Why Your Child’s Storybook Takes an Hour (and What It’s Competing With)

Sometimes a parent will start generating a book on Enchantably and then check back forty-five minutes later and it’s still going. I know this because they message me about it. Politely, but with the unmistakable energy of is this thing broken?

It’s not broken. It’s waiting in line.

And the line it’s waiting in is one of the most fascinating, absurd, and genuinely important queues in the history of technology. I want to tell you about it — partly so you understand why your book takes a minute, and partly because once you see what’s actually happening behind that loading screen, you’ll never think about AI the same way again.

Your book is sharing a brain with the entire world

Here’s the thing that nobody really explains about AI: it all runs on the same infrastructure.

The servers that generate your child’s personalized storybook — the illustrations, the poem, the layout — are drawing from the same global pool of computing power that is simultaneously being used to detect cancer in medical imaging, accelerate drug discovery, write code, generate memes, translate languages, power chatbots, compose music, and run the backend of nearly every major tech product you use daily.

It’s not that your book is on one computer and cancer research is on another. It’s that they’re all pulling from a shared, finite resource — AI compute — and right now, in 2026, there is not enough of it to go around.

This isn’t a hypothetical. AI companies are running out of capacity in real time. OpenAI scaled back its video-generation tool earlier this year just to keep its core services running. Anthropic’s users have been hitting usage caps faster than expected. Google’s CEO warned that compute limits were already constraining growth. The GPU chips that power all of this have wait times of 36 to 52 weeks — nearly a year just to get the hardware.

Your child’s bedtime story is standing in a line that stretches around the planet.

What’s actually in this line

I think about this a lot — the absurd democracy of the AI compute queue. If you could see the requests lined up alongside your child’s book, it would look something like this:

A radiologist’s AI assistant analyzing a mammogram for early-stage breast cancer. A pharmaceutical company running molecular simulations to find a drug candidate that could save thousands of lives. A teenager making a meme about a cat wearing a top hat. A developer asking an AI to write a sorting algorithm. Three separate startups generating marketing copy that nobody will read. A climate scientist modeling ocean temperature patterns. A college student asking an AI to explain the French Revolution. Someone generating 47 variations of a logo. A researcher training a model to predict protein structures.

And your kid’s book. Right there in the mix. Waiting its turn alongside all of it — the profound and the trivial, the lifesaving and the time-wasting, the meaningful and the absurd.

There’s no priority lane for importance. The meme and the mammogram wait in the same queue.

The token-maxing bros

And then there are the token-maxing bros.

If you’re not in tech, here’s what that means: in AI, a “token” is a unit of text or data that the model processes. Every word, every image prompt, every request eats tokens. And right now, in Silicon Valley, there is a competitive subculture of developers and AI enthusiasts who are essentially gobbling as many tokens as they can — running massive automated queries, stress-testing models, building elaborate AI agent chains that call AI to call AI to call AI, burning through compute at industrial scale not because they’re solving a problem, but because they can. Because it’s a flex. Because the leaderboard is measured in throughput.

I’m not here to judge what anyone does with their compute. But I am here to tell you that when your child’s book takes an hour instead of ten minutes, part of the reason is that the same pool of resources is being consumed by someone running recursive AI loops to see how many tokens they can burn in a day.

The demand for AI compute grew from about 6 million tokens per minute in October 2025 to roughly 15 billion tokens per minute by March 2026. That’s not a typo. That’s a 2,500x increase in six months. And the infrastructure — the physical data centers, the chips, the electricity, the cooling systems — simply cannot keep up.

What your book actually requires

Let me pull back the curtain on what happens when you hit “create” on Enchantably.

Your book isn’t one AI call. It’s many. The system generates a personalized story based on the arc you selected — your child’s name, their companion, the setting, the emotional journey. That’s one call. Then it generates a poem for each page, tailored to the story. Multiple calls. Then it generates an illustration for each page — and each illustration is its own compute-intensive request, because image generation requires significantly more processing power than text. Then there’s the layout, the dedication page, the cover.

A single book can involve dozens of individual AI requests, each one waiting its turn in that global queue. When the servers are busy — and in 2026, the servers are always busy — each of those requests takes a little longer. The delays compound. And what should take fifteen minutes stretches to forty-five, or an hour.

I wish it were faster. I’m constantly working on optimization — batching requests more efficiently, reducing redundant calls, making the pipeline leaner. But the fundamental constraint isn’t my code. It’s that the world wants more AI than the world has built the infrastructure to deliver.

The part that amazes me anyway

Here’s what I keep coming back to, even on the days when the generation times make me want to throw my laptop into the yard:

It works at all.

The fact that a parent in Virginia can open their phone, answer a few questions about their child, and receive — in under an hour — a fully illustrated, personalized, original storybook with their kid’s name on every page, a unique poem on every spread, and illustrations that look like they belong on a bookshop shelf?

That was science fiction three years ago. Not difficult. Not expensive. Impossible.

The same AI infrastructure that’s detecting cancer and discovering drugs is also generating a story about your child’s stuffed elephant navigating Tooth Town. And it’s doing it for the cost of an avocado toast. The sheer improbability of that — that these world-changing tools are available at this scale, at this price, to a solo founder building children’s books out of her kitchen — still knocks the wind out of me sometimes.

The wait time is real. I feel it. You feel it. But what’s happening during that wait is extraordinary: a global network of the most powerful computers ever built is painting a picture of your child riding a dragon, and the reason it takes a minute is because it’s simultaneously helping someone else fight cancer.

I can live with that. I think you can too.

What I’m doing about it

I don’t want to leave this at “the wait is worth it” and shrug. I’m actively working on making the experience faster:

I’m optimizing the generation pipeline to reduce the number of AI calls per book. I’m implementing smarter caching so that repeated elements don’t need to be regenerated from scratch. I’m monitoring the compute landscape and adjusting when I see capacity windows — yes, there are times of day when the queue is shorter, and I’m working to leverage that. And as AI infrastructure catches up to demand — more data centers coming online, more efficient chips, better memory — the generation times will come down. This is a when, not an if.

There’s one more thing I want to be transparent about. The image models I use for illustrations are top of the line — they’re the reason the books look as good as they do. But they’re also pre-GA, which is tech-speak for “not yet officially released to the general public.” I made a deliberate choice to use these models because the quality difference is night and day. The tradeoff is that pre-GA models come with lower service-level guarantees. They’re not running on the same battle-hardened infrastructure as a fully launched product. They’re faster some days, slower others. Occasionally they hiccup. It’s the price of being on the cutting edge rather than the safe middle — and for the quality of illustrations your child gets, I’ll take that tradeoff every time. But I want you to know it’s a factor.

In the meantime, I’d rather be honest about why it takes the time it takes than pretend the loading bar doesn’t exist.

A strange and wonderful moment in time

We’re living in a window — probably a short one — where the demand for AI has dramatically outstripped the supply. It’s a little like the early days of the internet, when you’d wait three minutes for a single image to load over a dial-up modem and your mom would pick up the phone and kill your connection. It was annoying. It was slow. And it was also the beginning of everything.

Your child’s book is being generated in the middle of a global compute crunch, alongside cancer research and meme generation and climate modeling and protein folding. It’s sharing a brain with the best and silliest of what humanity is building right now.

That’s not a bug. That’s the moment we’re in.

And your book — when it arrives, with your kid’s name on the cover and a story that was made just for them — is worth the wait.