Power & Grids
Google’s Michigan power plan shows what autonomous AI must build outside the server room
New generation, storage and demand flexibility are becoming part of the product plan for companies that want to operate AI at scale.
The lead / Chips & Memory
Micron’s HBM4 ramp shows why autonomy needs more than a faster chip
The memory beside the processor helps determine how much an AI system can remember, how many tasks it can serve, and what each task costs.
Read the analysis ↗
Across the publication
On our radar
Data Centers
Microsoft’s cooling debate shows why autonomous AI still has physical neighbors
A design that avoids evaporating water can still raise questions about noise, electricity and the place where the computing happens.
Networks & Edge
NVIDIA’s photonics push puts the network inside the autonomy cost equation
Larger AI systems need to move information as well as calculate. Optical networking changes where the engineering and investment must go.
The subject desks
View all ↗Chips & Memory
Understand chips & memory and the practical limits of autonomous systems.
Chips & Memory
Micron’s HBM4 ramp shows why autonomy needs more than a faster chip
The memory beside the processor helps determine how much an AI system can remember, how many tasks it can serve, and what each task costs.
Chips & Memory
NVIDIA’s Rubin rollout makes AI autonomy a systems migration
A new generation of accelerators also changes cooling, software, networking and the work needed before an agent can run reliably.
Chips & Memory
The new MLPerf tests ask a better question about autonomous AI
Agentic and end-to-end benchmarks broaden the evidence, but a leaderboard still needs to be read against the job a buyer wants done.
The subject desks
View all ↗Power & Grids
Understand power & grids and the practical limits of autonomous systems.
Power & Grids
Google’s Michigan power plan shows what autonomous AI must build outside the server room
New generation, storage and demand flexibility are becoming part of the product plan for companies that want to operate AI at scale.
Power & Grids
Google’s one-gigawatt flexibility deal separates AI work that can wait from work that cannot
Demand response could make large computing loads easier to accommodate, provided the promises match the actual workloads.
Power & Grids
The IEA’s new AI electricity outlook is a capacity warning, not a meter reading
Forecasts help explain the scale of the buildout, but mixing scenarios, geography and measured demand can distort the story.
The subject desks
View all ↗Data Centers
Understand data centers and the practical limits of autonomous systems.
Data Centers
Microsoft’s cooling debate shows why autonomous AI still has physical neighbors
A design that avoids evaporating water can still raise questions about noise, electricity and the place where the computing happens.
Data Centers
Cerebras’ OpenAI agreement turns chip ambition into a data-center delivery test
A large computing commitment creates demand, but useful capacity arrives only after hardware, power, software and customer acceptance meet.
Data Centers
NVIDIA’s liquid-cooled Rubin racks put the facility inside the autonomy strategy
The server, its cooling loop and the building are becoming harder to buy as separate decisions.
The subject desks
View all ↗Networks & Edge
Understand networks & edge and the practical limits of autonomous systems.
Networks & Edge
NVIDIA’s photonics push puts the network inside the autonomy cost equation
Larger AI systems need to move information as well as calculate. Optical networking changes where the engineering and investment must go.
Networks & Edge
Jetson Thor brings larger AI models onto robots—and brings the limits with them
On-device computing can reduce dependence on the cloud, but power, heat and timely decisions still define useful autonomy.
Networks & Edge
A Jetson-powered robot is also a software fleet that somebody must maintain
Product availability and update tools are only the beginning of keeping autonomous machines consistent in the field.
The subject desks
View all ↗Compute Economics
Understand compute economics and the practical limits of autonomous systems.
Compute Economics
Cerebras’ speed pitch raises the right question: what does a completed AI task cost?
Fast inference can make autonomous workflows more useful. The savings depend on the rest of the job, including retries and review.
Compute Economics
Figure’s Nscale commitment puts the price of future autonomy on today’s balance sheet
A multibillion-dollar computing plan can secure room to grow, but the capacity still has to arrive and be used productively.
Compute Economics
Google’s electricity growth shows why efficient AI can still need more power
Better chips and software reduce the cost of individual tasks. More users and more demanding autonomous workflows can push total consumption higher.