Infer.bid. The GPU Marketplace for the AI Era
UK 2025 · WEB APP DESIGN , LANDING PAGE , DESIGN SYSTEM , AND GRAPHIC DESIGN AND ART DIRECTION
We transformed Infer.bid from a backend-driven prototype into a complete marketplace experience. Alongside the core platform, we created a scalable design system, visual identity, LLM asset library, and marketing website, giving the founder a cohesive product ready for testing, growth, and future expansion.
Infer.bid is a platform with a bold promise: to connect people who need GPU compute (for AI models) with those who own spare GPU power.
When the founder first approached us, he already had a working backend built entirely on his own. The system could pair compute requests with providers, run LLMs, and track performance. But it had no UI. No UX. Just a command-line terminal and raw logic.
He told us, “I’m a tech guy and I know nothing about design. Just make it usable.” So, our goal was to help him transform a technical prototype into a product that could scale — one that worked intuitively for two very different types of users.
Mapping the Users
Through a series of workshops and user flow mapping sessions, we identified two distinct user personas at the heart of the product. On one side were the providers — GPU owners ranging from gamers and ex-miners to small-scale data centre operators. On the other side were the consumers, typically AI startup founders, indie developers, or researchers.
These two groups came with different priorities, expectations, and levels of technical fluency. Our task was to bring them together into a single, coherent platform — one that respected their unique needs while delivering a unified and intuitive user experience.
"The most impressive thing was The Gradient's knowledge of using AI tools properly, and to their advantage."
Platform design
We built a clean, scalable interface that made navigating hundreds of LLMs intuitive — even for first-time users. A real-time GPU Market dashboard gave providers insight into earnings, while a model ranking system helped both sides track what’s trending based on actual usage.
We designed two focused, frictionless flows — one for those renting compute, the other for those offering it. Providers could deploy their GPUs in under five minutes via terminal, set prices manually or based on AI suggestions, test new cards automatically, and monitor earnings through a clean dashboard. Power users could create GPU clusters, schedule availability (e.g. only at night), and adjust pricing dynamically based on demand.
We also supported hybrid users — people who rented high-end GPUs while monetising their own hardware. Switching roles was seamless, with no duplicated accounts or clunky toggles. The experience stayed unified, even as usage patterns blurred.
Prototyping in Replit
We delivered a clickable prototype in Replit (hello, early 2025) within the first week, helping validate product assumptions early. Skipping wireframes allowed us to work directly with real interface logic, and more importantly, to load real data into the system using the APIs. Many of the LLM configurations and marketplace mechanics were complex, so seeing actual model names, pricing tiers, and token behaviour inside the prototype made it feel closer to a working product than just a mockup. From there, the platform continued to evolve.
Video generation with Kling & Seedance
To bring movement and atmosphere to the landing page, we used Kling AI and Seedance to produce dynamic, AI-generated videos. These weren’t decorative fillers — they extended the product’s narrative, visually reflecting the technical depth of the platform. Each video loop was rooted in the model’s logic and vibe: from abstracted matrix-like flows to digital cloudscapes.
Prompt guide for scaling
Finally, we provided the client with a step-by-step prompt guide — a framework they could use to maintain and expand the visual system independently. This became the foundation for sustainable growth.
Art direction & identity
We partnered with Efficy to craft a visual language that balanced minimal UI with the expressive aesthetic of AI-generated art. We generated visuals and refined them manually, giving each LLM family a distinctive visual style. We also built a full LLM visual library with custom thumbnails for 200+ models, making the interface not just functional but also explorable.
