You press send. Your AI prompt disappears from your screen. What happens next is unseen, but the physical journey is remarkable.
As soon as you press submit, your text is converted to binary: zeros and ones. Your computer divides it into smaller data packets and sends them over your home network using Wi-Fi or Ethernet technology to reach your router.
From there, the data packets travel along fiber optic cables to your ISP’s regional facility. And then they head to their final destination: an AI data center, possibly thousands of miles away on another continent.
If your destination data center is outside your country, things get even more interesting. The data packets will be converted into infrared light rays and transmitted through fiber optic cables laid on the seabed. Light travels through glass at about 125,000 miles per second.
It is fast enough to cross the entire Atlantic Ocean and attack a data center on the opposite coast in 40 to 80 milliseconds. You probably won’t notice this, because your brain measures the delay which it perceives as 100 milliseconds, as if it is instant, or feels like it.
Once your prompt arrives, it is sent to a specialized server rack that houses AI chips, GPUs, TPUs, and other specialized processors. The server puts together your entire prompt and feeds it into a larger language model.
The LLM then processes what you sent, calculates on billions of parameters, and starts generating an answer.
But don’t sit there until the calculation of the complete answer is finished. Instead, as the GPU outputs token by token, the server immediately wraps each piece and sends it back to you immediately.
