A bunch of nerds don't even know this, and they're still acting like experts.

[00:00:00]Okay. We all know AI use a lot of electricity, but do you actually know why?

[00:00:05]Most of people think it's just about GPUs problem, right? And yes, GPU use a lot of power. But a huge amount of energy is actually spent on moving data around. Every second, massive amount of data have to move between GPU and networking hardware.

[00:00:24]So, traditionally, the data travel through copper cable by using electrical signal. But the problem is the more data you move, the more power you waste through resistant heat and signal loss.

[00:00:37]So, that is why semiconductor industry came out with a new technology, CPO.

[00:00:43]Co-packaged optics. So, in simple terms, in case your IQ not high, CPO move the optical connection much closer to the chips. So, the data get converted into light earlier, which reduce power consumption, lower latency, and improve efficiency.

[00:00:58]Right? So, at the Computex special this year, a Taiwanese company WeWynn showcased a CPO rack that can deliver up to four times faster data transfer while using five times less power.

[00:01:10]So, in this AI generation, a real bottleneck for AI isn't just about compute, isn't about AI model. It's how fast and efficient data you can move.

[00:01:20]Right? So, you people better pay attention on this kind of AI infrastructure.

[00:01:24]Because those are the reason you're paying the monthly fee for those AI company. You are not paying AI model.

[00:01:30]You're paying AI infrastructure. And honestly, if you don't care about it and you are even lazy to know how other people make money from you, then just like Andrew Tate say, stay broke.