVIDIA’s AI journey

Born in Tainan, Taiwan, in 1963, Jensen Huang moved to the U.S. in 1973. He earned a BS in Electrical Engineering from Oregon State University in 1984 and an MS from Stanford University. In April 1993, Huang teamed with Chris Malachowsky and Curtis Priem to found NVIDIA in a Denny’s booth near San Jose, CA. They started with $40,000 in capital. By 1999, NVIDIA went public. A near-bankruptcy in the mid-1990s gave way to a breakthrough: the GPU. Since then, NVIDIA has evolved from a gaming chip maker into an AI and accelerated computing powerhouse.

Under Huang’s leadership, NVIDIA reported $60.9 billion in revenue in 2023. The company now employs 29,600 people and ranks 65th on the Fortune 500. Huang retains roughly 3.5% of NVIDIA’s shares. He’s been recognized by Fortune, The Economist and TIME, and honored with the IEEE Founder’s Medal and election to the National Academy of Engineering.

The five layers of AI infrastructure

Drawing on common industry frameworks, Huang outlined the five core layers that power AI at scale:

1. Cloud & Compute: High-performance GPUs and servers that run training and inference workloads. These systems handle massive parallel computations for neural networks.
2. Networking: High-speed interconnects and switches. They link GPUs across racks and data halls, reducing latency and boosting throughput.
3. Data Centers: Physical facilities that house compute and networking gear. They require power, cooling, security and space planning.
4. AI Models: Pre-trained and custom neural networks. They translate raw data into actionable insights and drive end-user applications.
5. Applications: Customer-facing services—from chatbots to autonomous vehicles to healthcare diagnostics.

Across these layers, Huang emphasized a single refrain: “Every layer needs to be built, operated, maintained, secured and expanded.” That mantra highlights the full lifecycle work required to keep AI systems reliable and scalable.

Operational demands across the stack

Scaling AI isn’t just a software problem. It’s a physical and logistical challenge. Here’s what it takes:

• Build: Designing and constructing data halls, server pods and networking closets.
• Operate: Managing workloads, orchestrating resources and handling maintenance windows.
• Maintain: Replacing failed components, updating firmware and monitoring environmental controls.
• Secure: Protecting hardware and data from cyber threats and physical tampering.
• Expand: Adding racks, GPUs and network ports to meet growing compute demand.

Huang painted a picture of AI clusters that rival small cities. He pointed out that each new model—like the latest large language models—can consume tens of megawatt-hours during training. That drives data center growth by double-digit percentages year over year.

Collateral impacts and industry shifts

Huang’s framework carries wide-ranging effects:

• Data center construction: Hyperscalers and colocation providers must add capacity quickly. That fuels a $200 billion market for data-hall builds and facilities upgrades.
• Energy demand: AI clusters can draw as much power as a mid-sized town. Utilities must upgrade grids. Renewable sources and on-site generators join the mix.
• Water usage: Cooling systems rely on water for heat exchange. New technologies—like liquid immersion—are gaining attention to boost efficiency.
• Supply chain: Chandeliers of servers need GPUs, memory chips and switches. Semiconductor output and packaging services face fresh constraints.
• Talent: Data center operators, network engineers and security specialists are in short supply. Training programs and certifications are racing to catch up.
• Geopolitics: Nations see AI infrastructure as a strategic asset. Compute-rich regions gain leverage, while others push to build sovereign capabilities.
• End markets: Industries from healthcare to autonomous mobility stand to benefit. AI applications promise faster drug discovery, safer vehicles and smarter manufacturing.

Strategic signals for investors and policymakers

By outlining the full stack, Huang sent clear signals:

• Investment opportunity: Vendors offering power, cooling, racks, networking and security gear will see strong demand.
• Energy policy: Regulators must balance grid stability with surging data center load.
• Infrastructure finance: Long-term contracts and leases for data halls gain appeal as stable, high-growth assets.
• Public-private partnerships: Governments may co-fund AI campuses to spur local economies and R&D hubs.

In short, the session did more than evangelize GPUs. It mapped out where capital and policy need to flow to keep AI engines humming.

Looking ahead

As AI workloads climb, Huang’s five-layer model becomes a practical guide. It shows that success isn’t just about training ever-larger models. It’s about sustaining the entire ecosystem.

For NVIDIA, the framework underlines its central role. From GPUs at the compute layer to partnerships on data halls, the company sits at each junction. That bodes well for future revenue as enterprises chase scale.

At the same time, challenges remain. Energy grids need upgrades. Supply chains must expand. The workforce needs new skills. Yet, if the industry tackles these hurdles, we’ll see AI deliver on its promise across sectors.

In Davos, Huang reminded us that AI’s power is physical. You can’t just code your way to the future. You have to build it, room by room, rack by rack—and keep it running.