Akool

$25M Revenue in 2025 — From a Flat $100K Year to #1 on the Inc. 5000

November 1, 2026

$25M (2025)

Recognized revenue

20–30K

Paying customers

2022

Full-time since

60–70

Team (~40 engineers)

~$10M raised

Funding

What Akool Does

Akool is an AI video studio. You type or upload, and it makes the video.

Swaps: face, head, hair, motion or a whole character in a video
Avatars: talking avatars, live ones that chat back, or a copy of you
Translation: your video, in another language, in your voice
Agentic Canvas: AI agents that build long videos from a text prompt

Akool homepage: banners for Agentic AI Canvas, an AI film creator contest and Seedance 2.5, above a grid of tools including Face Swap, Motion-Swap, Talking Photo, Character Studio and Image Generator — The Akool homepage (November 2026). It opens straight into the tools — no hero headline, just a grid of things to make.

Prosumers pay by the month. Big companies pay by usage, often through the API. Avatars are the top seller with businesses.

The Problem They Solve

Video is the hardest content to make. You need cameras, actors, editors and time.

Jeff Lu had spent 10+ years on that problem:

Microsoft
Apple, working on Face ID
Google Cloud, working on video

By 2022 he could see AI video was getting good. Google wasn't ready to ship it. AI video can't be 100% controlled, and Google wanted full control. So Jeff quit and went all in on his side project.

The scoreboard in July 2026:

~$25M recognized revenue in 2025
$3M+ in June 2026 alone
~$60M expected for 2026
~500,000 images and videos made per day
#1 on the 2025 Inc. 5000 (37,364% growth, 2021–2024)

And Jeff counts it the strict way. Asked if $60M was ARR:

“Revenue, not ARR.”

Watch: How Akool Grew from $100K to $25M Revenue (Nathan Latka interviews Jeff Lu)

The Growth Story

2021–2022: two flat years

Jeff wrote the first code around 2020–2021, part-time, during COVID.

2021: about $100K in revenue
2022: he quits Google. Still about $100K.

Why flat? The team spent most of its time chasing big B2B contracts. Those deals took forever to close.

2023: add a checkout

Meanwhile, people kept finding the tool and using it on their own.

So they put a payment page on it. People started paying online, with no sales call. That year Akool did $2–3M.

The early users came from places where people already make videos:

Demo posts in Reddit communities, Facebook groups and Discord
1:1 calls with video creators to hear what they needed
Freelancers on Fiverr selling video work, who could do it faster with Akool

Then a giant brand showed up. Latka's notes say it was Coca-Cola, running a 6-month campaign that swapped millions of faces. Jeff didn't name them. He said:

They came to Akool. There was no sales team to find them.
They hid who they were at first.
They spent about 3 months testing tools before picking Akool.

Akool's own site has a Coca-Cola case study: a browser game where players swap their face onto a game character.

2024: the first salesperson

Akool hired its first account executive in 2024. Revenue that year: $12–13M.

Today there are 4–5 reps. Each carries a $1M quota.

2025–2026: enterprise takes over

2025: ~$25M recognized revenue
20,000–30,000 paying customers, mostly prosumers
Several hundred business accounts — where most new growth comes from
Biggest contract: about $4M a year, usage-based, through the API

All of this on about $10M raised in 4+ years. The last priced round was around a $100M valuation in 2024. Jeff thinks a new round could land near half a billion.

Would he sell to Google for $500M cash?

“Likely I will.”

What's next: the 1-hour movie

Short clips are mostly solved. Long videos are not.

Making long videos that still look real is hard
Sticking to the prompt for the whole video is hard
Making it easy enough for anyone is the last gap

That's the bet behind Agentic Canvas. Jeff thinks people will soon make movie-quality videos by themselves.

Akool revenue journey — about $100K in 2021 and 2022, $2-3M in 2023, $12-13M in 2024, about $25M in 2025 and about $60M expected in 2026

Key Growth Tactics

Lots of tools can do a face swap. Here's what Akool did differently.

Small models, one job each. For swaps, Akool designs and trains its own models from scratch. Each model does one task, like making a human in a video. So it's much smaller than a do-everything model, and cheaper to run.
Rent the GPUs. No server room. Akool trains on rented cloud servers and uses Lambda a lot.
Let people pay without talking to you. The checkout did more in one year than two years of chasing contracts.
Win the benchmark. Big brands test for months. The best results close the deal.
Price on usage. When a customer wires Akool into its own software, API calls grow with them. That's how one account reached ~$4M a year.
Count revenue the strict way. Jeff reports money actually earned over 12 months, not last month times 12. Latka called it the fastest way to spot a technical founder.

What didn't work:

Chasing big contracts first — two years stuck near $100K
A pricing page with “so many sliders” that even Latka got lost — Akool just added a Starter plan

Three plays behind Akool's growth — small single-task video models, a self-serve checkout, and letting self-serve feed enterprise deals

Key Takeaways for Builders

✓You don’t need a giant model to win. Akool trains small video models for one job each. Better results on that job, and a cost it can control.
✓If users keep showing up, let them pay. Big contracts stalled for a year. A simple checkout took Akool from ~$100K to $2–3M.
✓Go where people already do the work. Akool posted demos in Reddit, Facebook and Discord groups, and talked 1:1 with video freelancers on Fiverr.
✓Self-serve can feed enterprise. The first big deal was inbound. Akool hired its first sales rep only in 2024, at $12–13M in revenue.
✓Usage-based pricing lets big accounts grow on their own. The largest customer plugs in through the API and pays about $4M a year.