AI infrastructure demand is reshaping semiconductor supply chains today, especially memory production.
Recently I watched two videos that left me pondering… The comparison between the current AI boom and the dot-com era is interesting, but the question remains: how much can we actually learn from it? Even if the dynamics are similar, are we simply destined to repeat the same cycle?
One of the reasons I moved away from a purely natural science background toward the organisation and economics of technology was the human factor. What is considered rational, and what emerges when groups of self-interested actors pursue what appears to be the correct decision? Markets are not abstract mechanisms; they are the collective outcome of many individual choices made under uncertainty.
This is particularly relevant in technology sectors, where the line between necessary infrastructure and speculative overinvestment can be difficult to see in real time.
So here is a short reflection on an interesting topic at the intersection of business, globalisation, and technological development.
The current wave of AI investment is unprecedented in scale. Companies such as Microsoft, Google, and Amazon are investing tens of billions of dollars into AI infrastructure. At the hardware level, much of this investment flows into GPU clusters and memory-intensive systems built around accelerators from companies such as NVIDIA.
What is less visible is how this demand is reshaping parts of the semiconductor supply chain.
Modern AI systems require enormous amounts of memory, particularly high-bandwidth memory (HBM). HBM is produced by only a few companies, including Samsung Electronics, SK Hynix, and Micron Technology. Because this type of memory commands significantly higher margins than commodity DRAM, manufacturers have strong incentives to shift production capacity toward AI-focused products.
In practical terms, that means a larger share of manufacturing capacity is now dedicated to the infrastructure supporting AI systems.
This shift does not automatically imply shortages for consumer electronics, but it does change the balance of priorities. Smartphones, laptops, televisions, and embedded systems all rely on the same underlying memory ecosystem. When demand from one sector expands dramatically, it can influence pricing and availability across the rest of the market.
Several analysts have already pointed to rising memory prices and tighter supply in parts of the DRAM and NAND markets. Whether this represents a temporary cycle or a longer structural shift remains uncertain.
What makes the situation interesting is that none of the actors involved are behaving irrationally.
From the perspective of semiconductor manufacturers, prioritising higher-margin products is an obvious decision. For cloud providers, investing heavily in AI infrastructure is a strategic necessity; the cost of falling behind competitors could be far greater than the cost of overbuilding capacity.
Individually, each decision makes sense.
Collectively, however, these rational choices can produce outcomes that look very different from the perspective of the broader ecosystem.
This is where the comparison with the Dot-com Bubble becomes relevant. During the late 1990s, telecommunications companies such as Global Crossing and WorldCom built vast networks of fiber-optic cables in anticipation of explosive internet growth. When the market collapsed in the early 2000s, a large portion of that capacity remained unused for years.
The commonly cited figure that “90% of fiber was dark” is debated, but the underlying pattern is well documented: infrastructure investment significantly outpaced immediate demand.
Yet the story did not end there. As broadband adoption, streaming, and cloud computing expanded in the following decade, much of that infrastructure became essential.
In hindsight, the investment was both excessive and transformative.
The question is whether AI infrastructure will follow a similar trajectory. If demand for AI services continues to grow rapidly, today’s data centers and memory investments may simply become the backbone of future computing.
If adoption slows, the industry may experience a period where supply outpaces demand.
At this stage, both outcomes remain plausible.
In summary, he expansion of AI infrastructure is already influencing semiconductor supply chains, particularly in the memory market. Manufacturers are prioritising high-margin AI components, while hyperscale companies are investing heavily in data-center capacity.
None of these choices are irrational. In fact, they are exactly what economic incentives would predict.
But history suggests that when many actors make individually rational decisions under uncertainty, the collective result can sometimes look like overinvestment.
Whether the current AI boom ultimately resembles the early internet build-out or something entirely different remains an open question.
For now, it is simply a reminder that technological progress is rarely a straight line—and that infrastructure built during periods of uncertainty often shapes the systems we rely on decades later.
Written by Kailash de Jesus Hornig, Partner at Bayinco
Sources:
”https://www.youtube.com/watch?v=-YNk9_e4pg4”
”https://www.youtube.com/watch?v=Wcv0600V5q4&t=589s”
___________________________________________________________________
If you need help or advice now, we are here for you.
If you have interim or other staffing needs, we are of course happy to help you with that too. Don’t hesitate to get in touch. With us, you always get to talk to a specialist who understands your business in the areas of Finance, Tech, Sales, Marketing & HR.