Paicon

Paicon Accelerates Global Cancer Diagnostics with Vast.ai's High-Performance GPU Cloud

How a German AI-driven oncology company uses Vast.ai to train foundation models on large medical datasets while reducing GPU compute costs by more than 70% compared to hyperscale cloud providers.

70%+

Ongoing Cost Reduction

Agility

Run more experiments in parallel

Speed

Large foundation-model training jobs ran smoothly

Flexibility

Support multinational research footprint

Overview

Industry: Bio Tech
Key area: Medical imaging and machine learning

Paicon is a biotechnology and artificial intelligence company headquartered in Germany, focused on advancing AI-driven cancer diagnostics and pathology analysis.

The company develops deep learning models that identify and classify cancerous tissue from digital pathology slides, helping reduce diagnostic time and improve precision for oncologists and pathologists.

Paicon operates at the intersection of medical imaging and machine learning, developing both application-specific models and large foundation models for oncology, alongside data platforms that unify highly heterogeneous datasets sourced globally.

Founded by Kristin ("Chris") Eichmuller, CTO & Co-Founder, and his wife and co-founder, Dr. Manasi Eichmuller, CEO, the company collaborates with research groups and hospitals in India, Europe, Africa, Cambodia, and the United States, giving it access to one of the most geographically and technically diverse pathology datasets in the world.

Challenge

Paicon's mission to democratize cancer diagnostics comes with extraordinary data and compute challenges. Their research involves large-scale pathology datasets and multi-cancer model development — including work on colorectal, breast, and prostate cancer.

These workloads require substantial GPU compute to support:

Training and fine-tuning computer-vision models across globally diverse datasets
Running hundreds of experiments on new architectures and modeling approaches
Developing large-language-model (LLM) systems for diagnostic interpretation and pathology education

While AWS offered a robust ecosystem for production deployments and inference, its high-end GPU instances proved cost-prohibitive and frequently unavailable in Europe, making large-scale experimentation impractical.

Additionally, European data sovereignty considerations, particularly for medical applications, made reliance on U.S.-centric cloud regions less attractive. Paicon needed compute resources that were powerful, geographically flexible, and economically viable — without sacrificing operational control.

"At some point we said, for production workloads AWS is fine — but for experimentation, AWS is simply not acceptable. The bills are gigantic."

Solution

Paicon's engineering team found Vast.ai's setup intuitive, with straightforward access to large-machine configurations and predictable pricing.

Vast.ai's distributed data-center model also aligned well with Paicon's European operating requirements, supporting regional deployment strategies and reducing dependence on a single centralized cloud region.

"With Vast.ai, we can move from 4 GPUs to 8 GPUs instantly and stay within budget. That flexibility lets us shoot at more targets and move faster."

See How Vast.ai can Transform AI for You

Discover how Vast.ai powers AI innovation. Our team will guide you through a tailored platform walkthrough and show how we combine enterprise-level capabilities with startup-friendly pricing.

Results

70%+ Cost Reduction

By shifting large training workloads to Vast.ai, Paicon reduced the cost of high-end GPU compute by more than 70% compared to equivalent AWS configurations. An 8x H200-class setup priced around $80-85/hr on AWS was available on Vast.ai for approximately $25/hr, making large-scale experimentation economically viable.

Greater Experimental Agility

Vast.ai's on-demand access to large GPU instances allowed Paicon to run more experiments in parallel, explore new architectures, and iterate faster — without being constrained by availability limits or runaway cloud costs.

Faster Training Cycles

Large foundation-model training jobs ran smoothly across large, heterogeneous datasets. Early training runs demonstrated strong cross-tissue generalization after just two epochs, validating Paicon's approach to foundation-model development.

Global Collaboration & Regional Flexibility

Vast.ai's distributed cloud architecture supports Paicon's multinational research footprint, enabling regional deployment strategies and reducing dependence on a single centralized cloud region — an important consideration for sensitive medical research.

Beyond Pathology: Democratizing Diagnostics

Using Vast.ai also enabled Paicon to develop a large-language model (LLM) trained on decades of pathology data and expert discussions from a long-running global pathology forum.

The dataset — containing over one million images and expert diagnostic discussions — was used to fine-tune a public base model into a specialized LLM capable of assisting with pathological image interpretation and terminology clarification.

"Because of Vast.ai's GPU infrastructure, we could fine-tune larger language models and reach higher performance. It's now part of our product roadmap."

Why Paicon Stands Out

Paicon's innovation lies not only in its technology but in its global and responsible approach to AI in medicine.

By aggregating data from diverse populations and laboratories — from India to Africa to Europe — the company builds robust, bias-resistant cancer AI models that generalize across ethnic, genetic, and technical differences.

As a German-based medical AI company, Paicon also places strong emphasis on data sovereignty and regional control, benefiting from Vast.ai's globally distributed compute model.

"For medical applications in Europe, data sovereignty is a huge concern. Vast.ai's distributed architecture gives us more flexibility while scaling globally."

Looking Ahead

Paicon plans to continue using Vast.ai for foundation-model research, expanding into additional cancer types, and scaling its diagnostic AI offerings internationally.

About Paicon

Paicon is a Germany-based medical AI company developing deep-learning models for oncology and pathology diagnostics. Paicon collaborates with hospitals and research institutions worldwide to build scalable, explainable, and bias-resistant diagnostic systems.

Focus areas include: