The Scene: My AI Tools Suddenly Got Slow and Pricey
Last week I was using GPT-4 to draft Xiaohongshu posts — it froze three times before the output finally came through. My colleague Xiaozhang, who does Amazon product research, told me Midjourney renders now take twice as long as before. We both figured it was just a network issue on our end.
What's Going On: Musk Wants to Launch AI Servers Into Space
SpaceX just announced a partnership with Nvidia to send next-gen data centers — powered by the Vera Rubin NVL72 — into low-Earth orbit. Each rack packs 72 chips, and Nvidia claims each one delivers 25x the compute of the old H100. Musk says the first space racks will launch before end of 2027.
Background: US ground-based data centers are getting hammered by resident protests and power constraints. OpenAI's Sam Altman had previously mocked the "space data center" idea as absurd — then Musk went and signed a deal with Nvidia, planning to ship them up via Starship, one load at a time.
I never connected this to my own work — until my friend Ajie, who runs cross-border e-commerce, told me his ChatGPT monthly bill jumped 30%. For small sellers like us, the AI tools we use every day are just pawns in this space-level compute war.
Replication Cost: You Don't Need to Replicate This — Just Spend 5 Minutes Reading
Money: $0. Space data centers have nothing to do with the AI tools you use right now.
Time: 5 minutes to read this + 10 minutes to glance at your AI subscription dashboard.
Technical barrier: None. Just internalize one thing — over the next 2-3 years, AI tool pricing may swing because underlying compute is getting scarce.
First step: Open the subscription backend for ChatGPT / Claude / Midjourney and note what you spent this month. I made this mistake — never checking my AI bills, then getting blindsided at month-end.
Next Steps for Three Types of People
Just starting out (0 clients): Don't worry about this yet. Your AI spend is probably under $50/month. Start stressing when you cross $500.
1-2 clients: Start splitting your workload. My dumb method — which tasks need GPT-4 (core client deliverables) vs. which can use domestic alternatives (Kimi, Jimeng, Doubao). I roughly cut my spend in half.
Scaling up (team of 5+): This is directly relevant to you. AI tool costs could eat 10-20% of your profit. Consider locally deploying open-source models (Llama, DeepSeek both work), or negotiate annual payment discounts directly with vendors.
One last thing: this news isn't for everyone to care about. But if you're a small team relying on AI tools to get work done, understanding "why it's getting more expensive" matters more than "how to make it cheaper" — at least next month when your bill creeps up, you'll know it isn't your fault.