The physical side of AI
AI may begin with models and software, but it eventually becomes physical.
Chips have to be manufactured. Compute has to be housed. Electricity has to reach the site. Heat has to be removed. Buildings have to be constructed, connected and operated.
DC Decoded is where I try to make that physical system easier to understand — data centres, power, grids, cooling, supply chains and the commercial choices behind them.
I started writing these pieces separately. Over time, they became one part of a larger question: whether AI demand, infrastructure investment and the resources needed to support them are developing together.
New research programme · Southeast Asia
A public-source monitor that separates the operating base from the active pipeline, then follows approvals, physical power, construction, energization and customer commitments.
The layers still exist. What changed is how quickly Nvidia, Anthropic, OpenAI and the hyperscalers are willing to change the plan while executing it.
A plain-language guide to why data centres are measured in power, what a gigawatt means and where the enormous investment requirement comes from.
The old real-estate mental model for data centres is breaking down as rack density and reliable power become the defining constraints.
A closed strait, a disrupted helium supply and the hidden dependency connecting geopolitics, semiconductors and AI infrastructure.
The AI infrastructure buildout is not a single market. It is a chain of value pools, and different companies win at different stages.
Why the centre of gravity in AI infrastructure may move from building models to running them at scale.
The energy may exist, but delivering it reliably is a different challenge—especially as data-centre demand accelerates.
What changes when data-centre developers stop bringing power to computing and begin bringing computing to stranded energy.
A lesson from evaluating real businesses: revenue tells a story, but free cash flow reveals whether the business is truly healthy.
When conflict escalates, most industries turn cautious. Oil prices do the opposite, because production costs barely move even as fear pushes prices up.
Every AI conversation is about jobs. The bigger question: did the last wave of tech gains that made companies more productive ever reach everyone else?
AI's energy footprint isn't just data centres and water. It's also how verbosely AI responds, and whether three simple prompt changes can cut energy use.
Early results from a study testing whether a single instruction can make AI get to the point faster, without switching models or learning prompt engineering.