Amazon Web Services (AWS) has announced a landmark $110 million investment dedicated to advancing generative artificial intelligence (AI) research within academic institutions. Known as “Build on Trainium,” this program provides university researchers with computing resources, allowing them to explore new AI architectures, develop machine learning libraries, and optimize performance across large-scale AWS Trainium UltraClusters. This state-of-the-art infrastructure is built upon Trainium, Amazon’s custom chip designed to elevate deep learning and AI inference, opening new doors for researchers by supplying access to advanced hardware that has historically been financially inaccessible.

AWS Trainium: Powering a New Era in University AI Research

The Trainium chip, AWS’s proprietary solution for machine learning, offers universities a new level of computing power, essential for the complexities of AI training and inference. By offering free compute hours and substantial research credits, AWS is enabling university-led teams to tackle computational challenges in AI that were previously constrained by budget limitations. The Trainium UltraCluster, a colossal infrastructure supporting up to 40,000 Trainium chips, is now available to researchers who can benefit from optimized chips designed specifically for the demanding workloads of AI.

AWS executives stated that the goal of Build on Trainium is to empower academic researchers, granting them the resources necessary to conduct transformative AI work. The initiative spans various disciplines, from improving AI accelerator performance to conducting in-depth studies on distributed systems. Through Amazon’s latest endeavor, leading research institutions such as Carnegie Mellon University (CMU) are using the program to advance AI compiler optimizations, ML parallelization, and fine-tuning techniques for language models.

Broadening the AI Research Landscape and Building the Future of AI Expertise

This initiative doesn’t stop at providing compute power; it’s also dedicated to fostering the next generation of AI experts. AWS has established a funding pool for new research and educational opportunities, collaborating with institutions to establish research grants and increase student engagement in AI. Through a series of Amazon Research Awards, selected proposals will receive substantial AWS Trainium credits, alongside access to AWS’s UltraCluster infrastructure.

The program is already impacting institutions such as CMU, where Professor Todd C. Mowry stated that this access is allowing both faculty and students to expand AI research. Mowry emphasized that the resources provided through Build on Trainium are significantly enhancing the university’s work in compiler optimization and machine learning parallelization, offering a rare opportunity to leverage high-performance AI infrastructure in academic settings.

New Horizons for AI Research with AWS Neuron Kernel Interface (NKI)

A notable aspect of the Build on Trainium program is the inclusion of the AWS Neuron Kernel Interface (NKI), a revolutionary programming tool that provides deep access to Trainium’s hardware. This advanced interface enables researchers to manipulate the chip’s instruction set, allowing them to create customized computational kernels optimized for new AI model operations and scientific advancements. Christopher Fletcher, an associate professor at the University of California, Berkeley, highlighted the immense flexibility of Trainium’s architecture, explaining that it provides the “knobs” needed for fine-tuning at every step, making it an ideal platform for innovative AI research.

The NKI and the customizable nature of Trainium have positioned AWS as a pivotal player in academia’s AI landscape. Fletcher shared that the access to Trainium clusters provided by AWS has become essential for his lab’s research, with computing needs varying across each project. He emphasized that this AWS initiative is enabling “unexpected innovation” within AI research, with Trainium providing the versatility required for both foundational and cutting-edge investigations.

A Catalyst for Collaborative AI Innovation

Beyond individual research, Build on Trainium fosters an environment of collaboration within the AI community. The program links grant recipients with AWS’s Neuron Data Science community, a network dedicated to knowledge-sharing and innovation. By connecting academia, industry specialists, and AWS’s Generative AI Innovation Center, the initiative facilitates a collaborative ecosystem in which AI research can thrive.

In alignment with the open-source philosophy, AWS is encouraging researchers to share the outcomes of their work. Academics involved in Build on Trainium are expected to contribute their findings and code to open-source machine learning libraries, thereby supporting the advancement of AI capabilities globally. This collaborative approach aims to amplify the program’s impact, as researchers worldwide will be able to build upon the achievements realized through AWS’s investment.

AWS’s Strategic Support for Academia and the Next Generation of AI Development

AWS’s commitment to supporting academia, particularly in generative AI, reflects its strategy to expand the role of custom hardware in AI development. Since the introduction of AWS Inferentia chips in 2019, Amazon has steadily advanced its cloud AI capabilities. Now, by providing universities access to the high-powered Trainium infrastructure, AWS aims to contribute significantly to AI’s future while nurturing a new wave of experts capable of driving the next generation of advancements in machine learning and generative AI.

  AI research, Amazon Web Services, AWS, Carnegie Mellon University, Neuron Kernel Interface, Trainium, University of California Berkeley