Habana

What Habana does

Habana Labs (legal name: Habana Labs, Ltd. an Intel Company) is an Israel-based AI semiconductor company founded in 2016 that designs and sells the Intel Gaudi family of deep learning accelerators. The company was acquired by Intel Corporation in 2019 for approximately $2 billion and now operates as a wholly-owned subsidiary, with its products marketed under the Intel Gaudi brand while retaining the Habana name within the developer community. Its product portfolio spans three hardware generations — Gaudi (16nm), Gaudi 2 (7nm, May 2022), and Gaudi 3 (2024) — complemented by the SynapseAI software suite, which includes a graph compiler, runtime, TPC kernel library, firmware, drivers, the HCCL collective communications library, HLML management library, TPC SDK, and a GPU Migration Toolkit. The architecture differentiates from competitors through purpose-built Tensor Processing Cores and the integration of 24x 200GbE ports per accelerator, enabling near-linear scaling above 99% efficiency up to 64 cards.

The company sells primarily to cloud providers, hyperscalers, and enterprises running large-scale AI training and inference. Distribution runs through two main channels: cloud, via Amazon EC2 DL1 instances (AWS's first non-GPU AI training instance, launched October 2021), and on-premises, via Supermicro X12 Gaudi Training Servers. Direct enterprise sales target customers in healthcare (Leidos medical imaging), fintech (Riskfuel, Fractal Analytics), autonomous vehicles (Mobileye, an Intel subsidiary), data storage (Seagate), and retail. Revenue is generated through hardware sales and accompanying software licensing, with cloud pricing available on a usage-based pay-as-you-go basis through AWS; on-premises hardware pricing is not publicly disclosed. The company's competitive proposition is anchored on up to 40% better price/performance versus GPU-based instances and a software stack designed to ease migration from Nvidia CUDA environments.

The AI accelerator market in which Habana operates is projected to expand from $28.5 billion in 2024 to over $360 billion by 2032, providing a substantial tailwind. Habana's strategic significance for Intel is high: it is the flagship alternative to Nvidia for cost-sensitive enterprise GenAI deployments, validated through MLPerf benchmarks and third-party evaluations from Hugging Face and customer case studies. Its principal risks are Nvidia's incumbent CUDA ecosystem lock-in, dependence on a single hyperscaler cloud listing (AWS), and the lack of disclosed standalone revenue given its subsidiary status within Intel.

Name

Habana

Legal name

Habana Labs, Ltd. an Intel Company

Website

https://habana.ai

Company type

Private

Founded year

2016

Operating status

Operating

Headcount range

101–250 employees

Ownership category

akta.pro rank

Product category

AI Semiconductor Accelerators

akta.pro primary industry

AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA)

Keywords

• AI training accelerators
• Deep learning processors
• Data center AI chips
• Neural network hardware
• AI inference silicon

Headquarters

HQ city

Caesarea

HQ country

Israel

HQ region

Middle East

Offices1 record

Markets served

GTM type

B2B

Offering type

Hardware or Manufacturing

Cost components

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

Revenue model

1. AI Accelerator Hardware Sales: Sale of Gaudi AI training and inference accelerators to cloud providers, enterprises, and through channel partners. Revenue generated from hardware product sales across first, second, and third generation Gaudi processors.
2. SynapseAI Software Licensing: Software stack including SynapseAI compiler, runtime, drivers, and developer tools licensed alongside hardware. Includes integration with TensorFlow and PyTorch frameworks.

Pricing tiers

Go-to-market motion3 records

Distribution channels3 records

Marketing channels5 records

Core offering

Habana Labs develops the Intel Gaudi family of AI accelerator processors purpose-built for deep learning training and inference workloads. The portfolio spans three hardware generations (Gaudi, Gaudi 2, Gaudi 3) bundled with the SynapseAI software suite, delivering price-performance advantages of up to 40% over comparable GPU-based solutions for large language model training, computer vision, and generative AI workloads.

Product overview

Habana Labs (an Intel company) develops the Intel Gaudi family of AI accelerators—deep learning processors designed for training and inference of large-scale AI models. The product portfolio includes three generations of hardware accelerators (Gaudi, Gaudi 2, and Gaudi 3) complemented by the Intel Gaudi Software Suite (SynapseAI), which provides the compiler, runtime, kernel libraries, and developer tools needed to train and deploy deep learning models. The platform supports both cloud deployment (via Amazon EC2 DL1 instances) and on-premises implementations through server partners like Supermicro. Habana's solutions target enterprise AI workloads including generative AI, computer vision, NLP, and domain-specific applications in autonomous vehicles, healthcare, fintech, and retail sectors.

Differentiator

Problem solved

Functional benefit

Products and services

• Intel Gaudi 3 AI Accelerator Third-generation deep learning training and inference processor purpose-built for generative AI enterprise applications, delivering competitive throughput and efficiency over prior generations. Sold to cloud providers, hyperscalers, and large enterprises for large-scale AI training and inference workloads.
• Intel Gaudi 2 AI Accelerator Second-generation AI accelerator built on 7nm process for large-scale deep learning training and inference workloads including LLMs and computer vision models. MLPerf results position Gaudi 2 as the only viable alternative to Nvidia H100 for training large language models like GPT-3.
• Intel Gaudi AI Accelerator (1st Generation) First-generation AI accelerator built on 16nm process with integrated 200GbE networking, delivering price/performance leadership for deep learning training and inference workloads. Features 24x 200GbE ports for massive system scale-out.
• Intel Gaudi Software Suite (SynapseAI) Comprehensive software platform including SynapseAI graph compiler and runtime, TPC kernel library, firmware, drivers, and developer tools such as the Habana profiler. Licensed alongside hardware to enable deep learning model development and migration from GPU-based workflows with native TensorFlow and PyTorch integration.
• Amazon EC2 DL1 Instances Cloud-based EC2 instances (DL1.24xlarge) powered by 8 Gaudi accelerators, offered through AWS marketplace on a pay-per-hour usage basis. Targets enterprises seeking cost-effective AI training infrastructure without upfront hardware investment.

Quantifiable outcome

• 40% better price-performance than GPU-based instances on EC2 DL1
• +6 more outcomes

Named customers5 records

Segments5 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration15 records

AI capability12 records

Feature7 records

Partnerships

Two partnerships are on record, tiered core.

• Amazon Web Services (AWS)coreChannel Partner/ Reseller/ DistributorAWS launched Amazon EC2 DL1 instances powered by Habana Gaudi accelerators in October 2021 - the first AI training instance by AWS not based on GPUs. AWS provides cloud marketplace access to Gaudi-based training infrastructure.
• SupermicrocoreChannel Partner/ Reseller/ DistributorSupermicro serves as channel partner for on-premises Gaudi-based training servers. Offers Supermicro X12 Gaudi Training Server and other configurations featuring 8 Gaudi accelerators per system.

Scale indicators6 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Nvidia: Dominant AI accelerator vendor whose H100, H200, and Blackwell GPUs are the primary competitive alternative to Gaudi. Nvidia's CUDA software ecosystem is the central switching barrier Habana must overcome in every enterprise deal.
• AMD: Direct competitor in AI accelerators with the Instinct MI300X/MI325X series, also targeting price-performance advantages over Nvidia. AMD competes head-to-head with Gaudi in cloud training and enterprise inference workloads.
• Cerebras Systems: AI accelerator startup building wafer-scale chips (CS-3) for large model training and inference. Competes with Gaudi for the same hyperscaler and enterprise training workloads, with a different architectural approach (wafer-scale vs. integrated-ethernet scale-out).
• SambaNova: AI accelerator company with DataScale SN40L systems targeting enterprise and government GenAI deployments. Closely comparable in scale, stage, and customer profile — both are Intel-adjacent alternative-AI-silicon plays.
• Graphcore: Intelligence Processing Unit (IPU) vendor competing in the same AI training/inference accelerator market. Provides a comparable alternative-architecture challenge to Nvidia GPUs alongside Gaudi.
• Groq: LPU-based inference accelerator startup competing for the same enterprise AI workloads, particularly in LLM inference. A relevant comparable in scale and customer profile within the alternative-AI-silicon category.

Emerging players

• Tenstorrent: AI accelerator startup building RISC-V-based AI chips for training and inference. An emerging alternative-architecture competitor with growing customer traction, though at earlier commercial maturity than Gaudi.
• Lightmatter: Photonic AI accelerator company pursuing a fundamentally different compute substrate (light-based). A longer-horizon alternative to Gaudi targeting the same large-scale AI training and inference market.

Broad incumbents

• Google TPU: Google's custom AI accelerator family (TPU v4/v5) used internally and offered to Google Cloud customers. A broader incumbent alternative that competes for the same cloud AI training workloads but with the unique advantage of vertical integration at a hyperscaler.
• AWS Trainium / Inferentia: AWS's custom silicon (Trainium2 for training, Inferentia for inference) — the same hyperscaler that hosts Gaudi on EC2 DL1. Represents both a partner and a direct internal competitor for cloud AI accelerator share within AWS.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/HabanaAI
• LinkedInlinkedin.com/company/habana-labs
• Xx.com/habanalabs

Revenue estimate

Valuation estimate

Number of profiles

Profiles5 records

Funding overview

Funding rounds2 records

Investors7 records

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 Habana