A / 01LATEST NOTE
Power is the new compute constraint
The next AI campus is won or lost before a GPU is installed. Interconnection queues, power quality and time-to-energization now shape the real capacity curve.
AIJELLA INSIGHTS / 2026The systems behind
The systems behind
intelligence.
Focused research on the physical and economic infrastructure that makes AI scalable.
A / 01ALL RESEARCH
Ten concise field notes on compute, energy, data centers, semiconductors and the economics of AI deployment.
A / 02HBM and advanced packaging: the constraint inside the chip
AI accelerators are no longer defined by logic alone. Memory bandwidth, packaging yield and substrate capacity increasingly determine the output of the entire system.
READ THE NOTE ↗
A / 03Inference economics after the training boom
Training created the headlines. Inference determines whether AI becomes a durable operating layer with repeatable demand and measurable unit economics.
READ THE NOTE ↗
A / 04Liquid cooling becomes core infrastructure
As rack density rises, cooling moves from a facility accessory to a system architecture decision with direct consequences for uptime, water use and deployment speed.
READ THE NOTE ↗
A / 05AI factories and the contract behind the cluster
A GPU campus is not valuable because the racks exist. Its value depends on who buys the output, for how long and under which performance obligations.
READ THE NOTE ↗
A / 06The tools behind leading-edge chips
AI chip supply depends on a narrow chain of lithography, deposition, etch, inspection and metrology steps. The slowest qualified process sets the real output ceiling.
READ THE NOTE ↗
A / 07The network is part of the computer
In a large AI cluster, thousands of accelerators must behave like one machine. The fabric decides how much of their theoretical performance becomes useful work.
READ THE NOTE ↗
A / 08Edge AI meets the physical world
Robotics, industrial vision and autonomous systems move inference closer to the machine. The opportunity is defined by reliability and deployment economics beyond pilot conditions.
READ THE NOTE ↗
A / 09How to read AI data-center project finance
Land, power, construction, compute and customer contracts mature on different timelines. A robust capital structure must survive the gaps between them.
READ THE NOTE ↗
A / 10Where enterprise AI moats actually form
Model access is broadening. Durable software value shifts toward proprietary context, workflow control, distribution and verified outcomes.
READ THE NOTE ↗