Designs its own reconfigurable dataflow unit (RDU) AI chips and sells them as on-premises inference racks, dedicated hosted systems and a per-token cloud API, pitching banks, governments and cloud providers on running large open models on infrastructure they control.
What is SambaNova, and which banks use it?
SambaNova, founded in 2017 and based in San Jose, California, designs RDU AI chips and sells them as on-premises inference systems and through SambaCloud, a per-token API for open models. It is private: a July 2026 Series F first close raised $1 billion at an $11 billion valuation, led by General Atlantic. JPMorgan Chase is its one named bank customer on our pages, selecting SN40 and SN50 systems for on-premises AI inference in July 2026. SambaCloud prices are published; on-premises system pricing is not.
Category
Founded
2017
Headquarters
San Jose, California, US
Ownership
Private; a July 2026 Series F first close raised $1 billion at an $11 billion post-money valuation, led by General Atlantic with Seligman Ventures and T. Rowe Price, alongside investors including Intel Capital, BlackRock-managed funds, Vista Equity Partners and the Qatar Investment Authority.[1]
Deployment
On-premises systems (SambaRack, SambaStack), dedicated hosting in the cloud, or API (cloud) through SambaCloud
Pricing
Per-token API pricing for SambaCloud; on-premises systems (SambaRack, SambaStack) through sales-led contracts
Banks among the 120
1
- JPMorgan Chase selected SambaNova as an inference-infrastructure partner in July 2026, deploying SN40 and SN50 systems for on-premises AI inference; the bank's infrastructure-platforms CIO said it will test their speed and security, and CEO Rodrigo Liang told TechCrunch the deal tells banks it is 'time not to completely depend on cloud services'.
- Funding: a $676 million Series D led by SoftBank Vision Fund 2 in April 2021 (valuation above $5 billion), more than $350 million in a Series E in February 2026, and a $1 billion Series F first close at an $11 billion post-money valuation in July 2026, led by General Atlantic.
- Intel ties: Intel CEO Lip-Bu Tan has chaired SambaNova since 2017; Intel invested from 2019, joined the 2026 rounds and signed a multiyear collaboration in February 2026, after talks to buy SambaNova for about $1.6 billion reportedly fell apart (Bloomberg, as reported by CNBC).
- Chips: the SN40L launched in September 2023; the SN50, unveiled in February 2026, was due to ship to customers in the second half of 2026 with SoftBank as its first deployment partner.
- Other named customers include Saudi Aramco, Intel and Japanese firms (TechCrunch), Hugging Face, Meta and major AI labs (CNBC) and OVHcloud, whose AI Endpoints service runs on SambaStack (November 2025).
- Speed, efficiency and cost-of-ownership claims, such as the SN50's five times more compute per accelerator than the previous generation, are vendor-reported.
- SambaCloud per-token prices are published; pricing for on-premises SambaRack and SambaStack systems is not published.
What does SambaNova sell to banks?
| Product | What it does | Status |
|---|---|---|
| SambaStack[16][8] | Full-stack inference hardware and software deployable on-premises or with dedicated cloud hosting; runs pre-configured model bundles on SambaRack systems. | Generally available |
| SambaRack SN50[17][5] | Air-cooled rack of 16 SN50 RDU chips for agentic inference, scaling to 256 accelerators; SambaNova said in February 2026 the SN50 would ship later that year. | Announced |
| SN40L systems[11][2][1] | Systems built on the SN40L chip launched in September 2023, available in the cloud and on-premises from November 2023; the SN40 systems JPMorgan Chase is deploying. | Generally available |
| SambaCloud[18][10] | Per-token inference API for open models such as DeepSeek, Llama, Qwen and gpt-oss, launched in September 2024 as SambaNova Cloud, with OpenAI-compatible endpoints. | Generally available |
How much does SambaNova cost?
How it is charged: Per-token API pricing for SambaCloud; on-premises systems (SambaRack, SambaStack) through sales-led contracts.
| Product | Plan | List price | Unit | Checked |
|---|---|---|---|---|
| SambaCloud | gpt-oss-120b[19] | $0.22 input / $0.59 output | per 1M tokens | Oct 26, 2026 |
| SambaCloud | Meta-Llama-3.3-70B-Instruct[19] | $0.60 input / $1.20 output | per 1M tokens | Oct 26, 2026 |
| SambaCloud | DeepSeek-V3.2[19] | $3.00 input / $4.50 output | per 1M tokens | Oct 26, 2026 |
SambaCloud per-token prices are published on SambaNova's pricing page. Pricing for on-premises SambaRack and SambaStack systems, the kind JPMorgan Chase is deploying, is not published, and the value of the JPMorgan Chase agreement was not disclosed.
Which banks use SambaNova?
| Bank | What | Status |
|---|---|---|
| JPMorgan Chase | Inference-infrastructure partner: SN40 and SN50 systems (RDU architecture) for on-premises AI inference[1][3][2] | Selected |
Among the 120 largest US banks on this site, from sources that name the bank: the bank, SambaNova or the press. Status is the furthest stage a source confirms.
What is SambaStack?
SambaNova's full hardware-and-software inference stack, deployable on-premises or with dedicated hosting in the cloud and powered by SambaRack systems built on its RDU chips; it runs open models such as DeepSeek, Llama and gpt-oss-120b and can hot-swap pre-configured model bundles at inference time.[16][17]
What has SambaNova done, and when?
- Jul 8, 2026
- Jun 12, 2026
New CFO and software chief[7]
Matt Padfield joins as chief financial officer and Rich Heaton, formerly of AWS's Neuron SDK group, as executive vice president of software.
- Apr 8, 2026
Heterogeneous inference blueprint with Intel[6]
GPUs for prefill, SambaNova RDUs for decode and Intel Xeon 6 CPUs for agentic tools, due for enterprises and cloud providers in the second half of 2026; SambaNova standardises on Xeon 6 as its host CPU.
- Feb 24, 2026
- Nov 20, 2025
OVHcloud selects SambaNova[8]
OVHcloud's AI Endpoints inference service is to run on SambaStack, with a 99.8% uptime SLA.
- Jul 8, 2025
SambaManaged for data centers[9]
An inference data-center product SambaNova says can be deployed in 90 days and run on 10 kW of air-cooled power.
- Sep 10, 2024
SambaNova Cloud launches[10]
A free-to-start inference API on SN40L chips; SambaNova reports Llama 3.1 405B at 132 tokens a second, as benchmarked by Artificial Analysis.
- Sep 20, 2023
SN40L chip unveiled[11]
The SN40L addresses both high-bandwidth memory and DRAM and, SambaNova says, can serve a 5-trillion-parameter model on a single system node.
- Dec 21, 2022
GPT Banking offered as a service[12]
SambaNova says it launched Dataflow-as-a-Service GPT Banking in 2022, a large language foundation model for banks sold as a subscription.
- Apr 13, 2021
$676 million Series D[13]
Led by SoftBank Vision Fund 2, with Temasek and GIC; total funding passes $1 billion and the valuation exceeds $5 billion.
- Jan 1, 2017
Founded[14]
SambaNova starts operations in Palo Alto, California, founded by Stanford professors Kunle Olukotun and Christopher Ré with semiconductor executive Rodrigo Liang (exact date not disclosed).
What do banks use SambaNova for?
| Use | What it does | Status | Use case |
|---|---|---|---|
| On-premises inference for sensitive workloads[3][1] | JPMorgan Chase is deploying SN40 and SN50 systems so production AI inference stays behind the bank's own walls, with full control over the data and an auditable trail. | Announced | Data & privacy |
| Inference capacity for enterprise and agentic AI workloads[3][5] | JPMorgan Chase plans to test the RDU systems' speed and security on its enterprise AI workloads; SambaNova positions the SN50 for agentic inference. | Announced | Generative & agentic AI |
| Diversifying away from GPU and cloud-only inference[2] | Liang says banks of JPMorgan's calibre are building private, secure inference infrastructure rather than depending completely on cloud services, and want heterogeneous hardware. | Announced | Third-party & vendor AI |
What has SambaNova disclosed in numbers?
| Metric | Value | As of |
|---|---|---|
| Series F first close | $1 billion at an $11 billion post-money valuation[1][2] | Jul 8, 2026 |
| Series E | more than $350 million[5] | Feb 24, 2026 |
| Series D | $676 million; valuation above $5 billion; total funding above $1 billion[13] | Apr 13, 2021 |
| SN50 versus the previous generation (vendor-reported) | 5x compute per accelerator, 4x network bandwidth, up to 256 linked accelerators[5] | Feb 24, 2026 |
| SambaNova Cloud speed on Llama 3.1 405B (vendor-reported, Artificial Analysis benchmark) | 132 output tokens per second[10] | Sep 10, 2024 |
“At JPMorganChase, AI infrastructure has to meet a very high bar for performance, control and reliability.”[1][3]
Darrin Alves · CIO, Infrastructure Platforms, JPMorganChase · Jul 8, 2026
“It sends a message to the banking industry that it's time not to completely depend on cloud services.”[2]
Rodrigo Liang · Co-founder and CEO, SambaNova · Jul 8, 2026
Who runs SambaNova?
Which rules apply to a bank using SambaNova?
| Authority | Why it matters here | Documents |
|---|---|---|
| OCC | Buying inference racks and software support from a venture-backed chip startup is a third-party relationship under the 2023 interagency guidance: banks are expected to assess the supplier's financial condition, support continuity and exit options for hardware they run on-premises. | OCC Bulletin 2023-17 FDIC FIL-29-2023 |
| Federal Reserve | SR 26-2 leaves generative and agentic AI outside model risk management and points banks to broader governance, which covers the open models a bank serves on SambaNova systems; the Fed's third-party guidance applies to the hardware and software supplier. | SR 26-2 SR 23-4 |
| NIST | NIST's Generative AI Profile lists information-security and value-chain risks for third-party AI components, relevant when a bank runs open-weight models on a vendor's inference stack inside its own data centers. | NIST AI 600-1 (Generative AI Profile) |
Linked authorities have a page in the Regulation tracker; the others are named for completeness.
What should a bank note before buying SambaNova?
Analysis by BankingNewsAI from the sources cited on this page. Observations, not advice or a recommendation.
01
One named bank customer, at the selection stage
JPMorgan Chase selected SambaNova in July 2026 and its infrastructure CIO spoke of testing the systems' speed and security; no production volumes or results have been published.
02
Two ways to buy, priced differently
SambaCloud publishes per-token prices for open models; the on-premises SambaRack and SambaStack systems a bank would run in its own data center are priced through contracts that are not published.
03
Performance figures are vendor-reported
Speed, compute and cost-of-ownership comparisons come from SambaNova's own announcements, some citing third-party benchmarks such as Artificial Analysis.
04
Close ties to Intel
Intel's CEO chairs SambaNova, Intel invests in it and co-develops and co-sells products with it, and Intel reportedly discussed buying it; a buyer assessing supplier concentration and continuity will weigh that relationship.
05
The newest chip is recent
The SN50 was unveiled in February 2026 with shipments planned for the second half of 2026; JPMorgan Chase's deployment pairs it with the older SN40-generation systems.
What are the alternatives to SambaNova?
Amazon Bedrock (AWS)
AWS service giving one API to foundation models from Amazon, Anthropic, OpenAI, Meta and others, plus AgentCore for running agents.
How much does SambaNova cost?
SambaCloud, SambaNova's inference API, publishes per-token prices by model; on October 26, 2026 gpt-oss-120b listed at $0.22 per million input tokens and $0.59 per million output tokens. Pricing for on-premises SambaRack and SambaStack systems, the kind JPMorgan Chase is deploying, is not published, and the value of the JPMorgan Chase agreement was not disclosed.
Which banks use SambaNova?
Among the banks tracked on this site, JPMorgan Chase is the named customer: in July 2026 it selected SambaNova as an inference-infrastructure partner and is deploying SN40 and SN50 systems for on-premises AI inference. SambaNova also names Saudi Aramco, Intel, OVHcloud, SoftBank and Japanese firms among its customers.
Who owns SambaNova?
SambaNova is privately held. Its July 2026 Series F first close raised $1 billion at an $11 billion post-money valuation, led by General Atlantic with Seligman Ventures and T. Rowe Price; other investors include Intel Capital, BlackRock-managed funds, Vista Equity Partners and the Qatar Investment Authority. Intel CEO Lip-Bu Tan is its chairman.
What is a SambaNova RDU?
The reconfigurable dataflow unit is SambaNova's own AI chip design, which uses a dataflow architecture to minimise data movement. The SN40L (2023) and SN50 (unveiled 2026) are RDU chips; SambaNova packages them in SambaRack systems, runs them under its SambaStack software and also serves models from them through SambaCloud.
Can a bank run SambaNova on-premises?
Yes. SambaStack can be deployed on-premises or with dedicated hosting in the cloud, and JPMorgan Chase chose SambaNova's SN40 and SN50 systems specifically to keep production AI inference behind its own walls, with full control over the data and an auditable trail. SambaCloud is the hosted, per-token alternative.
Is SambaNova an alternative to Nvidia GPUs?
SambaNova positions its RDU systems as an alternative to GPU-only inference and compares the SN50 with the GPUs in Nvidia's B200 systems; those comparisons are SambaNova's own. Its 2026 blueprint with Intel pairs GPUs for the prefill stage with SambaNova RDUs for decode rather than replacing GPUs outright.
- SambaNova Completes First Close of $1B Financing at $11B Valuation
- AI chip maker SambaNova raises $1B at $11B valuation, 5 months after last mega round
- JPMorgan Chase picks SambaNova for on-prem AI inference
- Intel partners with AI chip startup SambaNova after acquisition talks reportedly failed
- SambaNova Unveils Fastest Chip for Agentic AI, Collaborates with Intel, and Raises $350M+
- SambaNova and Intel Announce Blueprint for Heterogeneous Inference: GPUs For Prefill, SambaNova RDUs for Decode, and Intel Xeon 6 CPUs for Agentic Tools
- SambaNova Names Engineering and Finance Leaders to Accelerate Growth
- OVHcloud Selects SambaNova to Power Flagship AI Endpoints Inferencing Service
- SambaNova Launches First Turnkey AI Inference Solution for Data Centers, Deployable in 90 Days
- SambaNova Launches The World's Fastest AI Platform
- SambaNova Unveils New AI Chip, the SN40L, Powering its Full Stack AI Platform
- SambaNova Accelerates AI for Banking and the Public Sector with a World Record Performance System, New Partnerships, and International Expansion in 2022
- SambaNova Systems Becomes World's Best-Funded AI Startup
- About Us | SambaNova (undated page, retrieved 2026-10-26)
- Meet Our Talented AI Team | SambaNova (undated page, retrieved 2026-10-26)
- SambaStack | Full-Stack Enterprise AI Platform (undated page, retrieved 2026-10-26)
- SambaRack | Purpose-Built AI Rack for Model Deployment (undated page, retrieved 2026-10-26)
- SambaCloud | Full-Stack AI Platform for Large Open-Source Models (undated page, retrieved 2026-10-26)
- SambaCloud Pricing (undated page, retrieved 2026-10-26)
When a bank signs with SambaNova, you'll read it here first.
when a bank picks an AI vendor, or a vendor ships for banks, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning