On pricing

People are very proud to share their revenue statistics and how much they’ve grown over the last N months.

But on the other side of that revenue are users - and hopefully happy ones. The beautiful thing about earning money in a free market is that it often coincides with improving someone’s life. A rule of thumb: The amount by which you improve someone’s life relative to their income is the price you can charge.

For instance, being on the receiving end of business-to-business products now, I’m shocked at how much a simple solution can cost if it’s targeted at businesses. I recently paid $5,000 for a 2-year contract with a cashflow planning web app specialized for the European market. They don’t allow shorter contracts than 2 years. Their product is essentially a slightly more refined Excel table but with exactly the right integrations for a German business, so I’ll happily pay for someone else to improve my life by a few percent. Well, “happily” might be an exaggeration - I’ve told everyone I know about their brutal pricing. But I still paid.

In the business-to-consumer space, the situation is different.

The products we use the most often don’t even cost us money - look at Instagram, X, TikTok, LinkedIn, etc.

Then there are subscription businesses, usually under $20/month, such as Netflix, Amazon Prime, or apps to track your gym progress.

For musicians, there are some services like $29 for a music video made from stock footage.

In the AI world, the pricing is still trying to figure itself out, and the model from the old world doesn’t really work well.

For example, typical SaaS apps barely have any usage costs. Sure, Netflix has some cost when you watch it, but they spend way more on creating content and acquiring users than on raw streaming expenses.

For AI apps, it’s different. The pure server costs - including GPUs - at neural frames account for about 30% of our revenue. I’m not sure if that’s good or bad, but it used to be higher and now has stabilized. That’s also why many AI platforms use credits - essentially “compute units” for subscribers - to keep usage in line with actual costs.

Our margins would be much lower if all users fully used their credits. But we’re a subscription business, like many others, and not everyone uses their credits all the time.

We used to offer an unlimited plan for $99/month. That came from a time when I thought, “Nobody would spend that much on what I hacked together.” Turns out people were willing to pay. Some even subscribed with two accounts, to render in two tabs simultaneously.

We recently introduced annual plans, and now, every few days, someone purchases the annual “neural ninja” plan at $792/year. It’s a 33% discount compared to 12 x $99, so it’s a great deal if you’re serious about AI animation. I wasn’t expecting it to resonate so well with people.

Back to the unlimited plan: it ended up costing us money because people used the platform so heavily. We had to sunset it, but some users remain grandfathered in, happily rendering away. Those are the people who pushed the platform forward, creating the wildest stuff without usage restrictions.

Honestly, the credits we need to charge just to stay profitable are a shame. Although you can create music videos for very little money on our platform, if you use our very best AI models and are enough of a perfectionist, you might end up spending a hundred dollars for a four-minute music video. Compared to a real music video, that’s still not much - but it’s still a significant investment, and nothing I’m thrilled about.

AI costs are tricky. You can’t charge consumers too much - even Sam Altman said they’re losing money on their $200/month O1 Pro usage.

Usually, compute becomes cheaper over time, and that’s true on a $ per FLOP scale. But image and video models still tend to get larger, and NVIDIA chips are getting more expensive, even adjusted for inflation.

There are some great open-source AI video models now, and the community is working on reducing costs. I do think this will lead to big changes in the next few years. But I just paid $0.13 to generate the above 5-second clip - purely for renting the H100 it ran on. We still have a way to go.

AI Video Film of the week

SPACE LOGIC has been one of our very early power users mentioned above. He has been pushing the limits of what’s possible on neural frames since the very early days. I am so happy to see his style evolving, still staying true to neural frames. The 34-minute banger below is a masterpiece in story telling, all created by one single person. Few years ago, this would have been absolutely unfathomable.

Hope you’re having a fantastic week! Let me know what you think of this newsletter, by hitting on reply! :)