Scene hook
Last night, I saw this news while standing outside a convenience store. My first reaction wasn’t the policy itself, but: will the AI we use get more expensive and slower later on? I’ve been slow to notice this too. I used to think infrastructure was far away from a small team like ours, until the bill went up and response times got worse at the same time.
What this is + who’s already feeling it
This isn’t a new tool. It’s a very real signal: New York has become the first state to put a “pause” on data centers. Put simply, it’s hitting the brakes first and re-checking the math on power, land, and environmental impact. Last Tuesday at 8 p.m., in a coworking office near Jing’an Temple in Shanghai, Chen Ting, who runs overseas ad placement for clients, was still using AI to rewrite 12 pieces of client copy. She told me what scares her lately isn’t not knowing how to use AI. It’s a platform suddenly getting slower or more expensive while clients still expect the same turnaround. For teams of 1 to 5 people like ours, what we end up absorbing is often not the tech itself, but the uncertainty around it.
What it costs to replicate today
If I wanted to leave myself a fallback option, the cost to copy this setup is actually pretty low: RMB 0-300, 30 minutes, and the technical barrier is basically just being able to sign up for two different AI tools and save the login info. My first step would be to open the homepage of the AI platform I use most now, then register one backup tool, run the same prompt in both, and save the results. Not everyone needs to do this right now. If AI isn’t directly making money for you yet, it’s fine not to bother. But if it’s already part of client delivery, this really can help you sleep better.
Advice by stage: just starting / 1-2 clients / scaling up
If I were just starting, I’d do one thing first: not tie all my work to a single platform, and keep at least one backup.
If I already had 1-2 clients, I’d store my common prompts, pricing language, and delivery templates in a document so I could still patch things together manually on a day when a tool gets shaky. I’ve messed this up before too—I assumed “tool online = business online.”
If I were scaling up, I’d start paying attention to two things: whether infrastructure policy might slowly push up my costs, and whether my service can switch models or platforms when needed. We don’t need to understand server rooms in detail, but I do think we need to accept this earlier: AI business isn’t just about knowing how to use the tools. It’s also about whether we have a backup route.