AI · Annual review · 2026-08-30
The Bottleneck Moved
Memory became an allocation calendar, optics received real money, neoclouds became capital structures, and the scarce product moved from a GPU to useful output per active megawatt.
AI Bottlenecks research
Research library across AI, Energy, and Robotics
Browse source-led essays and research notes across the physical constraints behind AI infrastructure, energy systems, and robotics.
Energy · Annual review · 2026-08-30
The Megawatt Became the Product
The power trade moved from grid avoidance to electrical equipment, then from order books to the harder question: which collection of assets can become an energized campus on time?
Robotics · Annual review · 2026-08-30
Robots Became Operations
A few humanoids crossed into measured work. The harder story was everything around them: qualified joints, hands, safety, field failures, integration and the economics of deployed labor.
AI · Flagship research · 2026-08-28
China Can Reshape Memory Without Winning HBM
China can pressure conventional DRAM prices long before it reaches HBM parity. The result could be a split memory market, faster incumbent migration into custom HBM, and a more violent cycle when supply turns.
Robotics · Flagship research · 2026-08-22
A motor is not an actuator: inside the joint that makes a robot move
A robot joint forces motors, gears, sensors, bearings, heat, control and safety to agree inside a small housing. The hard part is not making it move once. It is producing the same controlled force, at the same cost and lifetime, thousands of times.
Robotics · Pillar guide · 2026-08-22
The robotics stack: what has to scale before physical AI can leave the lab
A useful robot is not a model with legs. It is a six-layer industrial system that has to secure materials, make precise motion, perceive contact, control the body, pass production test and earn its keep in the field.
Robotics · Deep dive · 2026-08-22
The calibration bottleneck: why the thousandth robot is harder than the first
A prototype can be tuned by its creators. A production line has to measure, correct and trace every joint quickly enough that accepted robots leave the factory on schedule.
Energy · Deep dive · 2026-08-22
The transformer bottleneck: why an available megawatt can still wait
Power can exist on the system and still fail to reach a data-center rack. Between the grid and the load sits a custom chain of steel, copper, insulation, factory slots, testing, transport and site engineering.
Energy · Pillar guide · 2026-08-22
AI's energy bottleneck: the race for delivered power
A GPU reservation is only a promise. Capacity becomes real when turbine slots, transformers, interconnection studies, switchgear, cooling loops and crews converge at the same site.
AI · Pillar guide · 2026-06-01
AI's photonics bottleneck: why moving data is now harder than making chips
AI clusters are running out of room on copper. As back-end fabrics climb from 800G to 1.6T and 3.2T, the constraint stops being how many GPUs you can buy and becomes how cheaply you can move data between them. The case for optics as the next chokepoint, and a map of where it bites first.
AI · Deep dive · 2026-05-14
The von Neumann bottleneck: why AI's real constraint is memory, not compute
Eighty years after von Neumann, the bus is still the bus, and modern transformer inference has stretched it past the point where engineering can hide it.
AI · Essay · 2026-05-14
How to play the AI bottleneck: a field guide to the physical stack behind the model race
Stop trying to pick the winning model and start mapping the physical floor under all of them. The picks-and-shovels framework, the four chokepoints, and the dates on the calendar.
AI · Founder letter · 2026-05-01
The world is built out of a few narrow places
Why I built AI Bottlenecks, what the AI buildout actually rests on, and why the most interesting story in markets right now is happening five layers below where everyone is looking.