SambaNova Systems is an American artificial intelligence hardware and software company headquartered in San Jose, California. Founded in November 2017 by Stanford University professors Kunle Olukotun and Christopher Ré with semiconductor executive Rodrigo Liang, the company develops Reconfigurable Dataflow Unit (RDU) processors and integrated systems for training and inference of large language models.[1][2] Following an April 2025 restructuring, SambaNova concentrated its commercial strategy on inference through the SN40L processor, SambaCloud, SambaStack, and SambaManaged.[3][4] In July 2026, the company announced the $1 billion first close of a Series F round at an $11 billion post-money valuation, bringing its announced financing since 2018 to more than $2.48 billion.[5] Investors across its rounds have included SoftBank Vision Fund, General Atlantic, Vista Equity Partners, Intel Capital, BlackRock, GV, and others.[5][6]

SambaNova logo, from the official website sambanova.ai. Trademark of its owner.

SambaNova competes with NVIDIA, Groq, and Cerebras Systems in the AI accelerator market. The company emphasizes low-latency inference. At SambaNova Cloud's September 2024 launch, SambaNova said Artificial Analysis independently measured its full-precision Llama 3.1 405B endpoint at 132 output tokens per second, the highest output speed among endpoints of comparable intelligence tracked by the benchmarker at that time.[7]

History

Founding and Early Years (2017-2019)

SambaNova Systems was incorporated in 2017 by three co-founders with backgrounds in processor architecture and machine learning research at Stanford University. Kunle Olukotun, the Cadence Design Professor of Electrical Engineering and Computer Science at Stanford, is a pioneer of multicore processor design. He founded Afara Websystems, whose Niagara processor work became a basis for Sun Microsystems' multicore server chips.[1][8] Christopher Ré, an associate professor of computer science at Stanford and a MacArthur Fellow, brought expertise in machine learning systems and data management. He co-founded Lattice Data, which Apple acquired in 2017, and later co-founded Snorkel AI and SambaNova.[1][9] Rodrigo Liang, SambaNova's CEO, previously led semiconductor engineering teams at Sun Microsystems and Oracle and holds bachelor's and master's degrees in electrical engineering from Stanford.[1]

The founders argued that general-purpose processors were inefficient for the dataflow patterns of deep learning. Their approach mapped computation graphs onto reconfigurable compute, memory, and interconnect resources rather than treating a model as a sequence of GPU kernels.[10]

In March 2018, while operating in stealth, SambaNova raised a $56 million Series A co-led by Walden International and GV, with participation from Redline Capital Management and Atlantic Bridge. An SEC Form D records $56.6 million of Series A securities sold. The company remained in stealth until December 2020.[2][10][11] Walden is a venture capital firm founded by Lip-Bu Tan, who later became SambaNova's chairman and, in March 2025, Intel's CEO.[1]

Growth and Product Development (2019-2021)

In April 2019, SambaNova closed a $150 million Series B round led by Intel Capital, with continued participation from GV and existing investors.[10] The company's first-generation SN10 RDU was taped out on TSMC's 7nm process in the first half of 2019.[12]

The company's first commercial product, the DataScale SN10 system, entered production in 2020.[12] In February 2020, SambaNova announced a $250 million Series C round led by funds and accounts managed by BlackRock, with participation from GV, Intel Capital, Walden International, WRVI Capital, and Redline Capital.[10]

By 2021, SambaNova had secured deployments at several U.S. Department of Energy national laboratories. In April 2021, the company raised a $676 million Series D round led by SoftBank Vision Fund 2, with participation from new investors Temasek and GIC alongside existing backers including BlackRock, Intel Capital, GV, and Walden International. The company said the round valued it at more than $5 billion and brought total funding to more than $1 billion.[6]

Strategic Pivot to Inference (2022-2024)

In September 2022, SambaNova launched its second-generation DataScale SN30 system. The Cardinal SN30 used two compute dies in one RDU socket and doubled the number of hardware tiles from the SN10 generation. SambaNova disclosed 688 teraflops of peak bfloat16 performance and reported up to a sixfold speedup over an NVIDIA A100 system on selected training workloads; the performance comparison was a company benchmark.[13][14]

In September 2023, SambaNova introduced the SN40L, its fourth-generation RDU. A 2024 architecture paper describes a dual-die TSMC 5 nm device with 1,040 Pattern Compute Units (PCUs), 1,040 Pattern Memory Units (PMUs), 520 MiB of on-chip SRAM, 64 GiB of HBM, up to 1.5 TiB of attached DDR memory, and 638 BF16 teraflops of peak compute. The chip was co-designed for training and inference and introduced a three-tier memory hierarchy used in SambaNova's Composition of Experts system.[15]

This period also marked a shift toward generative AI inference. On September 10, 2024, SambaNova launched SambaNova Cloud on SN40L hardware. At launch, the company offered Meta's Llama 3.1 70B model at a stated 461 output tokens per second and the 405B model at 132 tokens per second, both at 16-bit precision. SambaNova's release included an Artificial Analysis statement that the benchmarker had independently measured the 405B endpoint at 132 tokens per second.[7]

Restructuring and Inference-Only Pivot (2025)

On April 22, 2025, SambaNova cut 77 positions in California. Reporting based on the state WARN notice described the reduction as about 15% of a roughly 500-person workforce. A company statement said the changes aligned the organization with a shift from model training toward fine-tuning and inference and would concentrate it on cloud-first deployment of open models.[16][17]

On May 29, 2025, SambaNova made its AI platform available through AWS Marketplace. Customers could use existing AWS billing and connect privately through AWS PrivateLink; the initial listing highlighted Meta's Llama 4 Maverick and DeepSeek R1 671B.[18]

In July 2025, the company introduced SambaManaged and reorganized its portfolio around SambaCloud, SambaStack, and SambaManaged. The three products shared SambaRack hardware and a Kubernetes-based control plane called SambaOrchestrator.[3] SambaNova described SambaManaged as a fully managed inference system that could be installed in an existing data center in about 90 days. Its product sheet describes air-cooled SN40L systems averaging 10 kW per rack and deployments scalable to 1 MW.[4]

Intel Talks, Series E, and SN50 (2025-2026)

In December 2025, Bloomberg reported that Intel was in advanced talks to acquire SambaNova for about $1.6 billion including debt. The reported figure was a prospective transaction value, not a new financing valuation, and the report cautioned that the terms could change.[19] Reuters reported in February 2026 that the talks had stalled.[20]

On February 24, 2026, SambaNova announced more than $350 million in Series E financing led by Vista Equity Partners and Cambium Capital, with strong participation from Intel Capital and investments from new and existing backers. The company did not disclose a valuation or the size of Intel's investment.[21] The same day, SambaNova and Intel announced a planned multi-year collaboration built around Xeon infrastructure. Intel said the collaboration complemented, rather than replaced, its own data-center GPU roadmap.[22]

SambaNova also announced the SN50, its fifth-generation RDU for agentic inference, with customer shipments targeted for the second half of 2026 and SoftBank as the first announced customer.[21][23] In April 2026, Intel and SambaNova described a heterogeneous blueprint in which GPUs perform prefill, RDUs perform decode, and Xeon 6 processors serve as host and action CPUs.[24]

On July 8, 2026, SambaNova announced the $1 billion first close of a Series F round led by General Atlantic, with significant investment from Seligman Ventures and accounts advised by T. Rowe Price. The company reported an $11 billion post-money valuation. It also said JPMorganChase had selected SambaNova as an on-premises inference partner and would deploy SN40 and SN50 systems.[5]

Technology

How does SambaNova's Reconfigurable Dataflow Architecture (RDA) work?

SambaNova's Reconfigurable Dataflow Architecture maps a model's computation graph onto a spatial network of compute, memory, address-generation, and routing resources. This differs from the bulk-synchronous execution model commonly used in GPUs, where operators typically run as successive kernels.[10][15]

The RDA uses four main on-chip elements:

Component	Function
Pattern Compute Units (PCUs)	Perform arithmetic through configurable systolic or pipelined SIMD datapaths.
Pattern Memory Units (PMUs)	Provide distributed, software-managed SRAM for model weights, activations, metadata, and intermediate values.
Address Generation and Coalescing Units (AGCUs)	Generate and combine memory accesses for dataflow operations.
Reconfigurable Dataflow Network switches	Route vector, scalar, and control traffic among the compute and memory units.

SambaFlow compiles a model graph into this hardware configuration. The architecture can fuse and pipeline operators while retaining programmability for different model structures. SambaNova's SN40L paper reports that this approach reduced off-chip movement and improved utilization on the paper's tested workloads, but its comparisons should be read in the context of the configurations and baselines used in that study.[15]

RDU Chip Generations

SambaNova has disclosed four named RDU generations:

Generation	Chip	Process node	Publicly disclosed specifications	Year
1st	SN10	TSMC 7 nm	640 PCUs, 640 PMUs, more than 300 MiB on-chip memory, more than 300 BF16 TFLOPS, 40 billion transistors	2020
2nd	SN30	TSMC 7 nm	8 tiles per RDU and 688 BF16 TFLOPS	2022
4th	SN40L	TSMC 5 nm	1,040 PCUs, 520 MiB SRAM, 64 GiB HBM, up to 1.5 TiB DDR, 638 BF16 TFLOPS	2023
5th	SN50	Not publicly specified	5 times the compute per accelerator and 4 times the network bandwidth of SN40L; scales to 256 accelerators	Announced 2026

SambaNova presented the SN10 at Hot Chips 33 in August 2021. Its presentation documented the 7 nm process, 40 billion transistors, 640 PCUs, 640 PMUs, and more than 300 BF16 teraflops.[12] A peer-reviewed SN10 paper appeared at the 2022 IEEE International Solid-State Circuits Conference.[25] The SN40L paper documented the newer chip's dual-die 5 nm design and three-tier SRAM, HBM, and DDR memory system.[15]

The SN50 was announced on February 24, 2026. SambaNova says a SambaRack SN50 contains 16 RDUs, provides five times more compute per accelerator and four times more network bandwidth than SN40L, and can connect up to 256 accelerators over a multi-terabyte-per-second interconnect. The company says the system supports models up to 10 trillion parameters and context lengths up to 10 million tokens. Shipment was scheduled for the second half of 2026, so these launch specifications and performance comparisons were forward-looking company claims rather than results from broadly available production systems.[21][23]

At launch, SambaNova claimed up to five times the maximum speed and roughly three times the throughput of NVIDIA's B200 on selected workloads. Its release stated that the B200 Llama 3.3 70B figure came from SemiAnalysis while the SN50 result came from SambaNova, so the comparison did not come from one independent, like-for-like benchmark run.[21] In July 2026, SambaNova reported that Artificial Analysis benchmarked a preview system using four NVIDIA H200 GPUs for prefill and 16 SN50 RDUs for decode on MiniMax M2.7. The company reported up to 850 decode tokens per second for short-context workloads and more than 450 for long-context workloads in that configuration.[26] SoftBank was the first announced SN50 customer.[21]

SambaFlow Software Stack

SambaFlow is SambaNova's compiler and runtime software stack that bridges standard ML frameworks and the RDU hardware. Developers write models in PyTorch or TensorFlow, and SambaFlow automatically extracts the computational graph, optimizes it for dataflow execution, and maps it onto the RDU's PCUs and PMUs.

The SambaFlow stack includes:

SambaFlow Compiler: Translates high-level model definitions into RDU configuration sequences, handling tiling, weight and data partitioning, and flow control automatically.
SambaFlow Runtime: Manages communication with the DataScale hardware, including hardware initialization, error handling, resource management, and process scheduling.
SambaFlow Python SDK: Provides a developer-facing API for creating, compiling, and running models on the RDU.

Because the compiler handles low-level optimization, developers do not need to write custom CUDA kernels or hand-tune memory layouts, which SambaNova positions as an advantage over GPU-based workflows that often require significant kernel engineering for peak performance.

Products and Services

Following the July 2025 "SambaNova 2.0" reorganization, SambaNova offers three primary product lines, all powered by SambaRack hardware containing 16 SN40L RDUs (with SN50-based SambaRacks rolling out from late 2026) and orchestrated by a Kubernetes-based platform called SambaOrchestrator.[3][23]

SambaCloud

SambaCloud, launched as SambaNova Cloud in September 2024 and renamed in July 2025, is the company's hosted inference service. It provides API access to open-source and open-weight language models on SambaRack systems.[3][7]

The service exposes an OpenAI-compatible endpoint. Its model catalog changes over time and has included Meta's Llama family, DeepSeek, Qwen, and other models.[18][27] Selected launch results were:

Model	Stated output speed	Qualification
Llama 3.1 70B	461 tokens per second	SambaNova launch result at 16-bit precision
Llama 3.1 405B	132 tokens per second	SambaNova launch result; the release included independent measurement from Artificial Analysis

These figures describe the September 2024 launch and are not a current service-level guarantee. Since May 2025, customers have also been able to procure the platform through AWS Marketplace and use AWS PrivateLink for private connectivity.[18]

SambaStack

SambaStack is SambaNova's dedicated-infrastructure offering, succeeding the earlier DataScale appliances. Customers can deploy systems on premises or use dedicated capacity hosted by SambaNova, with SambaFlow and the broader software stack provided for model deployment and management.[3]

System	Configuration	Form factor or status
DataScale SN10-8 (legacy)	8 SN10 RDUs	Quarter rack
DataScale SN30 (legacy)	8 SN30 RDUs	Node and rack-scale configurations
SambaRack SN40L	16 SN40L RDUs	Single air-cooled rack, typically about 10 kW
SambaRack SN50	16 SN50 RDUs	Announced; shipments targeted for the second half of 2026

SambaManaged

SambaManaged, introduced in July 2025, is a turnkey offering for data-center operators and cloud providers. SambaNova supplies and manages the AI hardware and software while the host provides power, space, and network connectivity. The company initially advertised installation in approximately 90 days instead of the 18-24 months it said a purpose-built AI facility could require.[3][4]

Key characteristics include:

Air-cooled racks: The SN40L system is specified at about 10 kW typical power and can operate with air cooling.
Modular scaling: The product sheet describes deployments from a small cluster up to 1 MW.
End-to-end management: SambaNova manages deployment, operations, and maintenance, while SambaOrchestrator handles workload scheduling across SambaRacks.[3][4]

What is Composition of Experts (CoE)?

The Composition of Experts (CoE) is a model-serving architecture developed by SambaNova that allows multiple specialized, fully trained models ("experts") to be orchestrated behind a single API endpoint. Unlike a mixture of experts architecture where expert sub-networks exist within a single model, CoE treats each expert as a standalone model and routes queries to the appropriate expert based on the task.[15]

Key characteristics of CoE include:

Multi-Model Serving: Hundreds of domain-specific models (for example, finance, legal, engineering, and healthcare) can reside in the RDU's large memory tiers and be served from a single system.
Dynamic Routing: An orchestration layer routes incoming requests to the best-suited expert model, enabling broad coverage without requiring a single monolithic model.
Model Ownership and Privacy: Organizations can fine-tune their own expert models with proprietary data while maintaining ownership and access controls.
Efficient Memory Utilization: The RDU's three-tier memory hierarchy allows many models to be loaded simultaneously, avoiding the latency of swapping models in and out of GPU memory; on the SN50, models resident in HBM and SRAM can be hot-swapped in milliseconds, a property SambaNova emphasizes as essential for agentic workloads that route between many specialized models.[23]

SambaNova released Samba-CoE v0.1 and subsequent versions as demonstrations of this architecture, showing that a collection of smaller specialized models can match or exceed the performance of much larger general-purpose models on domain-specific tasks.

Enterprise and Government Deployments

SambaNova systems have been installed at U.S. national laboratories and research centers, as well as at enterprise and cloud customers.

National Laboratory Partnerships

Institution	Deployment details
Argonne National Laboratory	Argonne's AI Testbed includes SN30 and SN40L systems. A 2024 deployment added 16 SN40L RDUs for inference and evaluation of scientific foundation models.[28]
Lawrence Livermore National Laboratory (LLNL)	LLNL integrated an SN10 DataScale system with the Corona cluster in 2020 and expanded the collaboration in 2023 for its cognitive simulation program.[29]
Los Alamos National Laboratory	A DataScale system was integrated into the Darwin heterogeneous platform, initially for quantum chemistry, and the laboratory later selected SambaNova Suite for generative AI work.[29][30]
Texas Advanced Computing Center (TACC)	TACC deployed SambaNova Suite for AI inference and provides a SambaNova-backed experimental inference service.[31]
RIKEN Center for Computational Science	RIKEN adopted DataScale in 2023 to study integration of AI with the Fugaku supercomputer, including digital-twin research.[32]

Enterprise Customers

Publicly announced deployments and partnerships include:

Aramco: SambaNova said its hardware in Saudi Arabia supported Aramco's internal Metabrain language model.[33]
OTP Bank: The bank selected SambaNova to build an on-premises AI supercomputer for Hungarian-language and financial-services models.[34]
SoftBank: SoftBank hosted SambaCloud in Japan and became the first announced SN50 customer.[21]
OVHcloud: OVHcloud selected SambaStack to augment its AI Endpoints service in Europe.[35]
stc: stc.ai and SambaNova announced a sovereign inference service in Saudi Arabia based on Llama 405B.[36]
JPMorganChase: In July 2026, SambaNova said the bank had selected SN40 and SN50 systems for secure on-premises inference.[5]

Funding History

SambaNova has announced six major funding rounds:

Round	Date	Amount	Lead investor or investors	Disclosed valuation
Series A	March 2018	$56 million	Walden International and GV	Not disclosed
Series B	April 2019	$150 million	Intel Capital	Not disclosed
Series C	February 2020	$250 million	Funds and accounts managed by BlackRock	Not disclosed
Series D	April 2021	$676 million	SoftBank Vision Fund 2	More than $5 billion
Series E	February 2026	More than $350 million	Vista Equity Partners and Cambium Capital	Not disclosed
Series F, first close	July 2026	$1 billion	General Atlantic	$11 billion post-money
Announced total		More than $2.48 billion

The Series A through Series C figures are documented in company materials and the 2018 Series A Form D; the Series D, Series E, and Series F terms come from the respective company announcements.[2][5][6][10][11][21]

The more than $5 billion Series D valuation was the company's disclosed peak until the Series F. The roughly $1.6 billion figure reported during Intel acquisition talks in December 2025 included debt and described a possible transaction, so it should not be treated as a financing-round valuation.[19] SambaNova did not disclose a Series E valuation. In July 2026, it reported an $11 billion post-money valuation for the first close of Series F.[5]

Competitive Landscape

SambaNova operates in a competitive market for AI inference hardware and services:

Company	Architecture	Main distinction
NVIDIA	GPU and CUDA	Broad accelerator portfolio and mature CUDA software ecosystem, including H100 and B200 systems
Groq	Language Processing Unit	Deterministic architecture focused on inference latency
Cerebras Systems	Wafer-Scale Engine	Wafer-scale processors with large on-chip compute and memory resources
AMD	GPU and ROCm	Instinct accelerators and the ROCm software stack
Google	TPU	Custom accelerators used internally and offered through Google Cloud
Intel	Xeon CPUs and data-center GPUs	Collaborates with SambaNova on heterogeneous inference while continuing its own GPU roadmap

SambaNova's main architectural distinction is its combination of reconfigurable dataflow execution and a three-tier SRAM, HBM, and DDR memory hierarchy. The company argues that this reduces model-switching and memory-movement overhead for multi-model inference. The SN40L paper reports performance advantages on its selected CoE workloads, but those results are specific to the paper's systems, models, precision, and baselines.[15] Intel's 2026 statements also make clear that its SambaNova collaboration does not end Intel's direct participation in the AI accelerator market.[22]

Leadership

Name	Title	Background
Rodrigo Liang	Co-Founder and CEO	Former semiconductor engineering leader at Sun Microsystems and Oracle; holds BS and MS degrees in electrical engineering from Stanford.[1]
Kunle Olukotun	Co-Founder and Chief Technologist	Stanford's Cadence Design Professor of Electrical Engineering and Computer Science; multicore processor pioneer, founder of Afara Websystems, and recipient of the 2023 ACM-IEEE CS Eckert-Mauchly Award.[1][8]
Christopher Ré	Co-Founder	Stanford computer science professor and MacArthur Fellow whose research spans data systems and machine learning; co-founder of Lattice Data and Snorkel AI.[1][9]
Lip-Bu Tan	Chairman	Founder of Walden International and Walden Catalyst Ventures; CEO of Intel since March 2025.[1]

See Also

References

SambaNova. "Our Team Pushing AI Forward." Accessed July 28, 2026. https://sambanova.ai/company/team
U.S. Securities and Exchange Commission. "SambaNova Systems, Inc. Form D." Filed March 15, 2018. https://www.sec.gov/Archives/edgar/data/1733073/000173307318000001/xslFormDX08/primary_doc.xml
SambaNova. "SambaNova 2.0: Build with Relentless Intelligence." July 8, 2025. https://sambanova.ai/blog/sambanova-2.0-build-with-relentless-intelligence
SambaNova. "SambaManaged" data sheet. July 2025. https://sambanova.ai/hubfs/SambaManaged%20data%20sheet%2007%2002%2025.pdf
SambaNova. "SambaNova Completes First Close of $1B Financing at $11B Valuation." July 8, 2026. https://sambanova.ai/press/sambanova-completes-first-close-of-1b-financing-at-11b-valuation
SambaNova Systems. "SambaNova Systems Raises $676M in Series D, Surpasses $5B Valuation." April 13, 2021. https://www.businesswire.com/news/home/20210413005263/en/
SambaNova. "SambaNova Launches The World's Fastest AI Platform." September 10, 2024. https://sambanova.ai/press/worlds-fastest-ai-platform
Association for Computing Machinery. "Oyekunle Olukotun: 2023 ACM-IEEE CS Eckert-Mauchly Award." 2023. https://awards.acm.org/award_winners/olukotun_3961927
Stanford Data Science. "Christopher Re." Accessed July 28, 2026. https://datascience.stanford.edu/people/chris-re
SambaNova Systems. "AI Is Changing Everything You Know About Hardware and Software." 2021. https://sambanova.ai/hubfs/23945802/SambaNova_AIisChangingEverything2021_Whitepaper_English.pdf
U.S. Securities and Exchange Commission. "SambaNova Systems, Inc. Form D filing detail." March 15, 2018. https://www.sec.gov/Archives/edgar/data/1733073/000173307318000001/0001733073-18-000001-index.htm
Prabhakar, Raghu, and Sumti Jairath. "SambaNova SN10 RDU: Accelerating Software 2.0 with Dataflow." Hot Chips 33, August 2021. https://hc33.hotchips.org/assets/program/conference/day2/SambaNova%20HotChips%202021%20Aug%2023%20v1.pdf
SambaNova. "Transition to DataScale SN30." SambaNova Documentation. Accessed July 28, 2026. https://docs-legacy.sambanova.ai/developer/latest/transition-to-sn30.html
HPCwire. "SambaNova Launches Second-Gen DataScale System." September 14, 2022. https://www.hpcwire.com/2022/09/14/sambanova-launches-second-gen-datascale-system/
Prabhakar, Raghu, et al. "SambaNova SN40L: Scaling the AI Memory Wall with Dataflow and Composition of Experts." arXiv:2405.07518, May 2024. https://arxiv.org/abs/2405.07518
California Employment Development Department. "WARN Report for July 1, 2024 to June 30, 2025." 2025. https://edd.ca.gov/siteassets/files/jobs_and_training/warn/warn-report-for-7-1-2024-to-06-30-2025.pdf
Data Center Dynamics. "SambaNova lays off 77 employees as company pivots focus from training to inference." May 6, 2025. https://www.datacenterdynamics.com/en/news/sambanova-lays-off-77-employees-as-company-pivots-focus-from-training-to-inference/
SambaNova. "SambaNova Launches its AI Platform in AWS Marketplace." May 29, 2025. https://sambanova.ai/press/sambanova-launches-its-ai-platform-in-aws-marketplace
Bloomberg News. "Intel Is Said to Near $1.6 Billion Deal for Chip Firm SambaNova." December 12, 2025. https://news.bloomberglaw.com/capital-markets/intel-is-said-to-near-1-6-billion-deal-for-chip-firm-sambanova
Reuters. "AI chip startup SambaNova raises $350 million in Vista-led round, signs Intel partnership." February 24, 2026. https://www.investing.com/news/stock-market-news/ai-chip-startup-sambanova-raises-350-million-in-vistaled-round-signs-intel-partnership-4522565
SambaNova. "SambaNova Unveils Fastest Chip for Agentic AI, Collaborates with Intel, and Raises $350M+." February 24, 2026. https://sambanova.ai/press/sambanova-unveils-fastest-chip-for-agentic-ai-collaborates-with-intel-and-raises-350m
Intel. "Intel, SambaNova Planning Multi-Year Collaboration for Xeon-Based AI Inference." February 24, 2026. https://newsroom.intel.com/data-center/intel-and-sambanova-planning-multi-year-collaboration-for-xeon-based-ai-inference
SambaNova. "Introducing the SN50 RDU: Purpose-Built for Agentic Inference." February 24, 2026. https://sambanova.ai/blog/introducing-the-sn50-rdu-purpose-built-for-agentic-inference
Intel. "Intel and SambaNova Advance Agentic AI with Xeon 6." April 8, 2026. https://newsroom.intel.com/artificial-intelligence/intel-and-sambanova-advance-agentic-ai-with-xeon-6
Prabhakar, Raghu, Sumti Jairath, and Jinuk Luke Shin. "SambaNova SN10 RDU: A 7nm Dataflow Architecture to Accelerate Software 2.0." 2022 IEEE International Solid-State Circuits Conference, pages 350-352. https://doi.org/10.1109/ISSCC42614.2022.9731612
SambaNova. "SN50 Runs the Fastest MiniMax Speeds in the World." July 8, 2026. https://sambanova.ai/blog/sn50-runs-fastest-minimax-speeds-in-the-world
SambaNova Documentation. "API keys and URLs." Accessed July 28, 2026. https://docs.sambanova.ai/docs/en/get-started/api-keys-urls
Argonne Leadership Computing Facility. "Argonne National Laboratory deploys a new SambaNova inference-optimized cluster to support AI-driven science." November 18, 2024. https://www.alcf.anl.gov/news/argonne-national-laboratory-deploys-new-sambanova-inference-optimized-cluster-support-ai
Lawrence Livermore National Laboratory. "AI gets a boost via LLNL, SambaNova collaboration." October 19, 2020. https://www.llnl.gov/article/46836/ai-gets-boost-llnl-sambanova-collaboration
SambaNova. "Los Alamos National Laboratory expands partnership with SambaNova." November 14, 2023. https://sambanova.ai/blog/los-alamos-national-laboratory-expands-partnership-with-sambanova
Texas Advanced Computing Center. "AI Sandbox." Accessed July 28, 2026. https://cep.tacc.utexas.edu/ai-sandbox/
RIKEN Center for Computational Science. "SambaNova Systems' SambaNova DataScale Adopted." March 1, 2023. https://www.r-ccs.riken.jp/en/outreach/topics/20230301-1/
SambaNova. "SambaNova CEO explains why only one AI company wants a monopoly." July 2024. https://sambanova.ai/blog/full-stack-approach-to-ai-succeeding
SambaNova. "OTP Bank Selects SambaNova to Build Europe's Fastest AI Supercomputer." October 21, 2021. https://sambanova.ai/blog/otp-bank-selects-sambanova-systems-to-build-europes-fastest-ai-supercomputer
SambaNova and OVHcloud. "OVHcloud Selects SambaNova to Power Flagship AI Endpoints Inferencing Service." November 20, 2025. https://sambanova.ai/press/ovhcloud-selects-sambanova-to-power-flagship-ai-endpoints-inferencing-service
SambaNova and stc. "stc Partners with SambaNova to Introduce KSA's Sovereign Inferencing-as-a-Service Cloud." February 11, 2025. https://sambanova.ai/press/stc-to-introduce-ksas-inferencing-as-a-service-cloud-worlds-largest-open-source-frontier-model

Editing: SambaNova Systems