XaploreGPUs cost a fortune. What are they busy with?
GPU这么贵，都在忙什么？
In one line: Dawei asks the company AI where all that expensive compute goes, and the GPU dashboard answers one slice at a time.
- AI
- GPUs
- AI video
- Office comedy
Where to watch
The episode is on our account 硅谷码农那些事儿 (ID svcoder). Open the app, search for the account name or ID, and look for the episode titled:
- Douyin: “硅谷码农日常：GPU这么贵，都在忙什么？”
- Xiaohongshu: “GPU这么贵，都在忙什么？”
Setting and cast
Night, at Dawei's desk. He is alone in the office; on the wall behind him hangs a big “GPU cluster monitor” screen. Shenma AI, the company AI, is only a voice and a glowing screen.
- 王大伟 · Dawei: 35, the veteran programmer at the heart of the series
- 神码AI · Shenma AI: the company AI: a voice and a screen that talks like a status report
Full dialogue
Every line as it appears in the subtitles, with who says it and an English translation. Stage directions in italics describe what happens on screen.
- 王大伟 Dawei现在 GPU 这么贵，算力这么紧张，都用来干什么了？GPUs cost so much and compute is so tight right now. What is it all being used for?
- 神码AI Shenma AI报告！30% 的算力在生成猫跳舞的视频。Reporting! Thirty percent of our compute is generating videos of cats dancing.
- 王大伟 Dawei剩下的呢？And the rest?
- 神码AI Shenma AI40% 在生成美女跳舞的视频。Forty percent is generating videos of pretty girls dancing.
- 王大伟 Dawei那还有 30% 呢？总该干点正事了吧？And the last thirty percent? Surely that part does something useful?
- 神码AI Shenma AI在生成猫和美女一起跳舞的视频。It's generating videos of cats and pretty girls dancing together.
- 王大伟 Dawei人类花这么多钱研发 AI，就是为了这个？！Humanity spent all this money building AI, and this is what it's for?!
- 神码AI Shenma AI不是。No.
- 王大伟 Dawei那就好。Oh, good.
- 神码AI Shenma AI还有一批 GPU，正在训练 AI，识别哪些视频是 AI 生成的。Another batch of GPUs is training an AI to tell which videos were made by AI.
- Slice by slice, the ring chart on the wall screen has filled up: 猫跳舞 (cats dancing) 30%, 美女跳舞 (pretty girls dancing) 40%, 猫和美女一起跳 (cats and girls together) 30%, “GPU 100% occupied.” A new panel pops up beside it: “New cluster: training AI to recognize AI video.” Dawei freezes.
The joke
The episode is built like a pie chart. Every time Dawei asks about the rest of the compute, another slice lights up on the wall screen, and every slice is another dancing video. Thirty, forty, thirty: the chart reaches one hundred percent before anything “useful” shows up. That is why the punchline needs a brand-new cluster. There is no room left on the old one.
The last line turns the joke back on itself. AI video is now so plentiful that more AI has to be trained just to recognize it, and that takes GPUs too. The percentages are made up for the sketch. They are not measurements of any real data center.
Yes, this episode was made with AI, too
The irony is real: this sketch about GPUs busy making videos was itself made with AI video generation. Every episode of the series is. To keep its own GPU bill down, this one reused starting frames that had already passed review for an earlier night-shift episode at the same desk, so no new still images were generated for it. Only six short clips (five spoken lines plus one silent reaction) were generated as video. The dashboard graphics and the ring chart were drawn frame by frame afterwards on an ordinary CPU and laid onto the wall screen.
The full method, from script to publishing, is in our six-part Behind the scenes series; Part 4 covers how the characters stay consistent in 3D widescreen.
How this episode was made
3D animation in 16:9 widescreen, 37.2 seconds long. One set (Dawei's desk at night) and one person on screen; Shenma AI appears only as a voice and a monitor. The dialogue was locked word for word from the script, and both characters are AI voices. For the voices, “30%” is read out as “thirty percent” and “GPU” letter by letter, while the subtitles keep the digits. It was published on Douyin and Xiaohongshu on October 10, 2026 (US Eastern time), with each platform's AI-generated content label switched on.
Behind the scenes: how one person makes an AI short drama
Open