AI Demand is Still Booming

AI demand is still booming — and the latest data suggests the boom is accelerating rather than cooling, even as skeptics question whether the infrastructure buildout can sustain its current pace.

Dylan Patel, head of SemiAnalysis, laid out the economics in a widely discussed interview that has become required viewing for anyone tracking AI infrastructure. The key takeaway: the capital being deployed into AI data centers, custom chips, and networking equipment is not speculative — it's backed by real and growing revenue from enterprises that are paying for AI compute and seeing measurable returns on that investment.

The demand signals are unmistakable. Microsoft, Amazon, and Google are all accelerating data center construction timelines. NVIDIA's order books extend well into 2027. The major cloud providers are reporting that AI-related services are the fastest-growing segment of their businesses by a wide margin. And unlike previous technology cycles, this one is being pulled by demand rather than pushed by supply — enterprises are adopting AI because it's delivering results, not because vendors are forcing it on them.

The counterargument — advanced by analysts who see echoes of the dot-com bubble — focuses on sustainability. The concern is that current spending levels assume continued exponential growth in AI inference demand, and that if enterprise adoption plateaus at any point, the infrastructure overhang could trigger a painful correction. The SemiAnalysis data suggests that scenario is unlikely in the near term, but acknowledges that the long-term trajectory depends on whether AI applications can continue to demonstrate ROI at scale.

The most interesting data point is the shift from training to inference as the primary driver of compute demand. Training large models was the first wave of AI spending — buying GPUs to build models. Inference — running those models in production for real users — is now the dominant cost, and it scales linearly with the number of users and applications. That's a fundamentally different economic model than training, which is episodic and bounded by model release cycles.

What This Means For You: The AI boom is not a bubble in the traditional sense — it's backed by real revenue and real enterprise demand. But the pace of spending means that any slowdown in adoption will have oversized consequences for the companies that have bet their capital expenditure plans on continued exponential growth. If you're investing, the picks-and-shovels play (cloud infrastructure, semiconductors, networking) remains the most rational approach. If you're building a business, the implication is clear: AI compute costs are going to keep rising as demand outstrips supply, so optimizing for inference efficiency isn't optional — it's survival.