Hammerspace Secures $100M Funding for AI-Driven Data Management Platform

Hammerspace, an unstructured data wrangler used by Nvidia, Meta and Tesla, raises $100M at $500M+ valuation | TechCrunch

Artificial intelligence services at their heart are massive data plays: you need data -- a lot of it -- to build the models, and then the models need efficient ways to ingest and output data to work. A company called Hammerspace has built a system to help AI and other organisations tap into data troves with minimal heavy lifting, and it's been seeing impressive adoption. Now, with customers including NVIDIA, Meta, Tesla, Palantir and the Department of Defense as well as other very recognisable names, Hammerspace is announcing $100 million in funding to expand its business. The funding is being described as a "strategic venture round," and it values Hammerspace at over $500 million, sources close to the company told TechCrunch. Its backers include Altimeter Capital and ARK Invest, alongside strategic investors that are not being disclosed. The investors are being described as "highly participatory." The funding is notable because it points to the ecosystem developing around the value that the market sees in AI companies, which are raising billions of dollars both to build their capital-intensive businesses and meet massive demand. But as Jamin Ball, a partner at Altimeter, noted, "You don't have an AI strategy without a data strategy." So a company that is building a platform to enable that data strategy can itself become very valuable, too. Hammerspace said much of its growth so far has been through word-of-mouth. It will be using a portion of this funding to expand on that more proactively with sales and marketing. Hammerspace previously raised $56 million from Prosperity7 Ventures (the venture arm of Saudi Aramco), ARK Invest, Pier 88 Hedge Fund, and other unnamed investors. Prior to that, it was self-funded by its CEO and co-founder David Flynn, the pioneer technologist known for his early work on Linux, supercomputers and flash computing. There are a vast number of companies that have set out to plug the big gap that exists in the data market today. "Vast" is an operative word here, as it is one of the companies that competes with Hammerspace, along with Dell, Pure Storage, Weka and many others in the worlds of data orchestration, file management, data pipeline, and data management. That gap goes something like this: The apps and other digital services we use to work and do everything else in life these days produce a lot of potentially valuable data. But data troves exist in silos -- they're fragmented, stored across multiple (competing) clouds and other environments, and are often unstructured. That makes them a challenge to use. This gap applies across a wide range of enterprise use cases, but perhaps the biggest of these at the moment is AI. "AI has been the perfect storm for needing what I have built," Flynn said in an interview. Hammerspace, as we've noted before, is named after the concept first coined from cartoons and comics, where characters pull objects they need out of thin air. This is, in effect, what Hammerspace does. The startup provides a way of making large amounts of data, regardless of where it lives or how it is used, accessible and available to an organization just when they need it, and keeping it out of the way when they do not. As Flynn describes it, typically the way that enterprises would have worked with data would be to port it from wherever it is to where it needs to be processed. "You need to install stuff on every system," he said. "It's a mess." It's also slow. "The AI arms race is such a sprint," he said. With "time to value" now a key priority for these companies, Hammerspace is signing up a lot of customers that are anxious about idle time. Flynn's background in flash computing is central to Hammerspace's breakthrough. Built on Linux, ubiquitous in the database world, he could see that the key to organising data across disparate locations was to create a file system to do so. The heart of this is the Linux kernel NFS client, ubiquitous across many of the data systems. Hammerspace's co-founder and CTO Trond Myklebust was the lead developer of the Linux kernel NFS client, and the startup remains its lead maintainer. The "file system" that the company has built for managing, moving and orchestrating data is based on a particular implementation in Linux that taps this. What it does, Flynn said, "is unique across the industry." Longer term, Flynn said last year that Hammerspace may go public as early as this year. That timeline has changed now but the direction has not. "Yes, IPO is absolutely the Hammerspace intended strategy," Flynn said. "We likely are still approximately two years out (dependent on market conditions)."

Hammerspace nabs $100M for its Linux-powered data management platform - SiliconANGLE

Hammerspace nabs $100M for its Linux-powered data management platform Hammerspace Inc., a startup that uses the Linux kernel to provide applications with faster access to data, has closed an $80 million funding round. The company detailed today that the Series B deal was led by Altimeter with participation from ARK Invest and several unnamed backers. Ark Invest also joined Hammerspace's previous $56 million raise in 2023. The software maker disclosed on occasion of the new funding that its customer base includes Meta Platforms Inc., the U.S. Defense Department and the National Science Foundation. The performance of an artificial intelligence cluster is determined not only by how much data it can process but also the rate at which it retrieves data from storage. If the cluster can theoretically process ten gigabytes of data per hour but only retrieves 8 gigabytes in that time frame, its compute capacity is left underutilized. As a result, storage performance is a major priority in machine learning projects. Hammerspace provides a data management platform that can be used to power AI clusters. Such clusters' storage environment usually includes multiple arrays. Hammerspace's platform can retrieve data from several arrays simultaneously rather than only one system at a time, which boosts performance. There are competing platforms that provide similar parallelization features. According to Hammerspace, the catch is that those platforms require customers to install proprietary data access programs, or clients. Deploying additional programs in an AI cluster complicates day-to-day maintenance and sometimes requires companies to rewrite the cluster's existing software. Hammerspace's parallel data management features are powered by a technology called NFS. The technology is built directly into Linux, which in turn powers most enterprise technology environments. As a result, companies don't have to install any additional software to use NFS and Hammerspace's parallelization features. The first iteration of NFS was released in 1984. To make the technology more suitable for use in modern data centers, Hammerspace has contributed a number of enhancements over recent years. The company's platform uses those additions to further boost storage performance. Usually, data access requests powered by NFS aren't sent directly to a storage system but first go through a so-called NAS server. Hammerspace skips that server, which speeds up information retrieval. To further save time, data requests can also bypass the central processing units of the systems involved in the workflow. Hammerspace says its platform can be used in not only AI clusters but also other types of technology environments. A feature of NFS called FlexFiles, which was likewise developed by the company, allows its platform to interact with many different kinds of storage infrastructure. Performance is not the software's only selling point. Hammerspace provides a so-called global namespace, which allows companies to manage multiple disparate storage environments as one big environment. That can simplify certain day-to-day maintenance tasks. The platform also automates the task of ensuring files are stored in accordance with a company's cybersecurity and reliability rules. There are scenarios where an application deployed in one data center may need to access files stored in another. That usually requires copying the files to the data center where the application is running. Hammerspace says that its platform allows workloads to access records stored in remote systems without creating copies, which reduces storage costs. "We orchestrate data to the GPU faster regardless of where it is physically stored," said Hammerspace co-founder and Chief Executive Officer David Flynn. "We instantly assimilate data from third-party storage so it's ready to process faster. We deploy and scale easily and quickly so our customers can achieve their business outcomes faster."