Revolutionizing Data Infrastructure for the AI Era
The Surge in AI and its Demands on Computing Power
The explosive growth of AI enterprises has dramatically escalated the need for computational resources. Companies such as CoreWeave, Together AI, and Lambda Labs have successfully capitalized on this demand, drawing significant investment and recognition for their capabilities in delivering distributed computing power. However, the majority of businesses continue to rely on the major cloud service providers—AWS, Google Cloud, and Microsoft Azure—for their data storage needs. These incumbent systems were primarily designed to keep data geographically close to their own processing units, rather than accommodating dispersed, multi-cloud, or regional computing strategies.
Tigris Data's Vision for Decentralized Storage
Ovais Tariq, co-founder and CEO of Tigris Data, articulated his company's mission: to provide a distributed storage solution that mirrors the flexibility and efficiency of modern AI-driven distributed computing. \"Contemporary AI applications and their supporting infrastructure are increasingly opting for distributed computing instead of large-scale cloud services,\" Tariq explained. \"We aim to extend this same principle to data storage, recognizing that processing power is ineffective without accessible data.\" Tigris, spearheaded by the team responsible for Uber's robust storage system, is developing a network of localized data centers designed to fulfill the complex storage requirements of advanced AI operations. Their AI-native platform intelligently moves data to be near GPU resources, seamlessly replicates information across locations, handles billions of small files, and ensures rapid access for tasks such as model training, inference, and agentic processes.
Securing Funding to Challenge Cloud Monopolies
To realize its ambitious goals, Tigris recently concluded a Series A funding round, raising $25 million. This round was spearheaded by Spark Capital, with existing investors, including Andreessen Horowitz, also participating. This significant investment underscores the industry's belief in Tigris' potential to disrupt the established order of what Tariq refers to as \"Big Cloud.\" Tariq contends that the dominant cloud providers not only impose higher costs for data storage but also offer less efficient services. Historically, these providers levied egress fees—often termed a \"cloud tax\"—on customers seeking to transfer data to a different cloud, download it, or relocate it to optimize for cheaper GPUs or concurrent model training in various global regions. This practice can be likened to a gym charging members extra for discontinuing their membership.
Addressing Costly Egress Fees and Latency Challenges
According to Batuhan Taskaya, Head of Engineering at Fal.ai and a client of Tigris, egress fees once constituted the largest portion of Fal's cloud expenditures. Beyond the financial burden of egress fees, Tariq highlighted another critical issue: latency within major cloud infrastructures. He stated, \"Egress fees were merely one symptom of a deeper systemic problem: centralized storage architectures that are incapable of keeping pace with the demands of a decentralized, high-speed AI ecosystem.\" The majority of Tigris' client base, comprising over 4,000 entities, consists of generative AI startups engaged in developing image, video, and voice models. These applications typically involve extensive, latency-sensitive datasets. Tariq illustrated the problem with a scenario: \"Consider interacting with an AI agent performing local audio processing. Low latency is paramount. You need both your compute and storage to be local and in close proximity.\"
Optimizing for AI Workloads and Data Sovereignty
Tariq further noted that existing large cloud systems are not optimally configured for AI workloads. The process of streaming vast datasets for model training or executing real-time inference across multiple geographical regions can lead to latency bottlenecks, thereby diminishing model performance. However, by enabling access to localized storage, data retrieval speeds are enhanced, empowering developers to execute AI tasks with greater reliability and cost-effectiveness through decentralized cloud environments. Fal's Taskaya affirmed, \"Tigris allows us to seamlessly scale our workloads across any cloud by offering consistent access to the same data filesystem from diverse locations, all without incurring egress charges.\" There are additional compelling reasons for organizations to maintain data in closer proximity to their distributed cloud options. For instance, in heavily regulated sectors such as finance and healthcare, a significant barrier to AI adoption is the imperative to guarantee data security and compliance. Tariq also pointed out a growing desire among companies for greater ownership and control over their data, citing Salesforce's recent decision to restrict its AI competitors from utilizing Slack data. \"Enterprises are becoming increasingly cognizant of the immense value of their data, and how it fuels large language models and other AI technologies,\" Tariq observed. \"They seek to exert more control over it, rather than ceding that control to external parties.\"
Future Expansion and Growth
With its recent capital injection, Tigris is poised to continue expanding its network of data storage centers to accommodate escalating demand. Tariq proudly noted that the startup has experienced an eight-fold annual growth since its inception in November 2021. Currently, Tigris operates three data centers situated in Virginia, Chicago, and San Jose, with plans for further expansion across the U.S., as well as into European and Asian markets, specifically targeting London, Frankfurt, and Singapore.
