Athos Omics AI—A MultiCloud Case Study

Athos Omics AI is building a software-as-a-service (SaaS) platform for other researchers doing genomic analysis to train on their models.

Athos Omics AI is using AI for various types of "omics"—a technical term referring to various biological and chemical studies. In Athos's case, omics include genomics, or the study of genes; proteomics, the study of proteins; metabolomics, the study of substances produced by metabolism; microbiome, the study of microbes, and more.

"We are very cost-conscious. We are a startup, and we are trying to save time and resources for our customers. So, we have to optimize different layers and combinations of services we provide," Guo said.

Athos Omics AI integrates these disparate data sets using NVIDIA HGXH100 GPUs on Vultr cloud and Dell Technologies infrastructure. Athos Omics AI turned to Vultr as a cloud provider offering a variety of benefits: scalability and agility to adapt to hardware upgrades, a user-friendly interface for simple interactions, cost efficiency by eliminating the need to build, operate and maintain their own data centers, engineering support and maintenance, disaster recovery, security and confidentiality and resource optimization.

But Vultr isn't Athos' only cloud partner. Vultr is part of a multi-cloud approach. The company uses AWS High Performance Computing (HPC) to generate intermediate files used as input for AI models. In addition, Athos uses Microsoft Azure to ensure its SaaS platform is cloud-neutral for its customers.

"We have to be able to train, test and infer everything and make sure everything runs, so we can provide these solutions to other companies to serve other customers," Guo said.

The company needs to understand its software and data structures and be able to explain costs and potential savings. Initially, Athos used two on-premises workstations for its analysis. "At the very beginning, this was fine," Guo said. But the company began to run into problems. "We had multiple power outages and instances where our buildings started leaking water."

Scalability became a concern, as growing data requirements demanded more—and more expensive—storage. To address these issues, Athos Omics AI needed to move to the cloud.

In selecting a cloud provider, Athos needed to go beyond GPU-as-a-service providers because the company needed access to more than just GPUs. It requires support for virtualization, containers, encryption and other security.

AWS provided those services and support for training, but AWS's pricing was prohibitively expensive and inflexible—Athos would have had to commit to locking down service agreements for a fixed number of years with AWS, while other providers would be willing to negotiate, Guo said. Transferring large volumes of data between AWS instances was costly as well.

Dell Technologies and Vultr together provide storage, GPU, Kubernetes application support, as well as disaster recovery, security and engineering support—all with flexible invoicing. Vultr offers engineering support and monitors vulnerabilities and patching.

AWS provides compliance with HIPAA, the European General Data Protection Regulation (GDPR) and California privacy regulations. "We have neither time and manpower to do these types of things. It's better to utilize existing services," Guo said.

With GPUs being extremely expensive, Athos Omics AI, like any company implementing AI, needs to be sure it's getting maximum usage from its processor investment.

When you send data from CPU to GPU, you want to be sure that the bus is full. You want to be sure that every single GPU is above 98% of utilization and RAM is almost full," Guo said.

Guo said that because AI and ML require huge data sets, organizations need to optimize the inference pipeline for maximum throughput. ML and AI also require specialized tools that use GPU hardware cores. "When we deploy AI models on the cloud, we have to think about how we optimize the whole pipeline, not just one part—data, model, inference pipeline, and utilizing the hardware," Guo said.

Looking ahead, Athos Omics AI is implementing agentic AI to help better explain its platform to users. "We want to democratize our platform to beginners and non-expert users. How can we do a better job explaining these concepts? How can we make this platform easier to use, guide our users and provide corresponding information so they don't have to search here and there?" Guo said. Agents could help users organize information and make decisions, not replacing humans but helping with decision-making. Likewise, agents can help users visualize data and integrate new information with existing knowledge.

Original Link: https://www.fierce-network.com/premium/research/1407307?pk=Vultr-RR-062025-Editorial