WekaIO Inc., a startup with a data management platform that can store billions of files, has closed a $140 million late-stage investment.

The software maker said in its announcement of the Series E. round today that it included not only equity funding but also a secondary stock sale. Valor Equity Partners, an existing backer, led the investment at a $1.6 billion valuation. It was joined by nearly a dozen other investors, including Nvidia Corp., Qualcomm Ventures and Hitachi Ventures.

Organizations spread their information across multiple types of storage infrastructure. The data a company uses to train its artificial intelligence models is likely to be kept in high-speed, relatively pricey flash drives. Lower-priority records are typically relegated to less expensive hardware, while backups are stored in a separate storage environment that likewise has its own unique set of technical parameters.

Each of a company’s storage environments is often powered by a different data management platform. That means administrators have to navigate multiple management interfaces, which complicates their work.

Weka says its data management platform allows companies to implement a unified software layer across their various storage environments. The platform can run on both cloud and on-premises infrastructure, as well as orchestrate an organization’s backups. According to Weka, all those environments can be managed through a centralized interface that allows one administrator to maintain up to several exabytes of data.

One of the company’s main target use cases is powering artificial intelligence workloads. AI training datasets typically comprise a large number of relatively small files such as webpages that are often stored in different formats. According to Weka, its platform is capable of storing up to billions of files containing both structured and unstructured information.

“Unstructured data management is a common data management challenge in the AI era, so enterprise data stacks must be able to handle a variety of IO patterns, data types and sizes at ever-increasing volumes and velocity,” Weka co-founder and Chief Executive Officer Liran Zvibel (pictured, right, with co-founder Maor Ben Dayan) wrote in a blog post today. 

Weka’s platform includes a number of AI-focused performance optimizations. Typically, transferring data between the Nvidia graphics cards that run an AI model requires creating a copy of the data for the host servers’ central processing units. Weka supports GPUDirect RDMA, an Nvidia technology that removes the need to create such copies and thereby speeds up information transfers.

AI models and other applications sometime experience performance drops because other workloads running in the same environment use too much hardware. Weka promises to address that challenge with a feature called Converged Mode. According to the company, it allows administrators to ensure an important workload will always have access to a predetermined amount of CPU capacity, storage space and other hardware resources.

“With the arrival of gen AI, enterprise data requirements are becoming increasingly complex, while expectations for speed are rising,” Zvibel detailed. “GPUs and modern networking have improved their performance by four to five orders of magnitude over the last decade, and they require much faster access to data to ensure they are balanced.” 

Weka says its installed base comprises more than 300 organizations. Those customers include enterprises such as Samsung Electronics Co. Ltd., research institutions and government agencies. Weka claims one customer, the University of Surrey, managed to speed some AI workloads by a factor of eight using its platform.

TechCrunch reported that the company is generating more than $100 million in annualized recurring revenue from its install base. Weka expects to become cash flow positive by December. The software maker will use the proceeds from its new funding round to build more features and  provide liquidity for early employees. 

Photo: SiliconANGLE