Uploaded August 2026 | Updated November 2026, 4 days ago
Are AI Data Centres About to Be Replaced by AI on a Chip?
Massive AI data centres packed with GPUs and specialised accelerators currently power the training and large-scale inference of today’s most advanced models. But a parallel revolution is underway: highly efficient AI-on-a-chip solutions — including neuromorphic processors, in-memory computing, edge NPUs, and even wafer-scale engines — that bring intelligence directly onto compact silicon.
These chips slash power consumption, cut latency, improve privacy by processing data locally, and enable real-time AI in devices from sensors and robots to smartphones. Companies are advancing brain-inspired designs and specialised inference silicon that can handle many workloads without constantly calling the cloud.
Yet full replacement of hyperscale data centres is not on the immediate horizon. Training frontier models and serving high-volume inference still demand the dense, interconnected compute of large facilities. The realistic future is hybrid: data centres for the heaviest lifting, and ultra-efficient AI chips for edge and on-device intelligence that reduce overall energy use and infrastructure pressure.
Advances in packaging, photonics, and specialised architectures are making both paths more powerful and sustainable.
Curious about custom electronics that bring smart ideas to life? Check out www.veeb.ch for bespoke electronic projects — from innovative displays and restorations to unique hardware builds.
https://www.veeb.ch
Link to Veeb YouTube and other counters store.
https://shop.dirtboy.industries
Get early access to advert free films and discussion forum here:
*** Join Prof Simon Patreon FREE ***
patreon.com/profsimon
Are AI Data Centres About to Be Replaced by AI on a Chip?
Massive AI data centres packed with GPUs and specialised accelerators currently power the training and large-scale inference of today’s most advanced models. But a parallel revolution is underway: highly efficient AI-on-a-chip solutions — including neuromorphic processors, in-memory computing, edge NPUs, and even wafer-scale engines — that bring intelligence directly onto compact silicon.
These chips slash power consumption, cut latency, improve privacy by processing data locally, and enable real-time AI in devices from sensors and robots to smartphones. Companies are advancing brain-inspired designs and specialised inference silicon that can handle many workloads without constantly calling the cloud.
Yet full replacement of hyperscale data centres is not on the immediate horizon. Training frontier models and serving high-volume inference still demand the dense, interconnected compute of large facilities. The realistic future is hybrid: data centres for the heaviest lifting, and ultra-efficient AI chips for edge and on-device intelligence that reduce overall energy use and infrastructure pressure.
Advances in packaging, photonics, and specialised architectures are making both paths more powerful and sustainable.
Curious about custom electronics that bring smart ideas to life? Check out www.veeb.ch for bespoke electronic projects — from innovative displays and restorations to unique hardware builds.
https://www.veeb.ch
Link to Veeb YouTube and other counters store.
https://shop.dirtboy.industries
Get early access to advert free films and discussion forum here:
*** Join Prof Simon Patreon FREE ***
patreon.com/profsimon











[Watch our previous episode!](#)
Nord VPN Link:
https://nordvpn.com/profsimon
Link to Joshua Bertrand X (twitter)
x.com/korea_ufo
Get early access to advert free films and discussion forum here:
*** Join Patreon FREE ***
https://www.patreon.com/profsimon Secret Spy missions 1950s](https://i.ytimg.com/vi/s725yZ2U6uQ/mqdefault.jpg)