SambaNova
SambaNova designs and sells the Reconfigurable Dataflow Unit (RDU) AI inference chip and a full-stack platform (SambaCloud, SambaStack, SambaManaged, SambaRack) serving enterprises, sovereign AI operators, neoclouds, government laboratories, and developers with low-latency inference on large open-source models.
- Company typePrivate
- Founded2017
- HeadquartersPalo Alto, United States
- Headcount251–500
- GTM typeB2B
- OfferingHardware or Manufacturing
What SambaNova does
SambaNova is a privately held AI infrastructure company founded in 2017 in Palo Alto and now headquartered in San Jose, California. It designs and sells the Reconfigurable Dataflow Unit (RDU), a purpose-built inference accelerator based on its patented Dataflow Architecture — a grid of Programmable Compute Units (PCUs) and SRAM-based Programmable Memory Units (PMUs) combined with a three-tier memory hierarchy (on-chip SRAM, HBM, DDR) — packaged into the SN40L (fourth generation) and SN50 (fifth generation) chips. The company markets itself as an alternative to general-purpose GPUs for production AI inference, with claimed 5x faster speed, 3x lower TCO, and 4x energy savings, and its SambaRack systems are air-cooled at roughly 10 kW per rack, enabling deployment in standard data centers without liquid cooling.
SambaNova's commercial offering spans four tiers: SambaRack (hardware systems of 16 RDUs per rack), SambaCloud (a self-serve, usage-based inference API with OpenAI-compatible endpoints), SambaStack (a full-stack enterprise platform with SambaOrchestrator for model bundling and hot-swapping at inference time), and SambaManaged (a 90-day turnkey managed service for data centers, telecoms, and governments to launch their own branded inference cloud). The company goes to market via direct enterprise sales, the AWS Marketplace, regional distributors (notably TEPCO Systems in Japan), and a growing channel of sovereign AI infrastructure partners. Revenue mechanics therefore blend hardware sales, usage-based cloud inference, multi-year enterprise subscriptions, and recurring managed services.
The customer base spans enterprise inference providers and neoclouds (Together.ai, General Compute, OpenRouter, Maitai, Parasail, BlackBox.AI), sovereign AI operators (SCX in Australia, Argyll in the UK, Infercom in Germany, OVHcloud in France, SoftBank in Japan), government and national laboratories (Argonne, Oak Ridge, Los Alamos, TACC, Lawrence Livermore), and large enterprises across telecommunications, energy, imaging, and life sciences (SoftBank, Ricoh, TEPCO, Aion Labs, Hume.ai). The company has raised more than $1.4 billion across eight funding rounds, most recently an oversubscribed $350M+ Series E in February 2026 led by Vista Equity Partners and Cambium Capital with Intel Capital participation, and was reported in June 2026 to be in talks for an additional $800M–$1B at a $10 billion valuation. The CEO has publicly discussed IPO prospects, but SambaNova remains a private, venture- and private-equity-backed company.
SambaNova firmographics
Firmographics- Name
- SambaNova
- Legal name
- SambaNova, Inc.
- Website
- https://sambanova.ai
- Company type
- Private
- Founded year
- 2017
- Operating status
- Operating
- Headcount range
- 251–500 employees
- Short description
- SambaNova designs and sells the Reconfigurable Dataflow Unit (RDU) AI inference chip and a full-stack platform (SambaCloud, SambaStack, SambaManaged, SambaRack) serving enterprises, sovereign AI operators, neoclouds, government laboratories, and developers with low-latency inference on large open-source models.
- Ownership category
- akta.pro rank
Where SambaNova is headquartered
LocationHeadquarters
- HQ city
- Palo Alto
- HQ country
- United States
- HQ region
- North America
Offices2 records
Markets served
SambaNova business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Supply Chain, Personnel, Marketing or Sales, Infrastructure, Operations
Revenue model
- SambaRack hardware sales: One-time/recurring hardware sales of integrated SambaRack systems containing 16 RDU chips each, sold to enterprises, data centers, sovereign AI operators and inference providers.
- SambaCloud cloud inference service: Subscription and usage-based cloud inference offering OpenAI-compatible APIs for the largest open-source models (Llama, DeepSeek, Qwen, gpt-oss-120b), billed per token/usage; available via self-serve at cloud.sambanova.ai.
- SambaManaged managed inference cloud: Fully managed end-to-end service for data centers, telecoms and governments to launch their own branded AI inference cloud in 90 days; includes hardware, software, and operational support; recurring managed services revenue.
- SambaStack on-premises enterprise deployments: Dedicated enterprise deployments (on-premises or hosted) of SambaStack full-stack platform; long-cycle enterprise license/subscription contracts.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Usage-based | Pay-as-you-go | SambaCloud usage-based API pricing |
| Other | Multi-year contract | SambaManaged / enterprise contract pricing |
Go-to-market motion1 record
Distribution channels1 record
Marketing channels9 records
SambaNova product offering
Product offeringCore offering
SambaNova designs and manufactures proprietary Reconfigurable Dataflow Unit (RDU) AI inference chips (SN40L and SN50 generations) and packages them into air-cooled SambaRack systems. It sells these as the SambaStack enterprise platform for on-premises deployment, delivers inference as a service via SambaCloud, and offers SambaManaged turnkey cloud deployment. Its Dataflow Architecture minimizes data movement to deliver high-speed, low-latency inference for large open-source models, targeting agentic AI workloads.
Product overview
SambaNova offers a unified full-stack AI inference platform combining proprietary chip-to-cloud hardware (RDU chips and SambaRack systems), enterprise software (SambaStack), cloud services (SambaCloud), and managed deployment (SambaManaged). At the foundation is the Dataflow Architecture powering the Reconfigurable Dataflow Unit (RDU) chips — currently in their fourth (SN40L) and fifth (SN50) generations. SambaCloud is the public cloud inference platform for developers, SambaStack is the enterprise full-stack offering for on-premises or dedicated cloud deployments with model bundling, SambaManaged is the turnkey managed service that brings SambaNova's stack into customer data centers within 90 days, and SambaRack is the integrated air-cooled hardware system that hosts RDU chips in production.
Differentiator
Problem solved
Functional benefit
Brands
- SambaCloud: Full-stack AI inference cloud platform for running large open-source models with fast inference
- SambaStack
- SambaManaged
- SambaRack
- Dataflow Architecture
Products and services
- SambaCloud
- SambaStack
- SambaManaged
Quantifiable outcome
- 5x faster AI inference than competitive chips (SN50)
- +8 more outcomes
Companies that use SambaNova
Customer profileNamed customers21 records
Segments5 records
Ideal customer profiles5 records
SambaNova technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration7 records
AI capability12 records
Feature7 records
SambaNova partnerships and signals
Strategic signalPartnerships
18 partnerships are on record, tiered strategic, flagship and core.
- Vector Core Compute (VC2)strategicVista Equity Partners and Cambium Capital launched VC2, a 'disaggregated agentic neocloud' built on Intel Xeon CPUs, SambaNova RDUs, and NVIDIA Blackwell GPUs. Together.ai is the first commercial customer. Live demonstration of B200+SN40 disaggregated inference (2x faster than B200-only).
- Together.aistrategicFirst commercial customer of Vector Core Compute's disaggregated inference cloud running on SambaNova RDUs.
- FoxconnstrategicAt Computex 2026, Foxconn and Intel announced a rack-scale AI infrastructure partnership in which Foxconn integrates and manufactures rack-scale AI infrastructure pairing Intel Xeon CPUs with SambaNova SN-50 RDUs, targeting hyperscale and enterprise deployments.
- General ComputestrategicAI inference neocloud startup with $300M of SambaNova SN50 chips on order; launched General Compute Cloud (first ASIC-native neocloud); claims fastest independently benchmarked speeds on MiniMax M2.7.
- TEPCO Systems CorporationflagshipSigned distributor agreement to bring SambaNova's energy-efficient AI infrastructure to enterprises across Japan; foundation for TEPCO Group's next-generation AI system platform and offered to external customers.
- Intel CorporationflagshipMulti-year strategic collaboration to deliver high-performance, cost-efficient AI inference solutions. Intel plans a strategic investment in SambaNova to accelerate rollout of an Intel-powered AI cloud; collaboration spans AI cloud expansion on Intel Xeon infrastructure, integrated AI infrastructure combining SambaNova systems with Intel CPUs/accelerators/networking, and joint go-to-market execution through Intel's global enterprise, cloud, and partner channels. Lip-Bu Tan (Intel CEO) serves as SambaNova's Chairman.
- OVHcloudflagshipSelected by OVHcloud to power its flagship AI endpoints inferencing service in Europe; provides premium low-latency AI inference on SambaStack infrastructure.
- SCX (SouthernCrossAI)flagshipSovereign AI partnership to launch Australia's first ASIC-based sovereign AI cloud, powered by SambaNova SN40L and SN50 chips; integrated with Equinix Fabric AI ecosystem; offers 10x lower inference energy costs and meets national data residency requirements.
- Argyll Data DevelopmentflagshipSovereign AI partnership to launch the UK's first containerized sovereign inference cloud powered by wind, solar, and wave energy; runs SambaNova air-cooled SN40L systems; part of the Killellan AI Growth Zone on Scotland's Cowal Peninsula (£15B investment, 100-600MW to 2GW).
- InfercomflagshipSovereign AI partnership to launch Germany's first Inference platform providing fully EU-compliant, GDPR-safe AI infrastructure for startups, enterprises, and government agencies; powered by SambaNova with data kept within European borders.
- MetastrategicSambaNova is the official launch partner for Meta's Llama 4 series, providing fast inference on Scout and Maverick models; SambaCloud was the first platform to support all three variants of Llama 3.1 (8B, 70B, 405B).
- Argonne National LaboratorycoreDeploys SambaNova hardware to support AI inference for research; member of SambaNova research ecosystem; Rick Stevens quoted in marketing.
- Oak Ridge National LaboratorycoreEcosystem partner using SambaNova for fast inferencing work in scientific research.
- Texas Advanced Computing Center (TACC)coreEcosystem partner using SambaNova for high-performance computing research.
- Lawrence Livermore National LaboratorycoreEcosystem partner using SambaNova for AI for science research.
- Los Alamos National LaboratorycoreListed as SambaNova customer/research partner in ecosystem.
- NSF NAIRR Pilot ProgramstrategicSambaNova participates in the National AI Research Resource (NAIRR) Pilot, connecting the private sector, government agencies, and academia to create a national AI infrastructure.
- LatticeFlow AIstrategicPartnered with SambaNova to deliver enterprise-grade security and compliance for open-source AI; co-developed blueprint for governing agentic AI in financial services with Unique AI.
Scale indicators11 records
Recent moves6 records
Expansion highlights7 records
SambaNova competitors and assessment
Company assessmentBroad incumbents
- NVIDIA: Dominant AI accelerator supplier whose GPUs (H100, B200, B300) define the inference performance and ecosystem (CUDA, TensorRT, Triton) that SambaNova's RDU chips compete against. SambaNova's own disaggregated blueprint actually pairs RDUs with NVIDIA GPUs, illustrating how the two products are complementary as well as competitive.
- AWS (Trainium / Inferentia): Amazon's in-house custom AI silicon for cloud inference (Inferentia) and training (Trainium), deployed inside AWS and available to external customers. Represents the hyperscaler in-house alternative that SambaNova competes against for cloud-provider workloads.
- Google (TPU): Google's Tensor Processing Unit is the most mature hyperscaler custom AI accelerator, available via Google Cloud. Sets the benchmark for performance-per-dollar on inference workloads and is the in-house alternative to external accelerators like SambaNova for the largest AI buyer in the world.
Direct peers
- Groq: Pure-play AI inference accelerator company whose LPU (Language Processing Unit) targets the same low-latency, high-throughput inference workloads as SambaNova's RDU. Both companies pitch dramatically lower latency and cost-per-token versus general-purpose GPUs, and both sell to inference providers and enterprise AI buyers.
- Cerebras Systems: AI chip startup whose wafer-scale engine (WSE) competes with SambaNova's RDU for large-model inference and training workloads. Both are private, well-funded US-based custom-silicon companies pursuing the same 'Nvidia displacement' thesis with fundamentally different architectures.
- Tenstorrent: AI accelerator company led by Jim Keller, designing RISC-V based chips for training and inference at scale. Targets the same enterprise and hyperscaler inference market as SambaNova with a similar full-stack software + hardware strategy.
- Intel (Habana Labs): Intel's Habana Gaudi accelerators compete directly with SambaNova's RDU for training and inference workloads. Notably, Intel is also SambaNova's largest strategic investor and partner, making this a hybrid competitor-collaborator relationship — Intel sells Gaudi into the same enterprise and cloud accounts while jointly marketing the heterogeneous inference blueprint with SambaNova.
- d-Matrix: AI inference accelerator startup whose digital in-memory compute (DIMC) chips target the same low-latency, high-throughput LLM inference workloads as SambaNova's RDU. Both are private companies targeting cloud providers and neoclouds for production agentic AI serving.
- Lightmatter: Photonics-based AI accelerator company targeting high-throughput inference and inter-chip communication. Competes for the same data center and cloud inference accounts as SambaNova, with a fundamentally different underlying technology (light vs. electronic dataflow).
Emerging players
- Positron AI: Early-stage AI inference accelerator company whose memory-heavy architecture targets the same large-model, low-latency inference use cases as SambaNova's RDU. Smaller and earlier-stage than SambaNova, but pursues a similar 'inference-first silicon' thesis for cloud and neocloud customers.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat6 records
Key risks7 records
Key highlights7 records
Customer concentration
SambaNova social profiles
Digital presenceSambaNova compliance and trust
Trust signalCompliance2 records
SambaNova financial estimates
Financial estimateRevenue estimate
Valuation estimate
SambaNova leadership team
Management profileNumber of profiles
Profiles18 records
SambaNova subsidiaries and ownership
Company hierarchySubsidiaries1 record
SambaNova funding detail
Funding detailFunding overview
Funding rounds9 records
Investors28 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
SambaNova M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about SambaNova
What does SambaNova do?
SambaNova designs and manufactures proprietary Reconfigurable Dataflow Unit (RDU) AI inference chips (SN40L and SN50 generations) and packages them into air-cooled SambaRack systems. It sells these as the SambaStack enterprise platform for on-premises deployment, delivers inference as a service via SambaCloud, and offers SambaManaged turnkey cloud deployment. Its Dataflow Architecture minimizes data movement to deliver high-speed, low-latency inference for large open-source models, targeting agentic AI workloads.
Is SambaNova a public or private company?
SambaNova is a private company. It is classified as venture growth investor backed and is currently operating.
When was SambaNova founded?
SambaNova was founded in 2017. It employs 251 to 500 people.
Where is SambaNova based?
SambaNova is headquartered in Palo Alto, United States, in the North America region.
How does SambaNova make money?
Four revenue lines are on record. SambaRack hardware sales are the primary driver. The others are sambaCloud cloud inference service, sambaManaged managed inference cloud and sambaStack on-premises enterprise deployments.
Who are SambaNova's main competitors?
Broad incumbents on record are NVIDIA, AWS (Trainium / Inferentia) and Google (TPU). Direct peers are Groq, Cerebras Systems, Tenstorrent, Intel (Habana Labs), d-Matrix and Lightmatter. Positron AI is listed as an emerging player.
Does SambaNova have an API?
Yes. SambaNova offers a public, OpenAI-compatible REST API for AI inference through SambaCloud. Developers can set OPENAI_API_KEY to their SambaNova API Key and change the base URL to port existing applications to SambaNova in minutes. The API supports large open-source models including DeepSeek, Llama, and Qwen, and provides Bring Your Own Checkpoint capability. Simple-to-integrate APIs enable quick onboarding, and developers can connect to the platform, manage models, and scale workloads with minimal changes. Developer documentation is at docs.sambanova.ai.