AI Magazine highlights SambaNova Systems' fast AI inference chips at a US$11 billion value.
In early 2026, takeover talks between SambaNova Systems and Intel had stalled. Six months later, SambaNova emerged with an US$11bn valuation, backed by Intel as a strategic partner and investor rather than an owner.
This turnaround reflects where the AI chip market is heading. Training giant models made NVIDIA the world's most valuable chipmaker, but running those models for millions of users presents a different challenge. Speed and cost per token now determine which companies win. SambaNova has therefore pivoted towards the inference market, and its new SN50 chip is pitched squarely at the agentic AI era.
Rodrigo Liang, Co-Founder and CEO at SambaNova Systems, says: "SambaNova's US$11bn valuation highlights the central role that fast inference now plays in the enterprise AI stack."
**What is SambaNova's technology ecosystem?**
Founded in 2017 and based in San Jose, California, SambaNova designs its own processors, which it calls Reconfigurable Dataflow Units (RDUs). Rather than selling chips in isolation, the company packages them into full systems and services. These include SambaCloud for developers, SambaStack and SambaManaged for operators who want to run their own inference clouds, and SambaRack hardware for on-premises deployment.
The SN50, unveiled in February 2026, is the centrepiece of this ecosystem. SambaNova claims it runs up to five times faster than competing chips at up to three times lower total cost of ownership (TCO). The chip links up to 256 accelerators and supports models with more than 10 trillion parameters. These performance figures have helped the company win serious customers across several sectors.
**Enterprise clients deploying SambaNova hardware**
JPMorganChase has selected SambaNova as an inference infrastructure partner and is deploying SN40 and SN50 systems on premises. SoftBank Corp. will be the first company to deploy the SN50 in its next-generation data centres in Japan. In addition, TEPCO Systems will distribute the technology across Japan's power sector. In Europe, OVHcloud and Infercom have built inference services on SambaNova hardware.
The relationship with Intel has evolved into a strategic partnership rather than an acquisition. A multi-year collaboration covers an Intel Xeon-based AI cloud and joint selling initiatives across global markets. It also includes a blueprint for heterogeneous inference that uses GPUs for prefill, SambaNova chips for decode, and Xeon 6 processors for agentic tools. Intel Chief Executive Lip-Bu Tan also chairs SambaNova's board.
**SambaNova Systems in numbers**
- **Valuation:** US$11bn after the first close of a US$1bn Series F led by General Atlantic in July 2026
- **Series E:** More than US$350m in February 2026, led by Vista Equity Partners and Cambium Capital
- **SN50 chip:** Claimed up to 5x faster than rival chips at up to 3x lower TCO (February 2026)
- **Scale:** SN50 links up to 256 accelerators and supports models of 10 trillion-plus parameters (2026)
- **Sales:** Beat its sales plan for the fiscal year ending January 2026
- **Intel stake:** About 9% after a further US$15m investment planned in April 2026
- **Founded:** 2017, headquartered in San Jose, California
**Backing SambaNova's growth**
Investors have followed this market momentum. A Series E funding round of more than US$350m, led by Vista Equity Partners and Cambium Capital, closed in February 2026. General Atlantic led the first US$1bn close of a Series F in July 2026. The company also beat its sales plan for the fiscal year ending January 2026.
Its European wins are central to this growth strategy. OVHcloud, which runs more than 500,000 servers across 46 data centres, chose SambaNova for its AI Endpoints service. Meanwhile, Infercom keeps all data and model execution inside EU borders for customers concerned about data sovereignty.
Darrin Alves, CIO of Infrastructure Platforms at JPMorganChase, says: "At JPMorganChase, AI infrastructure has to meet a very high bar for performance, control and reliability."
SambaNova has gone from a possible acquisition target to one of the best-funded challengers in AI hardware in less than a year. Its bet rests on a simple premise: as AI moves from training to everyday use, the market will reward the fastest and cheapest way to serve tokens.
The next 12 months will show whether the SN50 lives up to its performance claims at scale. Production proof is set to arrive soon, with implementations for JPMorganChase, SoftBank and Intel's heterogeneous blueprint all due to go live.
**Top technology partners**
- **Intel:** A multi-year collaboration from February 2026 covering an Intel Xeon-based AI cloud, joint selling, and a heterogeneous inference blueprint that pairs GPUs for prefill, SambaNova RDUs for decode, and Xeon 6 CPUs for agentic tools.
- **SoftBank Corp.:** Hosts SambaCloud for developers in Asia Pacific and will be the first to deploy the SN50 in its next-generation AI data centres in Japan, delivering low-latency inference for sovereign and enterprise customers across the region.
- **TEPCO Systems:** The digital arm of Tokyo Electric Power will distribute SambaNova's AI infrastructure across Japan under an April 2026 agreement and use it as the base of the TEPCO Group's next AI platform.
- **OVHcloud:** The European cloud provider chose SambaNova's SambaStack platform as the core of its AI Endpoints inference service, launched from France with a 99.8% uptime SLA. Further European deployments are planned under a November 2025 agreement.
- **Infercom:** The Luxembourg-based provider uses SambaManaged to run what SambaNova calls Europe's first sovereign inference-as-a-service platform. It keeps data storage, processing and model execution inside the EU to support GDPR and AI Act compliance.