AI + a16z: Tigris Data CEO on Building Your Own Datacenters
Podcast Host: a16z (Martin Casado)
Guest: Ovais Tariq, Co-founder & CEO of Tigris Data
Date: November 7, 2025
Episode Overview
This episode features Martin Casado (General Partner at a16z) in conversation with Ovais Tariq, CEO and co-founder of Tigris Data—one of the few startups building independent, core infrastructure for the AI era. They discuss the unique challenges of storage startups, why Tigris operates its own datacenters, the changing needs of AI workloads, technical and operational hurdles, developer experience, and how AI is reshaping software development and infrastructure companies.
Key Discussion Points & Insights
1. Why Independent Storage is Uncommon and Hard (03:25–06:51)
- Unlike other foundational cloud services (databases, data warehouses), independent storage is rare because most companies build on top of cloud providers like AWS S3, while Tigris competes at the foundational storage layer.
- “What we are effectively doing is building a product that competes with the foundational services that cloud provides. And that is something that is really hard to make cost efficient on top of the cloud itself.” – Ovais Tariq, 04:24
- Building storage involves not just software but also hardware and operational expertise.
2. Technical Architecture: Global, Active-Active Object Store (05:22–07:30)
- Tigris offers global storage with no concept of regions—data dynamically moves to wherever compute happens; ideal for AI workloads running in many locations.
- “With Tigris, there’s no concept of a single region. The storage is distributed and it gets dynamically placed wherever compute is running… The goal is to provide local access to storage everywhere, regardless of any region.” – Ovais Tariq, 05:26
- Managing global, S3-compatible storage requires deep knowledge of both distributed systems (for metadata) and hardware (for data durability, reliability).
3. Why AI Needs Specialized Storage (08:34–09:49, 10:12–12:25 )
- AI and ML workloads are inherently distributed—requiring storage close to compute to minimize latency and support new cloud providers focused on training and inference.
- Traditional object storage (S3) is poor for AI use cases:
- Struggles with massive volumes of small files (ML datasets)
- Can’t provide low-latency access needed for real-time AI (e.g., audio gen AI)
- Tigris focuses on the remote storage tier (the “data lake”) rather than the local storage on GPU clusters.
4. Developer Experience & Immutability (14:25–16:29)
- Strong focus on making high-performant infrastructure simple for developers:
- “Oh, I'm super focused on developer experience. I use the product myself as well.” – Ovais Tariq, 14:45
- Tigris is append-only and immutable, enabling features like instant, zero-copy snapshots and the ability to instantly fork even petabyte-size buckets.
- “You can instantly create a fork. Zero copy fork.” – Ovais Tariq, 16:22
5. Capacity Planning & Running Your Own Datacenters (16:30–22:34)
- Immutability means storage requirements continually grow. Capacity planning remains crucial—not just for space but also IOPS (input/output operations per second).
- Uber ran 100 petabytes of operational storage on SSDs entirely; Tigris follows similar practices with modern, high-density drives.
- “Per drive capacity has been increasing drastically—for example, 25 terabyte per drive now!” – Ovais Tariq, 18:41
- Setting up datacenters is simpler than a decade ago due to increased reliability and efficiency of modern hardware.
- Tigris pre-populates racks, shipping them to datacenters to minimize on-site work.
6. Team Structure, Skills, and Data Reliability (22:34–26:24)
- Tigris divides engineering into infrastructure (hardware, automation) and distributed systems (software) teams.
- For metadata consistency and reliability, they use FoundationDB—the same as Apple’s iCloud and Snowflake.
- “FoundationDB is actually one of the only databases I know of that has implemented simulation testing. They are really maniac about how they do testing.” – Ovais Tariq, 24:26
- The company takes a “batteries-included” philosophy—users get built-in caching and robust routing without needing to configure these themselves.
7. AI Coding Tools and Engineering Productivity (28:30–31:44)
- Senior developers at Tigris use AI code tools (notably Claude and Cuzr) for significant productivity gains.
- “For a long period of time, the only advantage new college grads brought is speed—[but] none of that matters anymore. What matters right now is systems thinking—whether you can design a system, you architect a system, you know how systems work. Because if you know that, then you can use AI assisting.” – Ovais Tariq, 29:02
- Team mandates using AI tools for code reviews, debugging, and log analysis.
- 80% of Ovais’s own code is AI-written.
8. Cost Structure & Eliminating Cloud Egress Fees (32:09–34:33)
- Tigris abolishes egress fees (the cost penalty for transferring data out of a cloud), a major pain point of incumbent cloud providers.
- “We have done away with the cloud tax, which is the egress bill. So there’s no egress.” – Ovais Tariq, 32:32
- “80% of [a customer’s] storage bill was egress.” – Ovais Tariq, 34:13
- Lower infrastructure and operational complexity allow Tigris to pass on cost savings and to optimize for AI-specific workload patterns.
9. Future of Infrastructure & Cloud (36:11–37:48)
- Tariq sees ongoing trend toward specialized infrastructure providers (compute, storage, higher-level services), catalyzed by AI workloads.
- Prefers inference workloads (over training) for storage efficiency—typical paradigms like Pareto (80% of requests for 20% of data) mean caching and IOPS are manageable.
Notable Quotes & Memorable Moments
-
On the challenge of independent storage:
“What we are effectively doing is building a product that competes with the foundational services that cloud provides. And that is something that is really hard to make cost efficient on top of the cloud itself.”
— Ovais Tariq, 04:24 -
On product philosophy:
“We are not trying to build something very general. We’re trying to... focus just on storage, GPU providers focusing just on the GPUs. And that’s what I feel like is going to be, that’s how the world is going to be.”
— Ovais Tariq, 11:23 -
On capacity planning:
“So the immutability of the system allows... the storage keeps on growing... capacity planning in general is important... not just the amount of space you have, but also the IOPS.”
— Ovais Tariq, 16:54 -
On developer experience:
“The infrastructure needs to be as simple to use as possible.”
— Ovais Tariq, 14:51 -
On AI for engineering:
“What matters right now is systems thinking... if you know that, you can use AI assisting, write things like code or culture and build a system. So that’s the thing that’s most important. That’s why the senior folks are succeeding.”
— Ovais Tariq, 29:02 -
On egress fees:
“We have done away with the cloud tax, which is the egress bill. So there’s no egress.”
— Ovais Tariq, 32:32
“80% of their storage bill was egress.”
— Ovais Tariq, 34:13
Timestamps for Key Segments
- [03:25] – Why storage startups are rare; difference from other cloud services
- [05:22] – Tigris’s global storage model; non-regional, API-compatible
- [08:34] – AI workload demands and Tigris’s fit
- [14:25] – Developer experience; interface design, append-only system
- [16:30] – Immutability, snapshots, and capacity planning
- [19:39] – Building physical datacenters: operational lessons
- [22:34] – Team structure, FoundationDB, reliability guarantees
- [28:30] – Using AI-assisted code tools throughout the stack
- [32:09] – Cost and pricing: No egress fees, user economics
- [36:11] – The future: Rise of specialty infrastructure providers
Conclusion
- The episode provides an insider’s look at the technical and operational challenges of building independent, global storage infrastructure for AI, underscored by an emphasis on developer experience, operational excellence, cost savings, and the transformative role of AI in productivity.
- Ovais Tariq’s hands-on perspective—spanning deep technical details, operational realities, and shifting industry economics—offers a window into the next wave of infrastructure innovation driven by AI workloads.
- For developers and companies building with or for AI, Tigris exemplifies the new breed of specialty providers: cost-efficient, globally distributed, simple to use, and optimized for the future’s most demanding workloads.
