How TOOL Scales Big AI Ideas on Runpod

Problem

For a creative studio, AI infrastructure complexity can kill creative momentum. 

TOOL is a creative studio working across content, AI and experiential. With Fortune 1000 clients including AT&T, AMD, Coca-Cola, and Miro, TOOL has built its reputation on providing creative production solutions across content and experiential leveraging AI.  They describe themselves as "more of a speedboat than a cruise liner" - able to pivot fast, yet trusted enough that the world's biggest brands let them take calculated creative risks and advise on AI content production solutions.

But as AI became central to their work, TOOL's infrastructure couldn't keep pace with their ambition. The team found themselves managing a patchwork of local machines, each with its own friction.

Beyond the expense, the complexity was staggering: port settings, security gateways, HTTPS configuration, load balancers. Getting a GPU instance actually running required significant DevOps expertise that distracted from creative work. On Runpod, that setup goes from days to hours, there are no ports to configure, no load balancers to stand up, no security gateways between the idea and the work. What’s more, always-on billing models created constant anxiety. Instances ran 24/7 whether the team needed them or not. 

The most tangible symptom of the problem? TOOL was literally transporting cases of GPUs through airports to run workloads on client sites. Wim Vanhenden, TOOL's Technical Director, still has the custom travel cases in his office, including one specifically designed for transporting 4090 GPUs.

"We had physical machines traveling for experiences. They break. They need to be repaired. You get a call at 3AM: the computer broke, what are we doing?"

Solution

Runpod became the GPU backbone for every AI-powered creative project.

TOOL transitioned all GPU workloads to Runpod, using both Pods and Runpod Serverless depending on project needs. AWS and GCP still handle APIs, CloudFront, and Lambda proxies – the team no longer uses those platforms for GPU work.

"We use a lot of AWS and GCP, but not for GPU or anything related to GPUs. Whenever we touch a GPU, we're not going there. When it comes to GPU, we're using Runpod," says Wim.

Prototype in ComfyUI, Dockerize the workflow, deploy as an API on Runpod, proxy through AWS for client delivery. Scripts automatically move outputs to Google Drive so directors and non-technical team members can review renders without ever logging into a terminal.

Technical depth varies across the team, and the setup accommodates it. Some team members run ComfyUI independently on Runpod instances. For others, TOOL created five-minute internal tutorial videos showing how to spin up machines.

Runpod powers the full range of TOOL's AI-driven production work. For an AMD campaign, the team ran hours of 4K video scaling using Wan 2.1 across multiple parallel instances. They handle LoRA training through the Kohya SS GUI. Background removal and segmentation APIs tackle essential preprocessing work.

Another standout project: a Pearson campaign featuring real-time avatar creation, where people with difficult backgrounds – including individuals who had experienced homelessness as children – interacted with AI versions of their younger selves. The technical challenges of real-time avatar generation required infrastructure that wouldn't fail during a physical shoot in New Orleans.

"We used many, many hours of Runpod instances creating training videos for those avatars, all done in Wan 2.1. I was emotional when I saw it the first time," Wim says.

‍

Result

Peace of mind, instant scalability, and no more 3 a.m. calls about broken hardware.

The transformation extended beyond technical improvements to operational confidence. TOOL eliminated the physical logistics of transporting GPU hardware entirely.

Scalability became a creative unlock rather than a constraint. When a video render faced a 27-hour processing time, TOOL simply spun up 27 parallel instances. Access to H200s enables projects that would have been impossible on their previous infrastructure.

Serverless eliminated the cost anxiety that plagued their AWS days. On workloads migrated from dedicated machines to Runpod Serverless, costs dropped 50–60% with savings varying per project. And when a project wraps, the spend stops.

"When you don't need it, you don't run it, and you don't pay for it. That's all gone now."

But the real win is reliability. Working from Belgium while projects run in the US, Wim can finally rest easy.

"I'm in a different time zone. I can literally sleep now instead of getting a call at 3AM. It's on Runpod. It doesn't break. It's fine."

For a creative and tech studio whose reputation depends on delivering breakthrough work on deadline, that peace of mind is everything.

"All of these projects, the renders for AMD, the Coca-Cola builds,

that has to do with scalability. If we can't scale, we can't deliver. Runpod makes that possible."

‍

Conclusion

Infrastructure is no longer the limiting factor.

For TOOL, the ability to dream big ideas with clients depends on knowing they can actually execute them. Partners like Runpod remove the uncertainty.

"Having partners like Runpod where we can dream those big ideas and know it's going to happen in the way we're presenting it and selling it to our clients is massive for us.," says Wim.

As TOOL continues pushing creative boundaries for the world's biggest brands, their infrastructure is no longer the limiting factor. The speedboat can move as fast as the ideas demand.

‍