Why My IoT API Hub Dream Died While OpenRouter Soared

Every so often, a headline makes you choke on your coffee. For me, it was the news that OpenRouter—a company that simply routes API calls to large language models—hit a valuation of 70 billion yuan. That's about $10 billion. For a middleware play. And I couldn't help but think: I had that idea five years ago. Except mine was for IoT data, not AI models.

Back in 2021, I wrote a business plan for a project called D³Link—short for Link Device and Developer with Data. The pitch was simple: just as OpenRouter would later unify access to AI models, my platform would unify access to IoT devices. One API to connect to any sensor, any gateway, any industrial machine. Instead of weeks of integration work, developers would get it done in days.

The plan was thorough. I had a tagline: "Let there be no hard-to-connect devices." I had a product roadmap: D³Link for data ingestion, D³Hub for developers, D³Solution for verticals, D³Mall for supply chain. I had a business model: free tier, then service fees, data usage fees, project packages. I even benchmarked competitors like Tuya Smart, Ping++, and Reekoh. I was careful to position as a neutral middleware—not competing with app developers, just enabling them.

And here's the kicker: that business plan won first prize in our company's innovation competition. I took the bonus and took the whole R&D team out for dinner. We laughed, we toasted, we felt like we were changing the world. Then the project went back on the shelf. No product, no users, no company. Just a PDF that looked great in a boardroom.

So why did OpenRouter win while my IoT dream stayed dead? Let me break down three things I got wrong.

Cold Start: The Trap I Walked Into

An IoT middleware's value depends entirely on how many devices are already connected. More devices mean more value, which attracts more developers, which brings more devices—but you can't start that flywheel without both sides showing up. That's the classic two-sided market cold-start problem.

My five-year financial model admitted it: the first two years showed losses of 2.37 million and 3.32 million yuan. I was betting on future network effects while bleeding cash up front. OpenRouter, on the other hand, had no such problem. The models were already there, the APIs were standard, and developers were already hammering at the door. OpenRouter just had to route traffic—a thin layer with near-zero cold-start cost. The network effect kicked in quickly because it was built on top of an existing ecosystem.

Heavy Middleware vs. Thin Aggregation

My D³Link was a beast. It had to handle protocol parsing, adapters, data lakes, cloud-to-cloud integrations, edge computing—you name it. Every new device meant custom engineering. That's the definition of a heavy, asset-intensive integration business. Margins get eaten by hardware costs and developer hours.

OpenRouter does something far simpler. It sits between the developer and the AI model, translates a few API calls, and takes a small cut. Pure software, near-zero marginal cost, infinite scalability. Same word—"middleware"—but completely different economics. Mine was a services company in disguise. Theirs is a platform with a toll booth.

Timing Is Everything

In 2021, the AIoT space was a mess. Every device manufacturer had its own proprietary protocol. Standards were a joke. Big tech was gobbling up consumer IoT, but the industrial and healthcare niches I targeted were too fragmented and too slow to pay. My "middleware gap" was real, but the market wasn't ready to pay for it.

By 2025, AI had flipped the script. Models were commoditized, developers were drowning in integration complexity, and they were happy to pay a small fee to skip the headache. The same logic—unify access, charge for convenience—became a 70-billion-yuan business because the window had finally opened.

The Real Lesson: It's Not About the Idea

For years, I told myself I failed because I was too hesitant, too cautious, or just not cut out for startups. But the deeper truth is simpler: I confused a beautiful PowerPoint with execution. I thought that if I could articulate the vision clearly enough, the world would somehow make it happen. That's not how it works.

OpenRouter didn't wait for a perfect plan. It started with one model, one developer, one awkward integration. It iterated in public, got feedback in real time, and adjusted. The product grew out of actual usage, not from a slide deck.

My mistake wasn't lack of vision. It was lack of motion. I spent months perfecting the pitch when I should have spent weeks building a prototype—even a clunky one—and putting it in front of a real user.

What I'd Tell Anyone Building a Middleware Business

If you're thinking about building a "unified API" or "middleware" play, here's my hard-won advice:

• Find the thin layer. The more you have to build and maintain, the harder it is to scale. Aim for the thinnest possible slice that delivers clear value.
• Attack an existing ecosystem. Don't create a new market from scratch. Bolt onto something that's already growing.
• Start with a manual process. Before you write a single line of code, do the integration by hand for one customer. Learn what they actually need.
• Charge from day one. Even a dollar. It forces you to find someone who cares enough to pay.

So What Am I Doing Now?

I finally quit my job in June. I'm all-in on AI now—not because it's trendy, but because it's the same game with better odds. I'm starting with AI training: content, community, and small paid services. It's lightweight, generates cash flow, and gets me in front of real users. My first product is slated for September. It won't be perfect. That's the point.

I'm done writing business plans. I'm done waiting for the stars to align. I'm done polishing the deck while the world moves on.

The lesson I wish I'd learned five years ago: Perfect is the enemy of done. A rough start beats a perfect wait. Every day you wait is a day someone else is out there shipping the thing you dreamed about.

So if you're sitting on an idea—any idea—ask yourself one question: What's the smallest, ugliest version of this that I can put in front of a real user today? Then go do it. Because the only thing worse than failing is never having tried at all.