Atlas Cloud Bets on AI Media: Compute Once, Earn Forever
How Atlas Cloud moved up the stack
Tang launched the company in July 2024 with plans to become a neocloud that rents GPUs. However, after a year with little traction, he pivoted to the software layer. Today, Atlas Cloud offers APIs to hundreds of video, image, language and audio models behind a single endpoint. In fact, Tang compares it to Expedia. The GPU clouds are the rental car companies, and his platform gives you access to all of them. Meanwhile, the company does not run its own data centers. Instead, it leases GPUs from neoclouds when it needs compute. As a result, Tang expects raw GPU rental to become a commodity once chips are no longer scarce.Media customers came first
After launching in November 2025, the team noticed that 70% to 80% of its users came from media. For instance, ad agencies, marketing agencies and short film studios now use it to build content. A team can also turn an interview into a short promo by using language, image and video models together. Some customers are even producing full-length films. Therefore, Atlas Cloud sharpened its focus on that audience.Compute once, earn forever
Tang argues that AI is splitting software into two accounting problems. For Atlas Cloud, the difference matters. On one hand, agents and AI services burn tokens every time a customer acts, so compute stays an ongoing expense. On the other hand, media works the opposite way. You spend the compute once and can then earn on the content for years, much like a classic TV series. Looking ahead, Tang expects AI video to grow as large as the LLM market as the models improve. He believes Atlas Cloud is well placed for that growth. To learn more, visit atlascloud.ai.Transcript
Hi, everyone. " My next guest is Mr. Jerry Tang.
" Jerry, how are you? It's nice to have you on here. I feel great.
Thanks for the opportunity, Alan. My pleasure. So Jerry, I mentioned you're the CEO of Atlas Cloud, and we're going to get into who and what Atlas Cloud is in a moment.
But before we do, let's tell people a little bit about you beyond, obviously you're the CEO, as I said, but what else should people know about you? So Alan, I have a financial background. I have been a banker for 14 years before I started Atlas Cloud.
Many people in the tech industry might say, you're an impossible founder doing finance for 14 years, and now you run an AI company catering to media content creators. Mm-hmm. So my background- I would imagine it...
So what made you do that, Jerry? Tell us the story of that. Yeah.
My background was in engineering. My undergrad school, I went to Columbia. For one reason or another, I got into banking and finance as my first job, so it was in commercial real estate.
I really enjoyed doing all that. As a matter of fact, I used some of the things I learned as an engineer to improve the process. I was joking that I was the smartest mathematician in real estate finance in New York.
Very cool. Now, so tell us how that led to Atlas Cloud. Yeah.
So I was looking at building data centers back in 2023, 2024, and I saw these GPU cloud companies popping up, serving GPUs and compute to AI companies at the time. So I thought- Neoclouds ... Neoclouds, exactly, Alan.
" The ROI at the time was 10 months, 12 months. I thought that was a very attractive business. And?
Is it? And July 2024, I started Atlas Cloud. I thought I was going to become a Neocloud.
I tried for 12 months, and I didn't get anywhere. I did a couple deals. That was it.
And then I decided to pivot into the SaaS layer, the IA model-as-a-service or IA infrastructure-as-a-service layer. I recruited 30, 40 engineers to build all the capabilities that we have today, to provide APIs to all AI models you can use for media content creation. So- So what you did is, as we call that, moving upstack, right?
Moving upstack, yeah. You moved up the stack from a pure infrastructure layer to, via API and maybe MCPs, offering connectivity to the models themselves, which aren't running on your infrastructure per se, but someone could build on your infrastructure and then hook into these. So, sitting on top of infrastructure-as-a-service is platform-as-a-service, right?
And this is almost more like a platform play, where now I give you the platform with the APIs, and you can go build your app on top of that. Yeah. I just thought of an analogy, Alan.
Yeah, that's exactly right. So let's say the GPU cloud companies are the Hertz, the rental companies. We are almost like Expedia.
We can give you access to all the rental car companies. Got it. Other people are calling that kind of thing a harness or a super harness, where you plug into my harness, and at the back, you could plug into multiple models and companies, and applications.
Yeah. And then do you actually operate your own data center, or are you taking space in other data centers, or how does that work? Yeah, we are taking space from other data center.
As a matter of fact, when we need GPU compute, we'll go out to one of the Neoclouds. Even though our name sounds like a Neocloud company, we'll go to them, and we'll lease X number of GPUs for our compute needs. Right.
Well, because that, to me, that's destined to become commodity, the Neocloud, right? Because they're trading in scarcity. Once GPUs aren't as scarce as they are now, it's kind of a commoditized thing, right?
I rent GPU and infrastructure, data center space from you. So what are the kinds of things people are building with Atlas Cloud? So Alan, when I got into this, I didn't think of ourselves as catering to the media content crowd.
We thought we are a platform providing all sorts of AI models, LLM, picture, video, sound models. But what's interesting is that when we launched in November 2025, a couple of months shortly after, we discovered that the biggest users are all coming from media space. I'll give you an example.
Higgs Field, for example, is one of our clients. They're on our website. They're one of the largest media content app player in the space.
They are one of our clients. So they create videos for ads for marketing purposes. So we decided to actually focus on the specific niche after we found out 70, 80% of our clients are from the space.
Really? And they just came to you. It wasn't like you set out to go after them, per se.
They came to us, and we took that as a sort of a market signal, a calling. We decided to really sharpen our focus just for these audiences. So what are they trying to...
What are they getting from you, Jerry? So, I'll give you an example. If you want to create, let's say, Alan, you have a show, right?
This interview, you can use AI to generate a 30-second marketing video out of it. Sure. And then you can come up with some prompts or script using LLM, and then you can even use picture models to generate some storyboards, and then you create this video to market this particular interview.
So it's an application, Alan, that uses every modality you can think of. So you developed... Okay, so now I'm seeing it.
So you developed an application plugging into multiple models and apps on the back end that allow people to create content for their media business. Yes. Now, so we use several AI tools here at Techstrong, right?
We use one tool that'll take this interview that you and I are doing and create literally dozens of different social media optimized cuts, if you will, of this interview, summaries and stuff like that. Is that the kind of stuff you're doing as well, you think, you'd say? Yes.
We supported that. So let's say whatever app you're using, they would call APIs from us- Mm-hmm. Got it ...
to produce better results. Yeah. All right.
And Atlas Cloud, what's the URL for that? ai. ai.
Okay. All right. Jerry, let me pivot over to here with you ...
they had talked to us about, which is AI is splitting software into two different accounting problems. Compute once, earn forever. What do you mean by that?
So there's a lot of discussion lately about token consumption or token overconsumption, right? You even hear companies going bankrupt. AI.
They were burning $9 million every month. Right? So what I think my hypothesis is media is a very interesting space.
Here's why. If you produce a piece of content, you can potentially earn money on that content forever. Right?
" It was produced in the '90s. They spent a total of $250 million. They have been earning money for the last 30-plus years, probably made over $7 billion.
So I think media is a very interesting space because, once you make the content once, you can make money off it forever. Versus the other applications. In order to produce a result and outcomes, you need to keep burning the tokens.
Sure. So, but what's the two different accounting problems then? Well, so for this, for media content, you can even amortize, right, the initial cost over time, even though you already spent money.
For other type of applications, you have an ongoing OPEX line item on the accounting system. I got it. All right.
Makes a lot of sense. Jerry, what kind of media companies are customers of yours? So we have ad agencies, marketing agencies, so they make UGC, they make high-quality ad contents.
There are also short film studios. They make these sort of one-minute episodes, maybe 90 episodes, all together, 90 minute. And nowadays, I even see 20 minutes, 30 minutes feature videos that are getting millions of views on YouTube.
All with AI. The latest, I guess, engagement I have is with a film production studio, making full-length, 90-minute, for theater films. So that's getting started as well.
Excellent. ai? ai.
ai. Oh, just Atlas. ai.
Yeah. ai. Yeah.
Excellent. So Jerry, with so much going on, where do you see it going from here? So, versus the LLM space, the video content space is at a baby infant stage, right?
LLM, you add up all the ARRs, is probably $200 billion, $250 billion right now. The ARR in the video space is probably $2, 3, 4 billion. I believe it's going to be as big, if not bigger, than LLM space.
So I see in the next 18 months, 36 months, there will be exponential growth as the model capability's getting better and better. Yeah. Agreed.
We do some of it here, so I'm seeing it firsthand. So Jerry, a long while away from doing finance, huh? Yeah, but it's a long way away, but the finance background gave me a lot of rigor and discipline, right, when running the business.
However, I'm a guy, like to learn new things, and being outsider in the media space may give an edge, right, to look at things differently. Fair enough. All right.
Well, Jerry, thanks for coming here on Techstrong TV. I wish you a lot of success with Atlas Cloud, and come back and keep us posted. Thanks, Alan.
Thank you. Jerry Tang, Atlas Cloud, here on Techstrong TV. We're going to take a break.
We'll be right back.