Antonio Neri on why chips alone can't power AI
The CEO of Hewlett Packard Enterprise argues that AI's real bottleneck is not the chip but the power, cooling, and efficiency of the machines around it, and that winning that fight is what unlocks the science everyone actually wants.
Somebody Has to Build the Machines
To the public AI is a software story; to Neri it is a machines story, and the fastest chip in the world is not enough on its own.
we work with great partners like NVIDIA, AMD, and Intel. They develop some of the amazing core CPU and GPUs. But that alone is not sufficient.
Power Is the Ceiling
Each new processor draws more power than the last, so cooling and energy, not raw speed, are the constraint Neri's engineers actually fight.
You have to have the entire infrastructure that goes around to be efficient because the power consumption on each of these processors is going up and up and up.
Exascale, Priced in Watts
His supercomputers reach two quintillion operations a second while using 35% less energy than the last generation, because the number that matters is now performance per watt.
two quintillion operations per second, but with 35% less energy than the predecessors.
Electronics in Water
Neri's answer to the heat problem is to submerge the servers in liquid entirely, a sealed, almost maintenance-free tank that drops into existing racks without redesigning the hardware.
People said, you know, I can't put electronics in water. Yes, this is it.
Agentic AI Will Pull the Demand
Consumers adopted AI fast and business is still cautious, but Neri expects agentic AI re-engineering business processes to bend the demand curve sharply upward.
as agentic AI takes place and you start re-engineering, re-envisioning these business processes, that's going to drive a tremendous demand.
A Minor in AI for Everyone
Neri wants every employee to keep their specialty but add a working fluency in AI, the way he taught himself the technology of each era.
I believe we as a company need to have a AI minor in every function.
Infrastructure Is the On-Ramp to Science
The reason to win the efficiency fight is the science it unlocks, from cancer and dementia cures to faster vaccines through accelerated molecular models.
Finding cure for cancer, finding cure for dementia and Alzheimer, finding new vaccination and medicine faster through molecular docking and acceleration of these models.
Fail Fast, Win the Award
HPE hands out a Failure of the Year award, because rewarding fast, honest failure is what lets engineers take on the hard problems in the first place.
Failure is part of the process, right? So we have a say in our culture, fail fast and improve.