The current economic climate has echoes of the Roosevelt era.

Just like the rapid rise of automobiles in the 1920s created entirely new industries, we're seeing a similar phenomenon with AI today. We even had signs of the Prohibition with many not having access to higher end GPUs. Back then, the Great Depression exposed a mismatch between supply and demand, which Roosevelt's New Deal addressed through government intervention and massive infrastructure projects like rural electrification.

Today, employers face a parallel challenge with AI agents and white-collar workers. The true potential of AI hinges on GPUs with large context memory. Smaller memory leads to more errors, hindering productivity gains and forcing companies to invest in ever-larger, more expensive data centers.

This mirrors the early days of motorization. Layoffs free up capital for investment, but the economy struggles to absorb the displaced workforce quickly enough. This can lead to social problems like alcoholism, joblessness, and even global conflict. Employers who lay off workers to boost productivity risk a potential backlash in the form of reduced revenue, especially in discretionary sectors like travel, hospitality, gaming, and entertainment.

Fortunately, we've developed some economic tools in the last century. Keynesian fiscal policy, which we see in action today, is one example. Large infrastructure projects are once again electrifying America, and massive data centers are being built worldwide, fueled by government incentives from Saudi Arabia to the East Coast. Cryptocurrencies are also there to fill the gaps.

So, where should you invest: data centers, office "fabs," or edge cloud? The answer might lie in another great invention of the 20th century: the PC. Companies that could scale their costs in line with their workforce by simply buying more PCs as they hired and fired were the ones that thrived over IBM mainframes.

Cloud computing offers a similar solution by allowing companies to rent servers by the hour or even minute. However, cloud productivity has its limits. When a company downsizes, its cloud resources are also scaled back. For cloud providers to maintain revenue, other companies must immediately pick up that rack. This dynamic can make cloud options surprisingly expensive. While training workloads can utilize GPGPU capacity, more specialized TPUs may sit idle, throwing off the balance sheet.

This is where governments and large organizations can step in with significant public projects. Defense is a prime example, as GPUs can potentially replace security researchers, scanning and fixing vulnerabilities to protect national networks live.

Government contracts enable investment in large data center projects without the immediate pressure to generate returns. Much of the cloud's growth over the past decade was built on cheap debt for stable companies like Amazon. Government contracts can create and pay for excess capacity, allowing private traffic to fluctuate more freely.

Entertainment and movies are likely to be major consumers of this extra capacity. Creator traffic can leverage GPUs for generative videos and user interfaces, potentially requiring 100 times the GPU capacity available today. This could easily justify and pay for data center investment.

However, this excess capacity could eventually lead to inefficient resource utilization and higher prices for companies that frequently hire and fire. Ultimately, low-latency, interactive generative user interfaces powered by AI models (rather than traditional computer programs) will likely replace the software stack. These "digital twin" workloads could be the most cost-effective if they are located closer to the user, whether in the office or at home without extra fiber, land and power costs. The desktop computer or edge server may eventually become the revenue generator of the age, supporting streaming laptops.

Copyright© Schmied Enterprises LLC, 2025.