Hi PH, I'm John. Went from food trucks to running video models on my own GPUs
by•
Hey everyone, I'm John, founder of EpochCore.
My last business was food. I started with one food truck, grew it into four brick and mortar locations, and my business partner bought me out in 2023. After that I learned to code through an IBM program on Coursera and I've been building in AI since 2025.
Now I'm working on Holina Factory. It started as something I built to make our own marketing clips on two GPUs I own, an RTX 5090 and a GX10. Ten people asked to use it before I listed it anywhere, and four of them pay for it now.
For anyone else here who came from a non-tech business: what carried over into building software, and what didn't?
14 views
Replies
Welcome John! Ten people asking before you listed it anywhere is the kind of signal most of us never get. I'm curious about the own GPU side, since I went the other way: my app generates songs through paid model vendors, so every job has a clear cost per run and someone else worries about capacity. With two cards of your own, the cost is fixed but capacity isn't. What happens when three of your paying customers render at the same moment, do they queue, and do they see where they are in line? From the food side I'd guess you already think about it like a kitchen during the lunch rush.
@gbasovv Thanks, George, and sorry for the slow reply. I’m only just seeing this. Great question.
When several people render at the same time, jobs go into a queue that shows an estimate of how long each render will take. A few users have told me renders run a bit slower than Higgsfield or Runway. In the same breath, they said the price makes up for it ($29 gets you 50 ten-second clips, plus more).
To deal with busy periods, we’re setting up overflow to cloud GPUs (Brev) so high-volume spikes don’t slow down the queue.
I was always told that in business you have to be either the best or the cheapest. We’ve chosen to be the cheapest, and the trade-off is slightly slower renders. The output quality matches the other major gen-AI video providers.
@holina_factory No worries, thanks for the detailed answer. Showing the estimate up front is the right call, people wait much more calmly when they know how long. Cheapest with honest wait times is a clear position, and Brev overflow for spikes sounds like a sane way to keep it. Good luck with it!