EP1. How Did Air Cloud Come Into Being?

Aieev
Jun 30
6 min read

Updated: Aug 17

Hello! 🐥I’m Yuna, a service planner on AIEEV’s Business Team, where I’m continuing to grow through hands-on project experience.

In this post, I’d like to share the planning journey behind Air Cloud—what kinds of challenges and considerations led us to launch this service.

I hope this story resonates with developers who build and utilize AI models, as well as planners designing user flows, and that it provides both empathy and a bit of practical insight. Let me take you through the birth of Air Cloud and its path to becoming a service.

1. How did Air Cloud start?

The cloud services we commonly know, such as AWS or Google Cloud, are centralized cloud services. This structure has limitations in terms of cost, scalability, and resource efficiency. Air Cloud is a “distributed cloud service” that started from these problems.

In Air Cloud, by using the Air Container function, anyone can easily upload and run their AI models or applications in the cloud. Traditional clouds had limitations such as high costs and complicated configurations that reduced resource efficiency. Here, I will share the planning process of Air Cloud that was designed to overcome these points.

In fact, when I was in the lab, our monthly cloud bills exceeded 15 million KRW (approx. USD 11,500), and I often stayed up for days expanding GPU servers and configuring them. Based on this personal experience, I conducted direct interviews with AI startups across different fields and confirmed that these difficulties were widely shared.

“Even though our monthly cloud expenses exceed 10 million KRW (around USD 7,700), whenever new customers increase, I end up staying up all night on marketing days worrying about server crashes.” — Chatbot Service (Company S)
“We need to scale, but our infrastructure can’t keep up.” — AI Education Service (Company T)
“As a startup, we don’t have infrastructure staff, nor the time or resources to manage it.” — AI Profile Service (Company S)

Many AI startups cannot avoid using cloud infrastructure, but they suffer from the burden of cost and operational complexity. Some teams relied on government support or promotional credits, while others attempted to run their own infrastructure—only to find it unsustainable in terms of cost and maintenance. Early-stage teams, in particular, faced a harsh reality: “We absolutely need infrastructure, but there’s simply no affordable way.”

That is why we planned Air Cloud—to solve these pain points and provide an environment where anyone with a good idea can bring AI services to life, even without abundant technical resources.

Air Cloud: a platform where developers can deploy quickly, without complex infrastructure, and at optimal cost.

2. The Planning Process for a User-Centered Service

Like most planners, when I first started working on this service, the biggest question in my mind was: “What do PRDs from similar services look like?” The toughest challenge was how to set priorities at the intersection of technical difficulty and user value.

To begin, we decided to clearly define our customer personas and then separate priorities step by step.

Because Air Cloud’s core differentiation lies in its distributed cloud foundation and cost efficiency, we set our primary target as early-stage AI startups—teams like ours—who feel the burden of infrastructure costs and complex setups when running AI services.

[Target Customer Personas]

AI-based startups that have just launched or are preparing to launch a product
Teams that have tested GPU cloud services but found them too costly for continuous use
Teams with limited development staff who face difficulties setting up a DevOps environment

Based on this, we set the direction of the product as a “service that lowers the entry barrier”, enabling easy and affordable deployment. Next, we analyzed competitors used by these customers to uncover additional pain points and adjust our priorities accordingly.

[Competitor Analysis]

At first, we simply listed competitor features. But as we reviewed real user feedback and interview data, we began to understand why users favored certain functions.

One important finding: users felt less psychological resistance when the deployment process was broken down into clear step-by-step stages. We made sure to apply this directly to Air Cloud.

(Because competitor-related data cannot be shared openly, please reach out via coffee chat or DM if you’d like to know more about our classification criteria and methodology. Just to give you a glimpse…)

From this analysis, we identified several key needs and defined our core functions accordingly:

Enterprise customers want to manage work at the project/organization level.
Users feel more comfortable with deployment when it is divided into step-by-step stages.
Some competitors offer cheap crowd-sourced rental pricing, but charge a fixed rate regardless of actual resource usage.
Providing incentives to long-term customers (e.g., enterprise users) increases the likelihood of continued usage.
Users prefer a real-time monitoring dashboard that directly shows metrics tied to billing, making infrastructure management easier.

[Core Function Definition] Rather than just listing functions, we aimed to capture the full cycle: user-centered problem definition → practical deployment flow design → cost strategy → feedback-based improvements.

Team-Based Collaboration Space

Air Cloud’s core target customers run projects as “teams.” They need to separate and manage work by project and by member.

So, we structured the platform as a two-level hierarchy: Organization → Project. Roles are divided between Owner and Member, with granular access rights per team. This UX was optimized for collaborative team workflows.

2. Pricing Optimization Strategy

Most startups cited GPU costs as their biggest burden. Especially for always-on services, expenses accumulate quickly. To address this, our business and development teams worked together to design ways to reduce costs across different usage scenarios.

As a result, we introduced three cost optimization features:

Case 1: For services that must run continuously→ Auto-Scaling: When container load increases, GPU instances automatically scale up; when load decreases, they scale down.
Case 2: For customers needing continuous GPU infrastructure→ Bulk Credit: Enterprise or repeat-use customers can purchase large amounts of credits upfront.
Case 3: For customers requiring fixed resources over long periods→ Saving Plan: Up to 30% discount for customers committing to more than 6 months of fixed usage.

3. Step-by-Step Deployment Structure

In customer research, we found users felt more comfortable when deployment was split into stages. To apply this, we designed Air Cloud’s deployment in five steps, making it easy even for users without DevOps experience.

Once deployment is complete, the container is provided with a Serving Endpoint URL, enabling external integration without additional network configuration.

4. Monitoring Metrics Design

Even after deployment, startups are highly sensitive about performance and billing data. Especially in an auto-scaling environment, unexpected charges can cause anxiety. That’s why we focused on designing a transparent, real-time dashboard with visualized resource and cost metrics.

Air Cloud provides a graph-based dashboard with 11 key metrics. Since we tried to include every metric mentioned during interviews with target customers, the dashboard became extensive—so we worked continuously to refine the design for clarity.

As a result, the dashboard now supports not only performance diagnostics but also resource optimization and cost management, addressing the needs of cost-conscious early-stage users.

👉 In the next episode, I will share in detail the PRD’s USER FLOW stage.

3. In Closing

When I first joined AIEEV and began working on the Air Cloud project, what I felt most strongly was a sense of uncertainty—where should I even start? During competitor research, I constantly asked myself: “What standards should I use for comparison?”, “Which functions are worth referencing?” I also remember how we conducted numerous customer meetings to decide on key features.

In meetings with the development team, we repeatedly discovered missing elements. Each time, I would go back to reorganize flowcharts and revise the PRD document dozens of times. I was always curious about how other planners might set their priorities—would they define things more strictly, or in greater detail? Today, I wanted to share my own first experience with fellow planners and startup colleagues.

For customers interested in trying out Air Cloud—the service I helped plan—we are offering a special promotion: two weeks of free PoC validation, plus consulting and training on cloud optimization.

If you sign up within 10 seconds here https://www.aieev.com/contact, we’ll reach out individually within one day.