What does DeepInfra actually sell?

DeepInfra sells inference by the token: OpenAI-compatible APIs over more than 150 open-source models with pay-as-you-go pricing and zero data retention. Its DeepCluster product also sells dedicated B300 GPU clusters of 256 to 5,000 GPUs on 3–5 year terms.

Who should choose DeepInfra?

DeepInfra fits application builders who want the lowest cost-per-token on open-source models with no infrastructure to manage — especially agentic workloads, which already account for roughly 30% of its weekly token volume.

How big is DeepInfra in 2026?

As of mid-2026 DeepInfra processes nearly five trillion tokens per week across eight US data centers it owns and operates, and closed a $107 million Series B in May 2026 with NVIDIA among the investors.

DeepInfra: Sells tokens — The Compute Ladder

The two-sided view

What DeepInfra sells — and what it buys

Atomic unit sold

Tokens. OpenAI-compatible APIs over 150–190+ open-source models, pay-as-you-go, zero data retention, SOC 2 / ISO 27001. GPUs fully abstracted away — plus DeepCluster: dedicated B300 clusters of 256–5,000 GPUs on 3–5 year terms, operated by DeepInfra but owned by the customer.

What they buy

Data centers and hardware outright — DeepInfra owns and operates GPU clusters across eight US data centers, with early Blackwell and Vera Rubin + NVIDIA Dynamo deployment. Vertical integration is the moat.

Verified 4 July 2026

Scale proof

~5T

tokens processed per week; token volume up 25× since Series A; revenue tripled since the start of 2026

$107M

Series B (May 2026) co-led by 500 Global and Georges Harik, with NVIDIA, Samsung Next and Supermicro participating

~30%

of weekly token volume already comes from agents — one of the clearest public data points that agentic workloads are now infrastructure-scale

200M+

users served by imo, the messenger the founding engineering team built and scaled before DeepInfra

The strategy

The bet

That inference — specifically always-on agentic inference at 50–100+ model calls per task — becomes the dominant compute workload, and that owning the full stack from data center to API beats renting spot capacity on cost-per-token.

Deal-making

Natural counterparty

Application builders who never want to see a GPU; and on the buy side, data-center and power providers as it expands beyond eight US sites — EU expansion is the obvious next move given AI-Act data-residency demand.

Buyer fit

Choose DeepInfra when

You want open-source model inference at the lowest cost-per-token with zero infrastructure ownership.

Look elsewhere when

You need proprietary frontier models or custom training infrastructure.

Private ledger

Five fields we track but don’t publish

Everything above is public and verifiable. The fields below change weekly and are verified directly with each provider — they live in a private supply-demand ledger, not on this page.

Available capacity, next 90 daysGPU type · quantity · location · online date

Effective price floorreserved · per-GPU-hour, at volume

Idle-inventory posturecurrent utilization pressure

Preferred deal structureattach terms · co-sell terms

Capacity-partnerships ownerthe person who signs

Ask about DeepInfra’s current posture → Answered case-by-case, with the provider’s consent.