Tigris

What Tigris actually is

Tigris is a globally distributed, S3-compatible object storage service. Point any S3 SDK at t3.storage.dev, swap your keys, and your code works — boto3, aws-sdk-go-v2, rclone, and anything else that speaks S3. The pitch is simple and structural: data is distributed globally and follows access patterns, pricing is identical in every geography, and egress is free on every tier, in every direction. For AI teams reading training sets repeatedly from scattered GPUs, or any shop tired of the egress line item, that last part is the product.

A note on the record here: older descriptions of Tigris — including our own database row — describe a serverless NoSQL database with search. That was an earlier product; the company is now single-mindedly an object storage vendor, as its own about page puts it: one thing, done right. If you arrive looking for a document database, what you will find instead is storage with a few primitives S3 never grew: bucket forks, snapshots, soft delete, and object events. The current feature set, not the old pitch, is what we review.

The company is Tigris Data, Inc., founded in 2022 by three engineers from Uber's storage platform team — Ovais Tariq (CEO), Himank Chaudhary (CTO), and Yevgeniy Firsov (Chief Architect) — who built and ran systems like Docstore and Herb at billions of requests per day. It raised a $25M Series A led by Spark Capital in October 2025, with Andreessen Horowitz, General Catalyst, and Basis Set on the cap table, and reports more than 4,000 customers including fal.ai, KREA, Hedra, and Fly.io, whose storage it backs. The company says it operates on its own hardware rather than reselling hyperscaler capacity — which is how the zero-egress pricing stays coherent. Compliance is further along than most vendors this size: SOC 2 Type II certified and HIPAA BAA available.

Pricing: what you really pay

Tigris publishes a full rate card — and even a machine-readable pricing page for agents — which we rate as top-of-class transparency. Standard storage is $0.02 per GB-month. Infrequent Access is $0.01 per GB-month with a 30-day minimum retention and $0.01 per GB retrieval. Archive is $0.004 per GB-month with a 90-day minimum and free retrieval after restore; Archive Instant Retrieval is also $0.004 per GB-month but adds $0.03 per GB retrieval. Requests are $0.005 per 1,000 Class A (PUT, COPY, POST, LIST) and $0.0005 per 1,000 Class B (GET, SELECT and friends) on every tier; DELETEs are free; object notifications are $0.01 per 1,000 events. Egress — regional, cross-region, or to the open internet — is $0.

The free tier resets monthly and needs no card: 5 GB of Standard storage, 10,000 Class A requests, and 100,000 Class B requests per month, on every account, indefinitely. Enterprise is custom-priced with a negotiated service contract. Tigris's own comparison tool puts a 100 GB stored / 30% egress workload at $1.90 on Tigris versus $8.75 on AWS and $11.00 on GCS — vendor-published arithmetic, but the per-GB rates behind it are public and checkable, and the egress asymmetry is real: AWS charges $0.09 per GB after 100 GB free and R2-style rivals still differ on storage.

Where the costs actually appear: minimum retention on cold tiers (an object deleted early from Infrequent Access is billed for 30 days anyway; Archive, 90 days), archive restore time billed as Standard-tier storage while the restore window is open, and object notifications from the first event. None of this is hidden — the FAQ spells out the retention arithmetic with worked examples — but lifecycle rules pointing young objects at cold tiers can cost more than expected. Storage is metered as GB-month via daily peak averaging, in binary GB.

Performance & infrastructure

The architecture is the differentiation. Tigris distributes objects globally by default and serves reads from the nearest replica, with bucket placement options for Global (default), single-region, multi-region, and dual-region (a USA multi-region bucket costs the same as a EUR one). The data layer follows access patterns rather than sitting in one region you then pay to move; writes land durably in the bucket's home configuration and reads fan out. Because the company runs its own hardware — a point it makes explicitly against the hyperscaler resale model — it can price traffic at zero without eating a provider egress bill underneath.

The compatibility surface is the other half. Tigris claims more than 90% of the S3 API, works with the standard SDKs and tools after an endpoint swap, and offers a zero-downtime migration path where objects copy over on first read and writes mirror back until you cut over. That is a materially lower switching cost than any proprietary storage API, and in our view it is the right posture for a challenger: keep the interface boring and compete on the bill. The extra primitives — forks, snapshots, soft delete, lifecycle rules, object events — cost nothing to enable; you pay for stored data at the tier rate.

We have not benchmarked Tigris ourselves. The honest framing for latency: globally distributed reads are the design goal, and where your data lives relative to your compute still governs the numbers you see. Teams with hot inference paths should measure their own access pattern rather than trusting any vendor's latency map — see our benchmarks approach for how we think about testing storage.

Where it falls short

Maturity cuts both ways. The founding team's pedigree is unusually strong and the customer list is real, but this is a company founded in 2022 with one product line and a Series A-sized balance sheet. Storage is the layer where you least want a vendor to disappear, and while the S3-compatible interface is a genuine exit ramp — you can repoint any client elsewhere — a multi-terabyte migration is still a project. For anything critical, budget for the exit even though the interface makes it cheap.

Second, no public availability SLA outside Enterprise. The free tier comes with no contractual uptime guarantee, and negotiated SLAs are part of Enterprise contracts only. A storage service with no published availability number is normal for this stage, but it is a gap on the procurement checklist: if you need a 99.9x% commitment with credits, you are signing an Enterprise agreement or accepting the risk.

Third, product depth is deliberately narrow. There is no CDN, no compute, no database (the old product is gone), no IAM integration with your identity provider beyond access keys and the S3 auth model, and object notifications cost from the first event rather than coming with a free allowance. Compare that with an AWS account where storage is one pane of a hundred. Tigris is a superb component, not a platform, and your architecture supplies everything around it.

Who should pick this

Our pick for AI and ML teams shipping training data and inference assets across regions and clouds, for anyone whose egress bill exceeds their storage bill, and for developers on compute platforms like Fly.io who want storage that matches the deployment model. The free tier is enough to run a real project — 5 GB plus generous request allowances — and the per-GB rates undercut AWS Standard outright before you count egress. Bucket forking and snapshots suit pipelines that branch datasets, and soft delete plus lifecycle rules cover the usual safety needs.

We would not pick Tigris as the sole repository for regulated archives without an Enterprise contract and a tested restore runbook, nor for teams that want storage inside a broader managed cloud account with consolidated IAM and support. If you are comparing object stores specifically, line it up in our cloud object storage category; the honest decision is usually Tigris versus Cloudflare R2 versus S3, and the tie-breakers are egress, geography, and how much you trust a young vendor with your data layer.

Key takeaways

• S3-compatible globally distributed object storage with zero egress fees on every tier and every direction — the structural pitch against AWS.
• Transparent rates: Standard $0.02/GB-month, Infrequent Access $0.01 (30-day min), Archive and Archive Instant Retrieval $0.004 (90-day min); Class A $0.005/1k, Class B $0.0005/1k.
• Standing free tier on every account, no card: 5 GB Standard storage, 10,000 Class A and 100,000 Class B requests monthly.
• Built by ex-Uber storage engineers (founded 2022, $25M Series A led by Spark Capital in 2025); runs on its own hardware; powers Fly.io storage and AI platforms like fal.ai.
• Extras S3 lacks — bucket forks, snapshots, soft delete, object events — with no separate enablement charge; watch cold-tier minimum retention billing.
• No public uptime SLA outside Enterprise; young vendor, so keep the S3-compatible exit path and your own backups in the runbook.

Sources

• https://tigrisdata.com/pricing
• https://tigrisdata.com/features
• https://www.tigrisdata.com/about/
• https://www.tigrisdata.com/docs/about/
• https://www.crunchbase.com/organization/tigris-data-inc
• https://app.daily.dev/posts/tigris-secures-25m-to-revolutionize-ai-storage-infrastructure-pfdgz7s5l