MIDJOURNEY

David Holz had already done the hard thing once. He co-founded Leap Motion, a gesture-control hardware company that raised over $100 million, got acquired in 2019, and left him with a very specific lesson: hardware is brutal and VCs make everything complicated.

So when he started Midjourney in 2021, he did the opposite of everything Silicon Valley told him to do. No Series A.

No pitch deck roadshow. No press release.

He launched in open beta in March 2022 directly inside Discord — not a standalone app, not a polished product, just a bot inside a chat server that anyone could type a prompt into.

The Discord strategy was either genius or laziness. Probably both.

It meant Holz didn't have to build authentication, payments, a dashboard, or a community platform from scratch. Discord already had all of that.

He just had to make the image generation good enough that people couldn't stop using it. Turns out, he did.

Within weeks of the open beta, the server had hundreds of thousands of users watching each other's generations in real time — which was, accidentally, one of the most powerful growth mechanics imaginable. You'd join, see what other people were making, get inspired, try it yourself, and immediately share the output everywhere.

The product demoed itself publicly, at scale, for free.

Midjourney makes money the old-fashioned way: subscriptions. You pay a monthly fee, you get a certain number of image generations per month, and if you want more or faster, you pay more.

Plans run from around $10 a month to $120 a month for heavy users and professionals who need commercial licensing and priority GPU access.

There's no advertising. No data licensing deals that anyone knows about.

No VC investors to satisfy with a 'path to monetization.' Just: people want the thing, people pay for the thing, the company keeps the money.

What makes the economics genuinely strange is the ratio. In 2023, Midjourney reportedly crossed $200 million in annual revenue with roughly 40 full-time employees.

That's $5 million in revenue per employee — a number that makes most SaaS companies look wildly inefficient. The GPU costs are real and significant, but the company has been profitable, which in the 2023 AI arms race made it basically a unicorn of a different kind: one that didn't need anyone else's money.

The Discord-first model also meant almost zero customer acquisition cost in the early days. The community acquired itself.

Users generated outputs, shared them on Twitter and Reddit, and other users followed the trail back to the Discord server. The product was the marketing.

The core product is the Midjourney image generator — a model you access by typing a prompt starting with '/imagine' in a Discord server or (increasingly) through a web interface. You describe what you want, the model generates four image options in about a minute, and you pick one to upscale or iterate on.

Simple loop. Stupidly powerful results.

What distinguishes Midjourney's output from competitors is its aesthetic. The images tend toward the painterly, the cinematic, and the dramatically lit.

It has a 'look' — one that many artists find either inspiring or threatening, depending on their disposition. That aesthetic is partly the model architecture and partly the years of community feedback and fine-tuning that Holz has used to steer its style.

Version 5 and V6 of the model made photorealism genuinely unsettling — the outputs are now good enough that distinguishing them from real photographs requires close inspection and sometimes not even then. This is both the product's greatest achievement and its most significant public relations problem.

The subscription plans gate access to faster generation speeds, more monthly credits, private image generation (so your prompts and outputs aren't visible to the whole server), and commercial usage rights. The basic plan is fine for experimenting.

The pro plan is what working designers and studios actually use. Midjourney has also introduced features like image-to-image generation, style references, and character consistency tools — all aimed at making it more useful for professional workflows and not just one-off experiments.

The counterintuitive move was Discord. Every startup instinct says you build your own platform — control the interface, own the data, brand the experience.

Holz didn't. He built inside someone else's house, and it turned out to be the smartest distribution decision in the 2022 AI wave.

Here's why it worked. Discord servers are public by default and built around communities watching each other.

When you generate an image on Midjourney, everyone in the server can see your prompt and your result. That's not a privacy flaw — it's a live tutorial and a gallery rolled into one.

New users would join, scroll through thousands of stunning generations from other people, and be hooked before they'd even typed their first prompt. The learning curve basically dissolved.

The viral loop was automatic. People generated wild, beautiful, unsettling images and posted them everywhere.

Each post was implicitly an ad. 'How did you make that?' was the most common reply, and the answer was always 'Midjourney.' The brand spread without a marketing budget.

Midjourney also timed the cultural moment perfectly — or got lucky with it. The open beta launched in March 2022.

DALL-E 2 launched from OpenAI in April. Stable Diffusion dropped in August.

Suddenly everyone was talking about AI image generation, and Midjourney was already the one with the best aesthetics and the biggest community. Being first and being good at the same time is rare.

They pulled it off.

The legal situation is the thing that could genuinely unravel everything. Midjourney, like every major image generation model, was trained on billions of images scraped from the internet — most of them created by artists who never consented to having their work used as training data.

Multiple class-action lawsuits have been filed against Midjourney, Stability AI, and others by artists and stock image companies arguing this constitutes copyright infringement at industrial scale.

The courts haven't settled this. The law hasn't caught up.

And if a ruling goes badly against AI image generators, the entire training pipeline for these models becomes legally questionable overnight. That's an existential risk, not a headache.

Then there's the competition problem. Midjourney was the quality leader in 2022.

By 2024, Adobe Firefly, DALL-E 3, Stable Diffusion XL, and Google's Imagen are all producing results that are genuinely competitive. The gap is closing.

Midjourney's moat is its community and its aesthetic reputation — but those are softer advantages than a patent or a proprietary dataset.

And the Discord-first model, which was brilliant for growth, is starting to feel limiting. Professionals want more control, better workflows, and integrations that a chat bot inside Discord can't easily provide.

Midjourney has been working on a standalone web interface, but it has been slow to ship. Every month it stays Discord-only is a month that Adobe and Canva get more integrated into professional workflows.

Midjourney has no outside investors. No VC funding.

No institutional money. No strategic rounds from Microsoft or Google or anyone else.

David Holz has said repeatedly that he doesn't want outside investment because he doesn't want to answer to investors whose interests might not align with building the product he wants to build.

This is either a principled stand or a convenient position to take after your product accidentally made $200 million without needing a dollar of outside capital. Probably both.

The practical effect is that Midjourney has complete control over its roadmap, its pricing, its pace of hiring, and its response to controversy. When artists filed lawsuits, Midjourney didn't have a board pressuring it to settle quickly or pivot strategy.

When competitors raised billions, Midjourney didn't need to match their headcount or compute spend to stay in the game. There's something genuinely unusual about a company at this scale that can just...

do what it wants.

The comparison that keeps coming up is Basecamp — another profitable, bootstrapped software company that rejected the VC playbook on principle and built a sustainable business on its own terms. Holz has cited similar philosophies.

Whether Midjourney can maintain that independence as the AI compute arms race intensifies and legal costs mount is the open question.