Quality assurance · Production

Climate tech startups Orbital Materials and Hum.AI train foundation models for climate solutions with Amazon SageMaker HyperPod

The problem

Climate tech startups need to train complex foundation models on vast and diverse environmental datasets but face high infrastructure complexity, cost, and fault-tolerance challenges at GPU scale that slow down innovation.

Workflow diagram · grounded in source
1
Environmental data collection
trigger
“They're training on 50 years of historic data collected by satellites, which amounts to thousands of petabytes of data”
2
Orb diffusion model training
ai_action
“Orb is a diffusion model that Orbital Materials trained from scratch using SageMaker HyperPod”
3
AI-suggested material design
ai_action
“They've released a generative AI model called "Orb" that suggests new material design, which the team then tests and perfects in the lab”
4
Lab validation of AI designs
human_review
“They've released a generative AI model called "Orb" that suggests new material design, which the team then tests and perfects in the lab”
5
VAE-GAN earth observation model
ai_action
“Hum.AI's FM architecture employs a variational autoencoder (VAE) and generative adversarial network (GAN) hybrid design, specifically optimized for satellite imagery analysis”
6
Cluster monitoring and node recovery
validation
“The SageMaker HyperPod monitoring agent continually monitors and detects potential issues, including memory exhaustion, disk failures, GPU anomalies, kernel deadlocks, container runtime issues, and out-of-memory (OOM) crashes. Based on t…”
7
Auto-resume from checkpoint
feedback_loop
“the SageMaker HyperPod auto-resume feature that automatically resumes a training run from the latest checkpoint, provides training continuity, even through node failures”
8
Carbon capture sorbent output
output
“The first product the startup designed with Orb is a sorbent for carbon capture in direct air capture facilities”
Reported outcome

Orbital Materials achieved a tenfold improvement in material performance and Hum.AI achieved the ability to see underwater from space for the first time; SageMaker HyperPod enabled both teams to train larger models faster and reduced operational overhead.

Reported metrics
Material performance improvementtenfold improvement
Development speed vs traditionalan order of magnitude faster than traditional development
Operational overhead reductionup to 40%
Time to begin training environmental modelsminutes rather than weeks
Show all 5 reported metrics
material performance improvementtenfold improvement
development speed vs traditionalan order of magnitude faster than traditional development
operational overhead reductionup to 40%
time to begin training environmental modelsminutes rather than weeks
cost saved from checkpoint recoverysaves thousands in lost progress between checkpoints
Reported stack
Amazon SageMaker HyperPodAmazon EKSAmazon CloudWatch Container InsightsAmazon Managed Service for PrometheusAmazon Managed Service for GrafanaAWS TrainiumAmazon BedrockAmazon SageMakerAmazon SageMaker JumpStartAWS MarketplaceAmazon EC2SlurmLlama 7B
Source
https://aws.amazon.com/blogs/machine-learning/how-climate-tech-startups-are-building-foundation-models-with-amazon-sagemaker-hyperpod?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Orbital Materials achieved a tenfold improvement in material performance and Hum.AI achieved the ability to see underwater from space for the first time; SageMaker HyperPod enabled both teams to train larger models fa…

What tools did this team use?

Amazon SageMaker HyperPod, Amazon EKS, Amazon CloudWatch Container Insights, Amazon Managed Service for Prometheus, Amazon Managed Service for Grafana, AWS Trainium, Amazon Bedrock, Amazon SageMaker, Amazon SageMaker JumpStart, AWS Marketplace.

What results were reported?

Material performance improvement: tenfold improvement; Development speed vs traditional: an order of magnitude faster than traditional development; Operational overhead reduction: up to 40%; Time to begin training environmental models: minutes rather than weeks (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Environmental data collection → Orb diffusion model training → AI-suggested material design → Lab validation of AI designs → VAE-GAN earth observation model → Cluster monitoring and node recovery → Auto-resume from checkpoint → Carbon capture sorbent output.

Related quality assurance cases