In a time when Nvidia (NASDAQ:NVDA) dominates graphics processing unit-accelerated data, a challenger has just turned the tables.
DataPelago announced last week that its Nucleus universal engine outperformed Nvidia's GPU-accelerated DataFrame library cuDF by wide margins, raising the roofline of GPU-accelerated data processing and transforming the economics of artificial intelligence infrastructure.
Don't Miss:
- The same firms that backed Uber, Venmo and eBay are investing in this pre-IPO company disrupting a $1.8T market — and you can too at just $2.90/share.
- ‘Scrolling To UBI' — Deloitte's #1 fastest-growing software company allows users to earn money on their phones. You can invest today for just $0.30/share.
A Software-First Disruption to Nvidia's GPU Dominance
DataPelago showed Nucleus outperforming Nvidia's cuDF in compute-intensive workloads on Nvidia GPUs. The universal engine operates across heterogeneous hardware, from central processing units to GPUs, without requiring any code or infrastructure changes.
DataPelago released benchmark results showing its Nucleus delivering up to 10.5 times faster project operations, 10.1 times faster filters, and 4.3 times faster aggregations compared to cuDF. In string-heavy hash join tests, Nucleus recorded throughput 38.6 times higher than cuDF on small strings and nearly four times higher on extra-large strings.
Inside the Benchmarks: How Nucleus Outpaces Nvidia cuDF
Nvidia introduced cuDF in 2018 as part of its RAPIDS suite, designed to unlock GPU acceleration for data science and AI workloads. While cuDF sped up pipelines over CPU libraries, Fast Company says it remained tied to Nvidia GPUs, compute unified device architecture compatibility, and memory limits that created vendor lock-in for enterprises.
Trending: An EA Co-Founder Shapes This VC Backed Marketplace—Now You Can Invest in Gaming's Next Big Platform
DataPelago CEO Rajan Goyal told Fast Company that the industry needs software that "fully leverages the hardware's strengths while compensating for its limitations" through fresh algorithms built for modern workloads.
Goyal added in the company statement that GPUs offer unmatched parallel power but remain underutilized due to legacy software. He said Nucleus "raises the roofline," leveraging innovative execution layers like kernel fusion, multi-column support, and zero-copy string optimization.
DataPelago says Nucleus is hardware-neutral, able to run across CPUs, GPUs, and other accelerators while integrating seamlessly with frameworks like Spark without code rewrites.
DataPelago President JG Chirapurath told Fast Company that this lets firms realize "far greater performance from existing hardware investments," a key appeal for enterprises already facing rising GPU cloud costs.
See Also: Dump Your Financial Advisors. One AI Platform Manages Everything — With Zero AUM Fees. Book Your Demo Today.
Funding, Industry Impact, and the Road Ahead
According to Fast Company, DataPelago has raised $47 million in seed and Series A funding from Eclipse, Qualcomm Ventures, and Taiwania Capital. Goyal, who previously held leadership roles at Cisco (NASDAQ:CSCO) and Oracle (NYSE:ORCL), recently hired Chirapurath to help accelerate adoption.
Analysts such as Forrester's Alvin Nguyen caution that while the benchmark results are striking, adoption could be slowed by developer reliance on Nvidia's ecosystem and the upfront effort required to transition to new tools. However, Fast Company says that reducing vendor lock-in and improving economics for AI pipelines will appeal to enterprises under pressure to optimize costs.
The broader implication is that AI performance gains may be defined by the intelligence of the software that orchestrates it, no longer solely by hardware. For years, the story centered on shortages of GPUs, but now, Fast Company says DataPelago's Nucleus suggests the next breakthrough may come from how those GPUs are harnessed rather than how many can be manufactured.
Read Next: From Chipotle to Red Bull, Top Brands Are Already Building With Modern Mill's Tree-Free Wood Alternative — Here's How You Can Invest Too
Image: Shutterstock
