Is there any way to avoid the computing capacity crisis?
It’s free, every week, in your inbox.
But thats easier said than done heres why.
In other words, today, demand cannot be simply met by ramping up data centre construction.
Nor should that be something we aspire to.
Credit where credit is due data centres have come a long way in minimising their environmental impact.
The trouble is, this isnt scalable globally, nor is boiling our seas a viable path forward.
All while still failing to meet the demand for compute resources.
Essentially, this means two things.
First, the development of more powerful and AI-centric chips.
Second, the enhancement of data transfer speeds.
Thats why everyones snatching them up, which has led to achip shortage.
Yet GPUs inevitably hit the same brick wall.
Improving data transfer capacity
No modern AI model operates on a single chip.
Instead, to get the most of available resources, you assemble multiple chips into clusters.
These clusters often form a part of larger networks, each designed for specific tasks.
The operative words here are can be.
Adoption would also be an uphill battle.
Plus, just like any cutting edge technology, optical computing has yet to prove itself in the field.
So, are we moving fast enough to avoid the crunch?
Definitely not to stop building data centres in the short term.
Professor Oskars Ozolins received his Dr.sc.ing.
it’s possible for you to follow him on LinkedInhere.
Story byOskars Ozolins
Professor Oskars Ozolins received his Dr.sc.ing.
