Sora shut down April 26 2026. Grok paywalled. Runway killed free video. Adobe Firefly removed credits. Every AI tool you currently use is on borrowed time unless its founder owns the GPUs.
I'm not exaggerating. I'm not hyping. I'm telling you the same thing every infra engineer who has ever looked at the unit economics of generative AI has been muttering at conferences for two years: the free tier you love is a marketing line item, and marketing line items get cut.
I'm Cemhan Biricik, founder of ZSky AI. This week we shipped something that might be the most important page on our entire site. It's not a feature. It's a promise.
It's called the AI Tool Refugee Center, and it exists because I'm tired of watching creators get evicted from their own creative practice.
The shutdown that broke me
When OpenAI announced Sora was sunsetting on April 26, 2026, my inbox lit up. Not with curiosity. With grief.
People had built workflows around Sora. Storyboards. Pitch decks. Client deliverables. A friend of mine used it for therapy — she has aphasia after a stroke and Sora let her describe scenes she couldn't draw or write. When the shutdown email landed, she texted me one line: "Where do I go now?"
I didn't have a clean answer. I had ZSky, sure, but I'd never built an explicit landing page for refugees. No migration guide. No "here's where your old prompts roughly map." Nothing that said we see you, this is hard, here's the door.
So we built it.
The unit-economics reckoning nobody wants to talk about
Let me show you the math that's keeping every AI video CEO awake at night.
Sora's reported infrastructure burn at peak was around $15 million per day. Lifetime consumer revenue, by most outside estimates, was around $2.1 million. That's not a rounding error. That's not "we'll grow into it." That's a structural impossibility wearing a free-tier costume.
Here's what nobody on the investor side will tell you out loud: most AI video generators are not businesses. They are demos that raised a Series A. The free tier exists to harvest training data, generate hype reels for the next funding round, and pump weekly active user charts that look great in a pitch deck. The moment the round closes — or worse, the moment the round fails to close — the free tier dies.
This is not a moral judgment. It's a balance sheet. If your cost per generation is $0.40 and your free tier serves 200 generations per user per month, every signup costs you $80 before they ever see a paywall. You can sustain that for 18 months on a war chest. You cannot sustain it forever. Nobody can. Not even Google, which is why Google Vids quietly stripped its free generation tier two months ago and nobody noticed because they're Google and the news cycle is busy.
Industry analysts I trust are now openly predicting 3 to 5 more major AI video shutdowns in the next 12 months. I think that's conservative.
Why ZSky won't be on the death timeline
I want to be very direct about this because it matters.
ZSky runs on 7 NVIDIA RTX 5090 GPUs that I own outright. No lease. No cloud credit. No "burst to AWS when traffic spikes." The silicon is in a rack in the United States. I paid for it with money I had. There is no burn rate to outrun because there is no burn.
I have no venture capital. Nobody on a board can vote to shut down the free tier to improve a metric for a Series B. There is no Series B. There is me, a small team, and a power bill.
26,000 users in 4 months, all served from those 7 cards. We're not the biggest. We're not trying to be. We're trying to be the one that's still here in 2028.
This is not a flex. It's a structural argument. When you ask "will this tool exist next year," the only answer that matters is: who owns the GPUs? If the answer is "a hyperscaler renting them to a startup that's burning $15M/day to look good for the next round," you already know how this story ends. You've seen it end nine times this year alone.
Introducing the Refugee Center
Go look: zsky.ai/refugees.html
It's a canonical home for displaced creators. If your tool died — or if you just woke up one morning and realized the writing is on the wall — there's a door here with your name on it.
What's on the page right now:
- Migration paths from 8 dead or dying tools — Sora, Grok, Google Vids, Runway, Pika, Kling, Luma, and Midjourney's video beta. Each one has a "what your old workflow looked like → what it looks like here" walkthrough.
- A free tier with no credit card — make stuff today. Watermarked, but yours. Forever.
- No vendor lock-in — every output you generate is yours, downloadable, no DRM, no subscription cliff.
- A founder you can email — me. I read everything. I respond to most of it. The reply might be slow, but it's a human.
- No model name marketing — I'm not going to wave brand names around to look impressive. The point is the output, not the recipe.
The Refugee Center isn't a campaign. It's a permanent fixture. As more tools shut down, we'll add more migration guides. The page is designed to outlive the news cycle.
Why a creator-led tool is more durable
I have aphantasia. I cannot picture images in my head. When I close my eyes I see darkness, not a beach, not my mother's face, not the sunset I watched yesterday. I learned this about myself a few years after a traumatic brain injury that took away most of what I'd built and most of how I thought.
Photography saved me. Holding a camera and capturing a thing I could see but not remember, pressing a button and getting a photograph back that I could look at — that was the first time after the injury that I felt like an artist again. Not a former artist. An artist now.
ZSky exists because I want everyone to have that moment. Specifically the people who have been told, in a thousand small ways, that they're not allowed to make beautiful things. Maybe because they can't draw. Maybe because they can't afford a $50/month subscription. Maybe because their hands shake. Maybe because the tool they used to use just shut down and the replacement costs more than their rent.
A creator-led tool is more durable than a VC-led tool because the creator's exit isn't an IPO. It's a body of work. I'm not optimizing for a liquidity event. I'm optimizing for the next ten years of being able to look people in the eye and say "yeah, the door is still open, come in."
You can read more about why ZSky exists if that resonates.
What to look for in any AI tool you adopt going forward
I want to give you something practical, because I think the next 12 months are going to hurt a lot of creators if they're not careful. Here's the list I would use if I were you:
Who owns the GPUs? If the answer is "we rent from AWS/GCP/Lambda Labs," that's not a deal-breaker but it's a yellow flag. If the answer is "a venture-backed startup whose last round was 18+ months ago," it's a red flag. If the answer is "the founder owns them," it's the green light.
What's the burn rate vs. revenue ratio? You can sometimes find this in TechCrunch profiles or interviews. If burn is more than 5x revenue, the free tier is a marketing budget and marketing budgets get cut.
Is there a credit card on the free tier? If yes, the company is preparing to convert you. That's fine, but understand the conversion event is coming.
Does the founder still post? Tools where the founder has gone quiet are tools where the founder is fundraising. Tools where the founder is fundraising are tools whose strategy may pivot in 90 days.
Can you export everything you've made? If the answer is no, you're not a customer, you're inventory.
Is there a written commitment to a permanent free tier? Most "free forever" promises evaporate with a board vote. A founder-owned tool with a published values doc is harder to renege on.
Where is the company located? US-hosted, founder-owned hardware sitting in a US data center is a structurally different risk profile than rented capacity in three regions with a stripe.com checkout in Delaware.
I'm not saying ZSky is the only tool that passes this list. I am saying we built the Refugee Center because we already know which tools won't.
What we're shipping next
This week alone we shipped 22 new landing pages, most of them migration-focused. Pages for "free Sora alternative," "Grok Imagine alternative," "Runway alternative," and so on. We're trying to be the page that comes up when somebody types "my AI tool just shut down what do I do" into a search engine at 2 AM.
Coming in the next two weeks:
- A bulk import tool for old prompts.
- A free conversion guide PDF.
- A shutdown-alert email list — we'll tell you when a tool you depend on shows signs of dying.
If you want any of that, the door is open at zsky.ai. No card. Just make stuff.
Closing
The free tier that survives isn't the one with the best free tier today. It's the one whose founder owns the silicon.
I built ZSky because everyone has the right to create beautiful things. That right doesn't expire when a venture round runs dry. It doesn't expire when a board votes to "rationalize the cost structure." It doesn't expire because you couldn't afford a $50/month subscription this month.
The Refugee Center is my way of saying: whatever happens out there, there's a door here with your name on it, and the lights are on, and the lights are going to stay on, because I own the building.
If you're a refugee from a shut-down tool, come in. If you're not yet but you can see the writing on the wall, come in early. If you just want to make something beautiful for the first time in your life, come in.
The kettle's on.
— Cemhan Biricik, Founder, ZSky AI
Top comments (0)