On the importance of on-site computing power to avoid latency, privacy, cost and centralisation issues.

April 2024

Written by He Ji Jun Hao


I’m becoming a skeptic of the future of Cloud computing. Don’t get me wrong - data centres aren’t going anywhere, and 24/7 remote access continues to be the number one use case for why we use Cloud operators. But I can’t help but feel that we may be witnessing the dawn of a new era of computing, and it doesn’t necessarily bode well for Cloud infrastructure operators.

Ultimately, Cloud Computing = Centralisation. It’s putting all your eggs in one basket. One that’s invariably owned by Microsoft, Amazon or Google. And the only reason it ever took off is because (1) the hardware/infra was too expensive to buy & maintain at home and (2) you needed fast remote access for multiple users.

But the technological landscape is changing and I see four main challenges cloud operators face:

  1. Latency / Synchronicity
  2. Privacy
  3. Cost
  4. Centralisation

Issue #1: Latency

As our processing power increases, so does the sophistication of applications we’re building. That’s always been true for every technology I can think of; Bicycles are more complex than they were 100 years ago, same applies to lightbulbs, refridgerators etc.

The challenge with advanced software is that computing requirements and therefore hardware requirements also go up; and we as customers have become accustomed to short waiting times - no one’s going to accept a novel technology if it means getting things done at the pace of dialling up a modem from 1998. We want things better and faster - not either or.

Humane’s AI pin launch this week is a great example of how latency issues are going to render many applications useless if they run exclusively in a remote Cloud environment. Unlike many online critics, I actually think the device looks and sounds like a great idea. I’d love to have a personal AI assistant sit on my shoulder, have it hear and see everything around me and then communicate with me in real-time. It could be providing me with recommendations, instructions, relevant information, warnings, directions etc. etc. All that kind of reminds me of having a personal droid straight out of Star Wars. Pretty cool if you ask me.

Untitled

The challenge Humane now has is delivering all this in a truly human-centric way. For one, humans don’t sound robotic (although I’m sure we could get accustomed to that after having heard Siri for a decade), but more importantly, we don’t process basic information at a transmission rate of 500Mb/s. At least most of us with an IQ of >80.