Why AI Is Humanity’s Largest Infrastructure Project, Not Just an App

1. From Software to Intelligence: A Paradigm Shift

Huang opens by contrasting traditional software, which is pre‑written and runs deterministic instructions, with AI that can understand unstructured data, infer context and intent, and generate intelligence in real time. This makes AI a "producer of capability" rather than a mere tool, shifting marginal costs from near‑zero to resource‑dependent.

2. The Five‑Layer Cake: A Complete AI Infrastructure Framework

Huang visualises the AI ecosystem as a five‑layer cake, each layer essential and dependent on the ones below.

Layer 1 – Energy : Physical power supply; training large models consumes hundreds of thousands of GPUs for weeks, equating to the electricity of a small city.

Layer 2 – Chips : High‑throughput, high‑bandwidth processors. NVIDIA GPUs dominate, but competitors such as Google TPU, Amazon Trainium, Huawei Ascend and Cambricon are also in the race.

Layer 3 – Infrastructure : The "AI factory"—massive data centers with specialized networking (InfiniBand/NVLink), cooling, and power management. Huang calls this the current global investment gap.

Layer 4 – Models : Large language and multimodal models (ChatGPT, Gemini, Claude, DeepSeek). Open‑source models like DeepSeek‑R1 are highlighted as key democratizers.

Layer 5 – Applications : End‑user value in drug discovery, autonomous driving, industrial robots, finance, education, etc., where most startups compete.

"AI can understand unstructured information, reason about context and intent. Most importantly, it generates intelligence in real time." – Jensen Huang

The framework shows that the AI industry is not homogeneous but a set of inter‑dependent ecosystems.

3. Three Counter‑Intuitive Insights

Insight 1 – AI Creates Jobs, Not Eliminates Them : Using radiology as an example, AI assists image analysis, freeing radiologists to focus on diagnosis, communication and care, thus increasing demand for their expertise.

Insight 2 – Building AI Factories Needs Technicians, Not PhDs : Large‑scale AI infrastructure requires electricians, pipefitters and installation engineers—high‑paid, dignified roles that do not demand advanced degrees, highlighting a broader labor opportunity beyond elite programmers.

Insight 3 – Open‑Source Accelerates Demand Across All Layers : Open models such as DeepSeek‑R1 lower entry barriers, prompting more organizations to adopt AI, which in turn drives higher demand for chips, infrastructure and energy, creating a positive feedback loop.

4. Chinese Perspective: Where Do We Stand in Each Layer?

Energy : China leads in renewable generation (hydro, solar, wind), giving it a natural advantage for locating AI data centers.

Chips : The biggest weakness; domestic efforts (Huawei Ascend, Cambricon, Suiyuan) are catching up but still lag behind NVIDIA H100/B200 and face export controls.

Infrastructure : Alibaba Cloud, Tencent Cloud, Huawei Cloud and Baidu Cloud are expanding AI‑specific data centers with investments in the hundred‑billion‑yuan range, though reliance on imported high‑end chips remains a risk.

Models : Rapid growth of domestic models (DeepSeek, Kimi, Wenxin Yiyan, Tongyi Qianwen, Hunyuan). DeepSeek is noted as a globally prominent open‑source model.

Applications : The most vibrant layer in China, with strong vertical use cases in industrial vision, financial risk control, and medical imaging, leveraging massive domestic data.

Key judgment: The top (applications) and bottom (energy) layers are relatively thick, while the middle (chips) layer shows a structural gap—both a risk and a decade‑long opportunity.

5. Closing Thoughts

Huang stresses that the narrative is not a NVIDIA advertisement but a framing of the era: AI’s story is just beginning and will be far larger than most imagine. Current multi‑hundred‑billion‑dollar investments are merely a prelude to a trillion‑plus‑dollar expansion, driven by a self‑reinforcing flywheel across all layers.

Although Huang benefits from the continued growth of this flywheel, the five‑layer framework offers valuable insight into the AI ecosystem’s dynamics.