What Horizon Compute does

Horizon Compute is a private GPU cloud infrastructure company headquartered in Pasadena, California, that owns and operates a fleet of 4,352 H100 and H200 GPUs interconnected via 3.2 Tbps InfiniBand or RoCEv2 fabric. The company sells two integrated service lines from this owned infrastructure: (1) GPU Capacity — reserved clusters and on-demand GPU instances priced per GPU-hour, marketed as direct-from-owner purchasing with stable pricing and guaranteed availability; and (2) Inference Endpoints — token-metered access to frontier open-weight AI models priced per million tokens, with no upfront commitment required.

The technical stack emphasizes high-throughput, low-latency AI workloads, with 3-minute automated provisioning to a running instance and a publicly committed upgrade path to next-generation B300 and Rubin silicon. Horizon Compute does not build proprietary AI models; instead it serves third-party frontier open-weight models from its owned hardware, positioning the inference endpoint product as a consumption-priced abstraction layer over raw GPU rental.

The go-to-market is API-first and direct, with all sales flowing through horizoncompute.com and a global reach via the website and API; no resellers, marketplaces, or channel partners are disclosed beyond a set of named technology collaborators (Canopy Wave, Hyperbolic, Dell Technologies, GMI Cloud, Hewlett Packard Enterprise, Shadeform). Customer segments are horizontal — AI developers and product teams, enterprise and partner buyers, and AI research organizations — with no named customer logos or revenue disclosures available in the source material. Headcount is reported at 1-10 employees, and no funding rounds, investors, or financial metrics are publicly disclosed.

Name

Horizon Compute

Website

https://horizoncompute.com

Company type

Private

Operating status

Operating

Headcount range

1–10 employees

Short description

Horizon Compute is a Pasadena-based GPU cloud provider that owns a 4,352-unit H100/H200 fleet and sells token-metered inference endpoints for frontier open-weight AI models alongside reserved and on-demand GPU capacity to AI developers, enterprises, and research organizations via a direct API.

Ownership category

akta.pro rank

Product category

Cloud GPU Infrastructure

NAICS

Computer Systems Design and Related Services (5415)

SIC

Electronic Computers (3571)

akta.pro primary industry

AI Compute Cloud & GPU-as-a-Service (HDAAAAAK)

akta.pro secondary industries

GPU/Accelerator Compute Servers (HDABADAC), GPU-Accelerated & AI Training/Inference Servers (HDACABAG), AI Compute Virtualization & Scheduling (GPU virtualization, cluster schedulers) (HDAAAAAG)

Keywords

• GPU cloud infrastructure
• AI inference services
• Token-metered inference
• GPU compute capacity
• AI training infrastructure

Headquarters

HQ city

Pasadena

HQ country

United States

HQ region

North America

Markets served

GTM type

B2B

Offering type

Software

Cost components

Infrastructure, Technology or R&D, Operations, Personnel, Supply Chain

Revenue model

1. Token-metered inference endpoints: Frontier open-weight models served from owned GPU clusters, metered per token consumed. Customers pay only for the tokens their product uses.
2. Reserved and on-demand GPU capacity: Partners and customers purchase reserved clusters and on-demand instances from Horizon's owned GPU fleet. Pricing that holds, availability that is real — buying directly from the owner of the silicon.

Pricing tiers

Go-to-market motion1 record

Distribution channels1 record

Marketing channels1 record

Core offering

Horizon Compute owns H100 and H200 GPU clusters outright and sells GPU compute capacity (reserved clusters and on-demand instances priced per GPU-hour) alongside token-metered inference endpoints for frontier open-weight AI models. Customers can consume AI inference via API at $ per million tokens, or rent raw GPU capacity from the fleet, with instances provisioning in approximately 3 minutes.

Product overview

Horizon Compute is an inference-first AI infrastructure company that owns its GPU clusters outright, offering two integrated service lines: GPU Capacity (reserved and on-demand GPU-hour based access to their H100/H200 fleet) and Inference Endpoints (token-metered access to frontier open-weight models). Both offerings leverage the same owned infrastructure for seamless scaling from capacity renting to fully managed inference.

Differentiator

Problem solved

Functional benefit

Products and services

• GPU Capacity
• Inference Endpoints

Quantifiable outcome

• 3-minute provisioning to a running instance

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability2 records

Feature5 records

Scale indicators1 record

Recent moves5 records

Expansion highlights5 records

Broad incumbents

• Microsoft Azure (ND H100 v5): Azure provides ND H100 v5 VM series and AI inference services as part of Azure AI. Broad incumbent competing for the same enterprise and AI-developer GPU workload demand.
• AWS EC2 (P5 / GPU instances): AWS offers P5 (H100) and other GPU-backed EC2 instances plus Bedrock inference services. Comparable on GPU capacity and inference but with much broader portfolio, hyperscaler scale, and deeper enterprise reach.
• Google Cloud (A3 / TPU): Google Cloud offers A3 instances (H100) and custom TPU capacity for AI workloads. Broad incumbent in the same customer segment, with proprietary accelerator options that compete with Horizon's NVIDIA-based fleet.

Direct peers

• RunPod: RunPod offers reserved and on-demand GPU instances plus serverless inference endpoints. Comparable on developer-facing GPU cloud and token/usage-based inference pricing.
• CoreWeave: CoreWeave is a specialized GPU cloud provider offering reserved and on-demand NVIDIA GPU capacity (H100, H200) and AI inference services. It is the most direct comparable to Horizon Compute's owned-fleet, inference-first model.
• Vast.ai: Vast.ai runs a marketplace and cloud for GPU compute (training and inference) at hourly rates. Comparable on GPU-hour pricing and developer-focused GPU access.
• Crusoe Energy: Crusoe owns and operates large-scale GPU clusters (including H100/H200) primarily for AI training and inference workloads. Comparable in owning the underlying infrastructure rather than reselling third-party cloud capacity.
• Lambda Labs: Lambda operates owned GPU clusters (H100/H200) for cloud rental and also offers hosted inference for open-weight models. Closely comparable in product mix (capacity + inference) and customer base (AI developers and researchers).
• Paperspace (DigitalOcean): Paperspace, now part of DigitalOcean, provides GPU virtual machines and notebooks for AI training/inference. Comparable on per-hour GPU compute targeted at AI developers and product teams.
• Together AI: Together AI provides GPU-backed cloud infrastructure and token-metered inference APIs for open-source/open-weight models. Directly comparable on inference-first GTM and frontier open-weight model focus.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks6 records

Key highlights6 records

Customer concentration

• Xx.com/horizon_compute
• LinkedInlinkedin.com/company/horizon-compute

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Horizon Compute