Research: GPU is the "Swiss Army Knife" in the Field of Data Processing, Not Just for AI-Specific Applications
Research shows that discussions about artificial intelligence (AI) have surged 383% since 2022. However, according to a research report recently released by data orchestration service provider Hammerspace, "The State of the Next Data Cycle: How to Use GPUs?", by 2024, only a small part of the $65 billion global GPU market will be used for AI-specific applications. This means that GPUs still have a broad application space beyond AI.
“The next wave of innovation will be driven by how effectively organizations activate and leverage their unstructured data,” said David Flynn, founder and CEO of Hammerspace. “Our research shows that the GPUs that many organizations initially purchased for AI projects are becoming the ‘Swiss Army Knife’ of data processing, unlocking value across industries in ways we never expected.”
The report highlights the changing dynamics of GPU utilization in the enterprise, revealing that many companies are repurposing these resources for a variety of non-AI applications to achieve measurable results.
Enterprises Increase Investment in GPU and AI
The report, based on approximately 17,000 digital conversations with approximately 200 industry leaders on platforms such as LinkedIn, Twitter, Reddit, GitHub and Discord, reveals how companies are responding to GPU and AI investments.
The report revealed that most companies are primarily focused on thought leadership (60%) and improving productivity (59%), while innovation discussions specifically aimed at achieving better AI outcomes account for only 18%.
Of the discussions on ethics, 51% focused on policy and best practices, reflecting the growing concern about responsible AI development.
Although many companies have invested heavily in AI infrastructure, including powerful GPU chips, these resources have not been fully utilized to handle AI workloads. Instead, GPUs are increasingly being used for traditional tasks such as strengthening big data processing and analysis projects.
As Hammerspace’s report outlines, GPUs are used in a wide range of industries, including large technology companies, scientific research, and media and entertainment, fully demonstrating their high flexibility and adaptability.
GPU Multi-field Application Cases
The report features case studies from companies such as Meta Platform, Los Alamos National Laboratory (LANL) and a well-known streaming provider, illustrating different uses for GPUs.
For example, Meta deployed more than 24,000 NVIDIA H100 GPUs to support the training of its Llama 2 and Llama 3 models, improving efficiency and resilience by optimizing GPU performance.
LANL has streamlined its hybrid supercomputer environment, which integrates CPU and GPU processing to support high-performance computing (HPC) and AI research. The infrastructure is being applied to projects covering national security, epidemic preparedness, and climate change mitigation. By consolidating isolated file systems into a unified platform, LANL has improved resource utilization for a variety of workloads and the efficiency of advanced data architectures.
Another example is a well-known streaming company that used GPU-CPU integration to improve its recommendation algorithms and optimize video streaming quality for millions of users. The report said that by combining GPUs with CPUs, the company increased the speed and accuracy of its personalized content recommendations, significantly improving streaming performance.
Key Assets Beyond AI
As the report notes, more companies are using GPUs not only for AI but also for a range of existing big data projects. This flexible and versatile approach has proven to be a beneficial move, bringing some unexpected benefits beyond initial expectations.
Although Goldman Sachs analysts noted that GPU supply constraints will continue to affect the deployment of AI projects until at least mid-2025, the report projects that the global GPU market will grow to $274 billion by 2029.
Meanwhile, another survey conducted by Tangoe found that 72% of respondents believe that AI-themed cloud spending is becoming difficult to manage (difficult to calculate ROI).
According to the agency's survey of 500 IT and financial professionals, spending in this category has grown 30% in the past year alone. For some companies that are still waiting to see the return on investment in computing power, this growth is obviously unsustainable.