Welcome To Operational Excellence (OPEX) Insight Article For The Paid Subscriber-Only Edition.

Subscribe now

This is the bilingual post in English and Vietnamese. Vietnamese is below.

Đây là bài viết song ngữ Anh-Việt. Tiếng Việt ở bên dưới.

English

When the thing blocking you no longer sits where everyone is looking

In 2026, a paradox recurring across artificial intelligence infrastructure forced many people to rethink how they understood growth. For several years, the biggest story was a shortage of chips, a shortage of GPUs, and whoever had more chips ran faster. But as 2026 arrived, a wave of industry analyses pointed to a reversed truth: what blocks the expansion of AI infrastructure is no longer the number of chips, but power capacity and the ability to cool. People can buy chips, but they cannot buy enough power, cannot erect substations fast enough, cannot connect to the grid fast enough, and cannot dissipate heat fast enough for machines that run ever hotter. The bottleneck has quietly moved, from the chip to the power socket and the cooling system, while most of the attention and money keeps pouring into the old place.

This is not a technical detail belonging only to the data center industry. It is a perfect illustration of one of the most profound and most misunderstood laws in operations management, a law that says every system is limited by a single point at any given moment, and that improving anywhere else is almost useless until you deal with that exact point. This law has a name, has a founder, has a concrete method, and has been applied in manufacturing, services, healthcare, software, and construction for nearly four decades. It is called the Theory of Constraints, in Vietnamese the theory of the limiting point, often abbreviated TOC.

The question this article pursues is very simple yet valuable for every business: when you want your system to run faster, make more, serve more customers, where should you pour your effort? Common intuition says improve every step, and if everyone tries their hardest the whole machine will get stronger. The Theory of Constraints says the opposite, and says it sharply enough to be uncomfortable: improving most of the steps is a waste, and only improving the right one step, the step that is currently the bottleneck, truly makes the whole system faster. Understanding this, and knowing how to find that one step, is one of the highest leverage management skills a person can learn.

This article will not argue where to build a data center or which cooling technology to use, because those are the industry’s technical matters. It will go into the operational law that lies beneath the story of power and cooling, trace back to where this law was born in a famous factory novel, dissect a few of its core and somewhat counterintuitive ideas, then come back to read the very case of the AI data center through that lens. And finally, we will draw out what an eatery, a clinic, a software company, or a repair shop can each take home and apply, because every system built by people has a constraint, whether its owner sees it or not.