CentML Secures $27 Million Funding To Enhance Efficiency Of AI Model Deployment


CentML, a startup focused on developing tools to optimize the cost and performance of deploying machine learning models, has announced a successful extended seed funding round, raising $27 million. Notable participants in the round include Gradient Ventures, TR Ventures, Nvidia, and Microsoft Azure AI VP Misha Bilenko. This investment brings the total raised by CentML to an impressive $30.5 million.

Key Takeaway

CentML, a startup specializing in AI model optimization, has raised $27 million in an extended seed funding round. The company’s software aims to decrease the cost and improve the performance of machine learning model deployment. By identifying bottlenecks and optimizing workloads, CentML can significantly reduce expenses while maintaining speed and accuracy. The funding will be used to support product development, research, and the expansion of CentML’s team.

The additional capital will be used to support CentML’s ongoing product development and research initiatives. Furthermore, the funding will be instrumental in expanding the company’s engineering team and overall workforce, currently consisting of 30 individuals across the United States and Canada, according to Gennady Pekhimenko, co-founder and CEO of CentML.

CentML was founded in 2021 by Gennady Pekhimenko, Akbar Nurlybaev, and Ph.D. students Shang Wang and Anand Jayarajan. The team shared a common objective of addressing the increasing challenges of accessing compute resources amidst the worsening AI chip supply problem. Pekhimenko explained, “Machine learning costs, talent and chip shortages… any AI and machine learning company faces at least one of these challenges, and most face a few at a time.”

One of the main challenges faced by companies in training AI models is the scarcity of high-end chips, particularly those required for GPU-based hardware. The demand from both enterprises and startups has made it difficult to acquire the necessary chips, resulting in companies having to compromise on the size of the models they can deploy or experience higher inference latencies for their deployed models.

CentML aims to address these challenges by optimizing model training workloads to run efficiently on existing hardware. Their software platform identifies bottlenecks during model training, predicts deployment time and cost, and provides access to a compiler that optimizes model training workloads for target hardware. According to Pekhimenko, CentML’s software can reduce expenses by up to 80% without compromising speed or accuracy.

While CentML faces competition from other software-based optimization solutions like MosaicML and OctoML, Pekhimenko emphasizes that CentML’s techniques do not result in a loss of model accuracy. Additionally, CentML’s compiler is touted as a “newer generation” and more performant than that of OctoML.

In the future, CentML plans to focus on optimizing not only model training but also inference, which involves running models after they have been trained. By reducing the memory requirements for model deployment, CentML enables teams to deploy on smaller and more affordable GPUs, potentially opening up new growth opportunities for the company.

Leave a Reply

Your email address will not be published. Required fields are marked *