The order form is a simple checklist: a pre-configured rack, a renewable power purchase agreement, a sovereign data residency guarantee. The product is not a software license or a cloud credit, but a physical object, a SambaRack, that will be installed in a specific geographic location, on land controlled by the customer, powered by a grid they can point to on a map. This is the new unit of sale for enterprise AI, and SambaNova Systems is its most committed architect.
Founded in 2017, the Palo Alto-based company has raised over $1.8 billion to build a full-stack alternative to Nvidia’s dominance, combining its own AI chips, systems, and software into a unified platform called the SambaNova Suite [Sacra]. For years, the pitch was about raw performance and efficiency for large language models. Today, as CEO Rodrigo Liang frames it, the next AI war is about inference costs, compute shortages, and scaling infrastructure profitably [Bloomberg, 2026].
The wedge is a shipping container
The company’s fundamental bet is that a significant slice of the enterprise and government market cannot, or will not, run its most sensitive AI workloads on rented, general-purpose GPU clusters in a public cloud. The wedge is vertical integration: from the custom SN40L inference chip up through the system software and managed service layer, all designed to be deployed as a turnkey appliance within a customer’s own data center or a partnered private facility [Sacra][LinkedIn, 2026].
Revenue flows from hardware sales, professional services for model tuning (an estimated 25-33% of new engagements), and subscription access to foundation models and compute [Sacra]. The flagship offering, the SambaNova Suite, is often sold in multi-rack configurations. Accenture, for instance, deployed a three-rack installation at its own data center for generative AI work in 2023 [Data Center Dynamics, 2026].
Founders, Backgrounds, and Bench
The trio of co-founders brings together deep systems architecture, academic research, and applied AI.
| Founder | Role | Key Background |
|---|---|---|
| Rodrigo Liang | Co-founder & CEO | Former SVP of Engineering at Oracle |
| Kunle Olukotun | Co-founder & Chief Technologist | Stanford professor; "father of the multi-core processor" |
| Christopher Ré | Co-founder | Stanford professor; co-founded Lattice.io |
This blend of commercial systems-building experience and foundational academic research has convinced investors like SoftBank, BlackRock, and Intel Capital to write nine-figure checks, culminating in a $5.1 billion valuation in 2021 and a further $350 million round led by Vista Equity Partners in early 2026 [Sacra][Reuters, Feb 2026].
The new frontier: sovereignty and watts
Recent moves reveal a strategy evolving beyond pure performance benchmarks. The company is layering its hardware stack with two powerful external narratives: data sovereignty and sustainable power. Its most illustrative deal is a partnership with Argyll Data Development to build the UK’s first renewable-powered sovereign AI cloud on Scotland’s Cowal Peninsula [SambaNova, 2026]. The project, dubbed the Killellan AI Growth Zone, plans to use SambaNova’s racks powered by local renewable energy to serve UK defense, healthcare, and finance sectors with strict residency requirements [Business Wire, 2025].
This aligns with Liang’s repeated commentary that power and energy will be the ultimate bottleneck for scaling AI [Bloomberg, 2024]. By offering a more efficient, purpose-built stack, SambaNova claims dramatic advantages, stating its inference of large models can be up to four times faster on a single rack than competitors require on forty [LinkedIn, 2026].
Navigating the Goliath in the room
The risks surrounding SambaNova are as substantial as its funding rounds. The company operates in the shadow of Nvidia, which has defined the market and continues to innovate at a blistering pace.
- The capital furnace. Designing and fabricating cutting-edge AI chips is astronomically expensive. The company has consumed nearly $2 billion in venture funding to reach this point [Sacra].
- The ecosystem gap. Nvidia’s CUDA platform is the entrenched standard for AI developers. SambaNova must convince enterprises to bet on a new, proprietary stack.
- Execution pressure. A 15% workforce reduction in 2025 underscores the operational pressures of navigating this market [EE Times, 2026].
The company’s answer to these pressures appears to be focus and partnership. The Intel collaboration, announced alongside the 2026 fundraise, is a key plank, potentially giving SambaNova’s technology a route to market through Intel’s vast enterprise sales channel [Reuters, Feb 2026].
The next twelve months
The immediate future for SambaNova will be measured in rack deployments and sovereign cloud launches. The success of the Argyll partnership in Scotland will be a critical proof point for the renewable-sovereign model. Internally, the focus will be on scaling the inference and cloud services business following its strategic refocus and new funding. The partnership with Meta to "unleash the power of Llama 4 models" is a direct attempt to embed its hardware at the foundation of the next wave of open-source AI [LinkedIn, 2026].
The cultural question SambaNova is implicitly answering is one of place and trust. In an era where AI is both miraculous and unnervingly opaque, its proposition is a return to the tangible. It sells a box you can touch, in a location you can visit, powered by a source you can name. The bet is that for the most important tasks, enterprises and nations will want their intelligence to have a concrete address.