SambaNova Systems builds AI chips and inference systems for enterprises, governments, and cloud operators. These customers want to run large AI models on private or controlled infrastructure. The SambaNova Systems funding news is significant. The Palo Alto company has raised $1 billion at an $11 billion valuation in the first close of its Series F round, led by General Atlantic. The funding comes as more organizations rethink where AI inference should run. Many large enterprises do not want their most sensitive models and prompts hosted entirely in public cloud environments. Founded in 2017 by Rodrigo Liang, Kunle Olukotun, and Christopher Ré, SambaNova is now turning that demand into a major AI hardware business.
Liang told TechCrunch that more investors are expected to join within weeks in a second close. The round lands roughly 5 months after SambaNova unveiled its SN50 chip and announced a $350 million Series E in February 2026. It also comes after acquisition talks with Intel last December that valued the startup at about $1.6 billion — a reminder of how fast AI infrastructure prices can move when demand gets hot. SambaNova hasn’t shut the door on an exit, but Liang’s public line is pretty clear: the company keeps getting approached, yet its momentum points more toward “being public at some point” than cashing out early.
What does SambaNova Systems actually sell?
At the product level, SambaNova sells a full-stack AI inference setup. Customers can use SambaCloud as a managed service, or deploy SambaStack and SambaRack on-premises inside their own data centers. The company’s software lets teams choose open-source models such as Llama, DeepSeek, and Qwen. They can stand up endpoints, test them in a playground or “try it” interface, and then connect those endpoints through curl, CLI tools, or a Python SDK for production use.
The hardware piece is the real pitch. SambaNova’s systems are built around its Reconfigurable Dataflow Unit, or RDU, rather than standard GPUs. The SN40L uses a dataflow design and 3-tier memory structure — SRAM, HBM, and DRAM — so models can stay closer to active compute and switch faster with less data shuffling. That’s why the company keeps framing inference as a data-movement problem, not just a raw-compute problem.
For a customer, the “before and after” is pretty straightforward. Before SambaNova, a bank or government buyer might piece together GPUs and software tooling. Model hosting, scaling logic, and privacy controls could come from several vendors. After SambaNova, it gets one stack with model management and monitoring. Auto scaling, load balancing, and infrastructure can live in the cloud, on-prem, or in hybrid and even air-gapped environments.
And the company is pushing a pretty blunt operating argument: speed without brutal power demands. SambaNova says a base SN40 rack averages about 10kW and is air-cooled, while some newer GPU-style systems can run far hotter and require liquid cooling. That difference matters a lot for enterprises that care less about training frontier models and more about serving them reliably, cheaply, and inside existing facilities.
Who founded SambaNova Systems and why are investors betting on it?
The founding story
SambaNova was founded in 2017 by CEO Rodrigo Liang, Stanford professor Kunle Olukotun, and Stanford computer scientist Christopher Ré. The core idea was simple enough: if AI models were going to get huge, then the underlying chip and system architecture couldn’t stay generic forever. That thesis looks a lot less contrarian in 2026 than it did at launch.
Why the founders fit this market
Liang isn’t a first-time semiconductor tourist. He previously led major processor teams at Sun Microsystems and Oracle, and started his career at Hewlett-Packard. Olukotun is one of the best-known academic names in parallel computing and founded Afara Websystems, which Sun acquired in 2002. Ré brought deep machine learning and data systems expertise. He had already co-founded a company that was acquired in 2017. That’s a serious blend of chip design, systems engineering, and AI research.
Product traction and commercial signals
SambaNova’s products are no longer early demos. The company launched SN40L in September 2023, made it available in the cloud first, and then on-prem from November 2023. SN50 was unveiled in February 2026 and is scheduled to start shipping in the second half of 2026, with SoftBank lined up as its first deployment partner.
The latest commercial proof point is JPMorganChase, which selected SambaNova as an inference-infrastructure partner for secure on-prem AI inference using SN40L and SN50 systems. Liang called that win “a big deal.” He’s right. A bank that size doesn’t pick new infrastructure lightly, especially for sensitive workloads. SambaNova also names Saudi Aramco, Intel, and Japanese firms among its customers, while Liang says the company serves 3 buyer groups: sovereign clouds, neoclouds, and enterprises building internal AI capacity.
Fundraising details
The new SambaNova Systems funding round is a Series F first close of $1 billion at an $11 billion valuation, led by General Atlantic. Other named investors include Seligman Ventures, T. Rowe Price Associates, Capital Group, A&E Investment, Assam Ventures, Battery Ventures, Cambium Capital, BlackRock, Kabila Capital, QFO Capital, Qatar Investment Authority, Vista Equity Partners, Volantis, and Intel, which has backed the company since Series C. SambaNova will use the capital to scale the business and secure supply chain capacity over the next 12 months.
How SambaNova compares with Nvidia, Cerebras, and Groq
SambaNova isn’t competing head-on with Nvidia everywhere. It’s carving out the inference-heavy part of the market, especially where buyers want privacy, big-model performance, and controlled deployment. Its direct startup rivals are companies like Cerebras and Groq. They’re also selling specialized AI hardware instead of general-purpose GPUs. Cerebras closed a $1 billion Series H at about a $23 billion post-money valuation in February 2026, while Groq raised $750 million at a $6.9 billion valuation in September 2025.
The incumbent alternative is still Nvidia infrastructure, and sometimes Intel Gaudi for customers that want a more open or lower-cost route. But SambaNova’s pitch is different. It says it can fit multi-trillion-parameter models into a single rack, co-develop around Intel Xeon, and give buyers a route to premium inference without total dependence on hyperscale cloud vendors. That’s the wedge investors are backing.
Why does the SambaNova Systems funding round matter?
This round matters because SambaNova isn’t raising just to pad the balance sheet. It’s raising to lock down supply. Liang has been explicit about that, and it’s a pretty revealing detail. In AI hardware, demand doesn’t mean much if you can’t get wafers, packaging, memory, and the rest of the bill of materials on time.
It also matters because of what the customer mix says. JPMorganChase choosing on-prem inference infrastructure, Intel deepening its partnership, and SoftBank becoming SN50’s first deployment partner all point to the same thing: buyers want alternatives to a single cloud-centered AI model. They still want performance. They just want more control over where inference runs and how the stack is assembled.
There’s a valuation story here too. A company once discussed as a roughly $1.6 billion acquisition target is now priced at $11 billion in a fresh financing. That gap tells you how much investor appetite has shifted from “interesting chip startup” to “maybe this is real infrastructure.” The jump is ambitious. But it’s not random. SambaNova has live product, named enterprise customers, and a sharper market message than a lot of AI hardware startups had 2 years ago.
What is the AI inference chip market telling us?
The timing lines up with a broader shift in AI compute. Deloitte expects inference workloads to account for roughly 2-thirds of all AI compute in 2026, up from about 1-third in 2023, and it puts the market for inference-optimized chips at more than $50 billion this year. That’s the part of the market SambaNova is chasing.
The wider AI hardware market is already enormous. Grand View Research values it at $151.3 billion in 2026 and projects it to reach $691 billion by 2033, a 24.2% CAGR. Processor hardware made up 54.5% of revenue in 2025. That tells you something useful: if AI keeps moving from experiments to production systems, chip vendors don’t just ride the trend — they become the trend.
And there’s a second tailwind. Sovereign AI and private infrastructure are gaining traction because governments, banks, and large enterprises increasingly care about data residency, latency, and security. SambaNova keeps talking about sovereign clouds for a reason. It’s not a niche talking point anymore. It’s becoming an actual procurement category.
What to watch after SambaNova Systems funding
The next thing to watch is execution. SambaNova Systems funding gives the company a lot more room, but AI hardware is brutal if supply slips or promised deployments drag. SN50 customer shipments in the second half of 2026, the JPMorgan rollout, and whether the second close of Series F expands the investor roster will tell us a lot more than the valuation headline does.
If SambaNova can turn private inference into a repeatable enterprise sale, this round will look smart. If not, it’ll just look expensive.
Read how Econovus Packaging raised ₹40 crore in a pre-Series A round led by Rainmatter, with participation from Rockstud Capital, to build engineered industrial packaging solutions that make transporting batteries, automotive parts, and export cargo safer, lighter, and more sustainable.
FAQ
- What is the latest SambaNova funding round? SambaNova has raised $1 billion in the first close of its Series F round at an $11 billion valuation. General Atlantic led the financing, and Rodrigo Liang said more investors are expected to join in a second close within weeks.
- How does SambaNova Systems work for customers? SambaNova gives customers a full inference stack that can run in the cloud or on-premises. A team can pick an open-source model and deploy an endpoint. It can test it in a playground, connect it through APIs or SDKs, and then manage scaling and monitoring on top of SambaNova’s RDU-based hardware.
- Who founded SambaNova Systems? SambaNova was founded in 2017 by Rodrigo Liang, Kunle Olukotun, and Christopher Ré. Liang came from senior processor work at Sun and Oracle, Olukotun previously founded Afara Websystems, and Ré built his reputation in Stanford’s AI and data systems world.
- Is SambaNova a cloud company or an AI chip company? It’s both, but the chip story is the foundation. SambaNova designs its own AI hardware for inference and packages it with cloud and on-prem software so enterprises, banks, and sovereign buyers can run large models without relying entirely on public cloud infrastructure.
