The Infrastructure Thesis

  AI is often described as a race for better algorithms and faster processors. But behind the rapid progress of AI lies a complex global supply chain. Building modern AI systems requires far more than designing powerful chips. It involves a network of industries working together: semiconductor foundries, memory manufacturers, advanced packaging providers, equipment suppliers, and data center infrastructure companies. Each layer depends on the others. When one part of the chain slows down, the entire system feels the impact. This interdependence makes the AI supply chain more fragile than it appears. Consider the components required to build a modern AI accelerator. The processor itself must be fabricated at advanced semiconductor nodes. High-bandwidth memory must be manufactured and stacked using specialized processes. Advanced packaging is required to integrate the processor and memory into a single module. Finally, the finished hardware must be insta...

  The expansion of AI infrastructure is beginning to reshape more than technology. It is also starting to influence local politics and public finance. Large data centers require significant physical resources. They need land, electricity, cooling systems, and network connectivity. When a major technology company decides to build a data center campus, the project can involve billions of dollars in investment. For local governments, this creates both opportunity and debate. On one hand, data centers bring economic development. Construction projects create jobs. Local businesses benefit from increased activity. Property taxes and infrastructure investment can also boost local government revenue. For regions seeking new economic growth, attracting data centers can become a strategic priority. Many local governments compete to attract these projects. They offer tax incentives, infrastructure support, and fast permitting processes. But the growth of data...

  The early narrative around artificial intelligence suggested a simple idea. Compute is infinite. Cloud platforms promised virtually unlimited computing power. If more capacity was needed, companies could simply rent more servers. But the rapid expansion of AI is beginning to challenge this assumption. Compute is not infinite. It depends on physical systems. Every AI workload requires processors, memory, networking hardware, and electricity. Each of these components must be manufactured, installed, and powered. The scale of modern AI systems makes this reality more visible. Training large models requires thousands of GPUs operating simultaneously. These processors must be connected through high-speed networks and supplied with enormous electricity. Building this infrastructure takes time. Factories must produce chips. Data centers must be constructed. Power grids must supply electricity. These constraints mean compute cannot expand instantly. ...

  For decades, one assumption shaped the digital economy. Electricity would remain cheap. Data centers expanded under this expectation. Cloud computing scaled rapidly because power costs were predictable and relatively stable. But the rise of artificial intelligence may begin to challenge this assumption. AI systems consume far more electricity than traditional computing workloads. Training large models requires thousands of GPUs operating continuously for weeks or months. Even after training is complete, running these models at scale demands massive computing clusters. Each processor consumes power. Each processor also produces heat that must be removed. The result is an infrastructure system that depends heavily on electricity. As more AI data centers are built, electricity demand begins to rise in concentrated regions. Utilities must expand generation capacity, upgrade transmission networks, and reinforce grid infrastructure. These upgrades require time and capital. In some regi...

  Advanced chip design used to focus almost entirely on the processor itself. Smaller transistors, faster clocks, and more efficient architectures drove most performance gains. But as semiconductor technology approaches physical limits, another layer of innovation has become increasingly important. Packaging. In modern AI systems, performance depends not only on the chip but also on how different components are connected together. GPUs, memory stacks, and interconnects must operate as a tightly integrated system. This is where advanced packaging comes in. Advanced packaging technologies allow multiple chips and memory stacks to be combined into a single high-performance module. Instead of relying on a single monolithic chip, designers can assemble complex systems using several specialized components. For example, modern AI accelerators often combine powerful GPUs with multiple stacks of high-bandwidth memory. These components must be placed extremely clo...

  The AI boom has pushed GPU designers and AI software companies into the spotlight. Much of the attention focuses on companies building the chips that power modern AI systems. But behind those chips sits another critical layer of the industry. Manufacturing. Most advanced AI chips are not produced by the companies that design them. Instead, they are manufactured by specialized semiconductor foundries. Among these foundries, one company plays an especially important role. TSMC. Taiwan Semiconductor Manufacturing Company produces the majority of the world’s most advanced chips. Companies like NVIDIA, AMD, Apple, and many others depend on TSMC to fabricate their designs. This relationship makes manufacturing capacity one of the key constraints in the AI hardware ecosystem. Designing a powerful chip is only part of the process. Producing it at scale requires extremely advanced fabrication facilities. Modern semiconductor fabrication plants cost tens of billion...

  Training large AI models requires enormous computing power. Thousands of GPUs must process data simultaneously across massive clusters. But compute alone does not determine performance. Data must move quickly between processors and memory. This is where high-bandwidth memory, or HBM, becomes critical. HBM provides the bandwidth needed to feed modern AI accelerators with data. Without it, even the most advanced processors cannot operate efficiently. As AI systems grow larger, demand for HBM has surged. Every new generation of AI GPUs requires more memory stacks. Larger models, larger clusters, and higher compute density all increase memory demand. But supply has not expanded at the same pace. Producing HBM is far more complex than producing standard memory chips. Manufacturers must stack multiple memory layers vertically and connect them using microscopic pathways. The process also requires advanced packaging technologies and tight integration with GP...