This Distributed Data Storage Startup Wants To Take On Big Cloud

The detonation of AI companies has pushed request for computing powerfulness to caller extremes, and companies for illustration CoreWeave, Together AI and Lambda Labs person capitalized connected that demand, attracting immense amounts of attraction and superior for their expertise to connection distributed compute capacity.

But astir companies still shop information pinch nan large 3 unreality providers, AWS, Google Cloud, and Microsoft Azure, whose retention systems were built to support information adjacent to their ain compute resources, not dispersed crossed aggregate clouds aliases regions.

“Modern AI workloads and AI infrastructure are choosing distributed computing alternatively of big cloud,” Ovais Tariq, co-founder and CEO of Tigris Data, told TechCrunch. “We want to provide the aforesaid action for storage, because without storage, compute is nothing.” 

Tigris, founded by nan squad that developed Uber’s retention platform, is building a web of localized information retention centers that it claims tin meet nan distributed compute needs of modern AI workloads. The startup’s AI-native retention platform “moves pinch your compute, [allows] data [to] automatically replicate to wherever GPUs are, supports billions of mini files, and provides low-latency entree for training, inference, and agentic workloads,” Tariq said. 

To do each of that, Tigris recently raised a $25 cardinal Series A information that was led by Spark Capital and saw information from existing investors, which see Andreessen Horowitz, TechCrunch has exclusively learned. The startup is going against nan incumbents, who Tariq calls “Big Cloud.”

Ovais Tariq, CEO of Tigris, astatine a Tigris information halfway successful Virginia.Image Credits:Tigris Data

Tariq feels these incumbents not only offer a much costly information retention service, but also a little businesslike one. AWS, Google Cloud and Microsoft Azure person historically charged egress fees (dubbed “cloud tax” successful nan industry) if a customer wants to migrate to different unreality provider, aliases download and move their information if they want to, say, usage a cheaper GPU aliases train models successful different parts of nan world simultaneously. Think of it for illustration having to salary your gym other if you want to extremity going there.

According to Batuhan Taskaya, caput of engineering astatine Fal.ai, one of Tigris’ customers, those costs erstwhile accounted for nan mostly of Fal’s cloud spending.

Beyond egress fees, Tariq says there’s still nan problem of latency pinch larger unreality providers. “Egress fees were conscionable 1 denotation of a deeper problem: centralized retention that can’t keep up pinch a decentralized, high-speed AI ecosystem,” he said. 

Most of Tigris’ 4,000+ customers are for illustration Fal.ai: generative AI startups building image, video and sound models, which thin to person large, latency-sensitive datasets.  

“Imagine talking to an AI supplier that’s doing section audio,” Tariq said. “You want nan lowest latency. You want your compute to beryllium local, adjacent by, and you want your retention to beryllium local, too.” 

Big clouds aren’t optimized for AI workloads, he added. Streaming monolithic datasets for training aliases moving real-time conclusion crossed aggregate regions tin create latency bottlenecks, slowing exemplary performance. But being capable to entree localized retention intends data is retrieved faster, which intends developers tin tally AI workloads reliably and much costs efficaciously utilizing decentralized clouds. 

“Tigris lets america standard our workloads successful immoderate unreality by providing entree to nan aforesaid information filesystem from each these places without charging egress,” Fal’s Taskaya said.

There are different reasons why companies want to person information person to their distributed unreality options. For example, successful highly regulated fields for illustration finance and healthcare, 1 ample roadblock to adopting AI devices is that enterprises request to guarantee information security.

Another motivation, says Tariq, is that companies progressively want to ain their data, pointing to how Salesforce earlier this twelvemonth blocked its AI rivals from utilizing Slack data. “Companies are becoming more and more aware of really important nan information is, how it’s fueling the LLMs, how it’s fueling nan AI,” Tariq said. “They want to beryllium much successful control. They don’t want personification other to beryllium successful power of it.” 

With nan caller funds, Tigris intends to proceed building its information retention centers to support increasing request — Tariq says nan startup has grown 8x each twelvemonth since its founding in November 2021. Tigris already has three data centers in Virginia, Chicago and San Jose, and wants to proceed expanding successful nan U.S. arsenic good arsenic successful Europe and Asia, specifically successful London, Frankfurt and Singapore.