SalmonCat Studios
A mechanical-keyboard company I co-founded and ran end to end. Our product, the Aozora 69, set out to do something the custom-keyboard market said you couldn’t: deliver a highest-end finish at an entry-level price. Making that math work is the whole story.
Where it sat in the market
Custom mechanical keyboards are an enthusiast hobby with enthusiast prices. A milled-aluminum board with a custom PCB normally ships through a group buy at $250–$500+. That price gates the hobby — and it’s where every competitor lived.
SalmonCat’s wedge was to deliver that exact premium spec — milled 6061 aluminum, custom hotswap PCB, brass plate, VIA — at roughly half the going rate. The position is easy to describe and brutally hard to fund: at an entry-level price there is almost no margin to waste, so the entire product had to be engineered for cost from the first sketch. The development story below is really the business story.
The development process — engineering the cost down
The launch price was only possible because we drove the unit cost from an opening factory quote of $350+ down to $138 — a 61% reduction — without touching the spec that made the board desirable. That gap is what funded the whole company.
Five factory revisions
Co-developed the milled-aluminum tooling with the factory over five formal design-for-manufacture passes — simplifying geometry, loosening tolerances where they didn’t matter, and cutting finish steps the factory billed for. Dozens of internal CAD iterations of the top, bottom, plate and badge preceded the cut.
Sourced through the shortage
Launched into the 2021 global chip shortage. Rather than pay spot-market markups that would have blown the bill of materials, I sourced the controller and the scarce PCB components directly — protecting the unit cost the whole price depended on.
Designed for self-assembly
A custom alignment mechanism let buyers assemble the board themselves — removing a paid factory assembly step worth ~$7–8 a unit, and turning the build into part of the product experience enthusiasts actually wanted.
The cost didn’t fall in one negotiation — it fell across five formal factory revisions and a long internal CAD history (the top, bottom, plate and badge each went through many iterations before tooling was cut). Later revisions kept improving the board after launch too — a USB-C bottom case, a flex-cut plate — without raising the price.
The product — Aozora 69
Aozora (青空, “blue sky”) is a 69-key, 65%-style kit. Premium materials, a swappable cloud badge that became the brand mark, and five colorways.
The launch — a $231k group buy
We sold the Aozora through a direct-site group buy — a pre-order run on our own storefront, not a crowdfunding platform. The community decided whether it lived.
It cleared the $50k goal in under a week and closed the first 30 days at $231,849 — 4.6× target, 849 orders at a $273 average. No marketplace fees, no platform cut: we owned the channel, the customer relationship, and the data. The demand was built first, in the places enthusiasts actually live:
Demand before product
Months of interest-check threads, renders, sound tests and community feedback meant launch day opened to an audience that already wanted it — the group-buy model only works if the want is real first.
The data told us what to build
Brass plates outsold aluminum 4:1; the Hakushi white colorway broke away at 34% of orders. Owning the storefront meant we saw those signals directly and stocked to them.
Then we had to ship it
Launch demand outran fulfillment fast. Rebuilding the pack-and-ship flow took us from ~20 to ~100 orders a day — the unglamorous half of a launch, and the half that decides whether customers come back.
Operating it — and winding it down well
Past the spike, the work was the work: a real P&L, supplier relationships, inventory, customer support, and the slow decisions about a product whose moment had passed.
We’re winding the company down through 2026 — and trying to do it responsibly. The plan is to open-source the Aozora’s files (PCB, plate, case CAD, build guide) under a non-commercial license so the community can keep building, repairing and modding their boards long after support ends. A product is a promise to the people who bought it; closing one down is the last part of keeping it.
What I took from it
SalmonCat is the clearest proof I have of owning something commercial end to end. I didn’t analyze a business from the outside — I made the calls that decided whether it worked. The pricing position, the supplier negotiation, the launch mechanics, the fulfillment recovery: each one was a problem where the constraint was real and the money was real.
The throughline I keep coming back to is that the strategy and the engineering were the same problem. “Premium board, entry price” wasn’t a slogan — it was a unit-cost target that every CAD revision and every sourcing decision had to hit. Holding a business goal and a technical constraint in the same head, and refusing to let go of either, is the part of consulting and analysis I’m actually built for.
It also taught me where I was weak. Our inventory lived in spreadsheets far longer than it should have — the load-bearing thing held together with the flimsiest tool. A proper system from day one would have saved real money and real stress. I’d build that first next time.