Learn how LEIP accelerates your time to deployment Schedule a demo


Reduce Your AI/ML Cloud Services Costs with Latent AI

by Mark Painter | Posted Feb 21, 2023
AWS SageMaker / LatentAI

Cloud computing offers more operational flexibility than privately maintained data centers. However, operational expenses (OPEX) can be especially high for AI. When deployed at scale, AI models run millions of inferences which add up to trillions of processor operations. It’s not just the processing that’s costly. Having large AI models also means more storage costs. And with more data coming, those costs are only going to rise.

How to lower your OPEX with LEIP

The Latent AI Efficient Inference Platform (LEIP) is an MLOps SDK (Machine Learning Operations Software Development Kit) that can compress, optimize, and secure AI/ML model runtimes. LEIP can significantly lower your recurring OPEX by optimizing model runtimes for improved inference at smaller sizes. LEIP can provide additional cost savings when making the AI/ML model portable across different hardware.

Amazon SageMaker Benchmarking Study shows improved inference at lower costs

A recent benchmarking study compared models optimized with SageMaker alone versus those optimized with SageMaker and LEIP together.  By using more efficient processing, LEIP reduced the size of SageMaker models and transformed their computation workload. As a result, inference costs were dramatically lowered. 

When compared to Amazon SageMaker alone, LEIP increases your cloud performance while reducing your OPEX. 

LEIP optimized models:

  • Saved 40% on OPEX (improved inference plus less storage)
  • Increased edge inference by 66% and cloud inference by 48%
  • Reduced storage requirements with 10x smaller edge models   

LEIP on Edge: 66% Faster Inference      LEIP on the Cloud: 48% Faster Inference

LEIP on the edge: 66% Faster

With only a negligible loss in accuracy, a LEIP Optimized AI model offers 66% faster inference speed when compared to SageMaker models alone. The study used a Yolov5-large object detection model targeting an embedded NVIDIA Jetson NX processor. 

Cloud-based LEIP: 48% Faster Machine Learning

A LEIP Optimized AI model offers 48% faster inference speed on an AWS processor instance (G4Dn.xlarge) with only a negligible loss in accuracy. LEIP technology not only offers performance portability, but also model optimization to provide the most efficient acceleration of the AI model runtime. LEIP can also optimize AI models for other target hardware with only a simple command line change.

Faster time to market with LEIP end-to-end MLOps

LEIP offers an end-to-end workflow to train, optimize, deploy, and secure AI/ML runtimes for a large variety of models and hardware targets whether embedded or cloud based.


If rising cloud computing costs are a concern to you, LEIP can:

  • Improve model inference without requiring upgraded hardware  
  • Reduce operating cost without impacting your current cloud workflow
  • Deploy AI/ML models to market faster with a simpler, transparent, and dedicated edge MLOps workflow
  • Do it all via the same end-to-end workflow.

Ready to integrate LEIP with your SageMaker workflow? Visit latentai.com or contact us at info@latentai.com today!

View All