SambaNova Systems

PUBLIC

Company Name	SambaNova Systems
Tagline	Full-stack enterprise AI infrastructure company building custom AI chips, systems, and software for LLMs.
Headquarters	Palo Alto, California
Founded	2017
Stage	Series D+
Business Model	Hardware + Software
Industry	Deeptech
Technology	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (3+)
Funding Label	$100M+
Total Disclosed Funding	~$1,840,000,000

Links

PUBLIC

• Website: https://sambanova.ai/
• LinkedIn: https://www.linkedin.com/company/sambanova

Executive Summary

PUBLIC

SambaNova Systems is building a vertically integrated alternative to Nvidia's general-purpose GPU stack, betting that enterprises will pay a premium for a hardware-software platform designed from the ground up for large language models [Sacra]. The company's full-stack approach, which pairs custom AI chips with proprietary software for training and inference, aims to address the compute shortages and escalating energy costs Rodrigo Liang, the CEO, identifies as the next major bottlenecks in AI scaling [Bloomberg, 2024][Bloomberg, 2026]. Founded in 2017 by a trio of Stanford professors and an Oracle engineering veteran, SambaNova has secured over $1.8 billion in disclosed funding, including a $676 million Series D at a $5.1 billion valuation in 2021 and a $350 million round led by Vista Equity Partners in February 2026 [Sacra][Reuters, Feb 2026]. Its business model blends capital-intensive hardware sales with recurring subscription revenue from its Dataflow-as-a-Service and cloud offerings, targeting government agencies and large enterprises with sovereignty or performance requirements that public clouds cannot meet [Sacra]. The next 12-18 months will test whether the company's recent strategic pivot toward inference and cloud services, underscored by a workforce reduction in 2025 and a new multi-year collaboration with Intel, can translate its technical differentiation into sustainable commercial momentum against an entrenched incumbent [EE Times, 2026][Intel].

Data Accuracy: GREEN -- Core claims corroborated by multiple independent sources including Sacra, Reuters, and Bloomberg.

Taxonomy Snapshot

Axis	Classification
Stage	Series D+
Business Model	Hardware + Software
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (3+)
Funding	$100M+ (total disclosed ~$1,840,000,000)

Company Overview

PUBLIC

SambaNova Systems was founded in 2017 in Palo Alto, California, by a trio of technologists with deep roots in high-performance computing and academia [Sacra]. The founding team includes Rodrigo Liang, a former SVP of Engineering at Oracle responsible for SPARC processors, Kunle Olukotun, a Stanford professor known for pioneering multi-core processor designs, and Christopher Ré, a Stanford professor and MacArthur Fellow who previously co-founded Lattice.io [Sacra]. The company's formation coincided with the early stages of the current AI hardware wave, positioning it to build a full-stack alternative to general-purpose GPU infrastructure from the ground up.

Key operational milestones follow a path of significant capital raises and strategic product launches. The company's first publicly disclosed major funding was a $56 million Series A round in 2018, led by GV [SambaNova, 2018][TechCrunch, 2018]. This was followed by a $676 million Series D round in April 2021, led by SoftBank Vision Fund 2, which valued the company at $5.1 billion post-money [Sacra]. A strategic pivot and workforce adjustment occurred in April 2025, when the company laid off 77 employees (approximately 15% of its workforce) to refocus its efforts on inference, fine-tuning, and cloud services [EE Times, 2026]. The most recent capital event was a $350 million fundraise in February 2026, led by Vista Equity Partners with participation from Intel, which followed reported, ultimately unsuccessful acquisition talks with Intel [Reuters, Feb 2026][Sacra].

Data Accuracy: GREEN -- Confirmed by multiple public sources including Sacra, Reuters, and company press releases.

Product and Technology

MIXED SambaNova Systems builds what it calls a full-stack enterprise AI infrastructure platform, a vertically integrated suite of custom hardware and software designed to run and fine-tune large language models. The company's public positioning centers on its flagship SambaNova Suite, which combines proprietary AI chips, rack-scale systems, and a software layer to enable organizations to deploy generative AI within their own data centers or private clouds [Sacra]. This approach is a direct alternative to assembling a stack from general-purpose GPUs and disparate software tools.

The hardware foundation is built around SambaNova's custom-designed Reconfigurable Dataflow Unit (RDU) chips. These are packaged into rack-level systems, previously branded as DataScale and now often referred to as SambaRack, which are optimized for the parallel processing demands of LLMs [Sacra][SambaNova, 2026]. The software component includes tools for model training, fine-tuning, and, increasingly, high-performance inference. A key public claim is performance efficiency: the company states its inference of a 405-billion parameter Llama model reaches up to 200 tokens per second at full precision on a single rack [LinkedIn, 2026].

SambaNova go-to-market includes multiple consumption models. [PUBLIC] Customers can purchase hardware systems outright. [PUBLIC] They can also access the platform via a subscription-based cloud service, Dataflow-as-a-Service, which provides access to pre-trained models and compute [Sacra]. The company offers a $5 free credit for developers to test its cloud platform [SambaNova]. Recent public announcements highlight a strategic pivot towards inference and partnerships, including a multi-year collaboration with Intel to deliver inference solutions on Xeon-based infrastructure and a project with Argyll to build a renewable-powered sovereign AI cloud in the UK [Intel][SambaNova, 2026].

PUBLIC The market for enterprise AI infrastructure is defined by a tension between the explosive demand for large language model capabilities and the practical constraints of cost, control, and compute availability. SambaNova positions itself at the intersection of these forces, targeting enterprises and governments for whom the public cloud is insufficient due to sovereignty, data privacy, or performance requirements.

Quantifying the total addressable market for custom AI hardware and full-stack systems is complex, as it spans several overlapping sectors. Public analyst reports on the broader AI chip market provide an analogous view. For instance, the market for AI accelerators, which includes GPUs and custom ASICs, is projected to grow from approximately $30 billion in 2023 to over $100 billion by 2028, according to a 2024 report from Gartner [Gartner, 2024]. The segment for on-premises AI infrastructure, which is SambaNova's primary delivery model, represents a significant portion of this spend, driven by sectors like defense, finance, and healthcare.

Key demand drivers extend beyond simple compute needs. CEO Rodrigo Liang has framed the next phase of competition as a war on inference costs and energy consumption [Bloomberg, 2026]. This aligns with broader industry tailwinds: the shift from model training to widespread inference workloads, the rising strategic importance of sovereign AI capabilities for national governments, and the search for architectural efficiency beyond general-purpose GPUs. These drivers create a wedge for vertically integrated solutions that promise better performance-per-watt and total cost of ownership, though they require customers to adopt a proprietary stack.

Adjacent and substitute markets exert significant pressure. The most direct substitute is the dominant Nvidia GPU ecosystem, paired with cloud hyperscalers offering managed AI services. This commoditized path offers flexibility and a vast developer ecosystem. Another adjacent market is the burgeoning field of AI software platforms and middleware, which aim to abstract hardware complexity entirely. For SambaNova, success depends on convincing customers that the performance and control benefits of its integrated stack outweigh the lock-in and the competitive might of these established alternatives.

Regulatory and macro forces are increasingly favorable to SambaNova's on-premise and sovereign cloud narrative. Data residency laws in regions like the European Union, combined with national security concerns around AI development, are pushing governments and regulated enterprises to build internal capacity. The company's partnership with Argyll to create a renewable-powered sovereign AI cloud in the UK is a direct play on this trend [Business Wire, 2025]. However, these projects are often long-cycle and subject to political shifts, introducing a different kind of execution risk compared to commercial sales.

AI Accelerator Market 2023 | 30 | $B
AI Accelerator Market 2028 | 100 | $B

The projected near-tripling of the AI accelerator market underscores the sheer capital flowing into the sector, but it also highlights the intensity of competition for that spend. SambaNova's bet is that a meaningful slice of this growth will prioritize integrated efficiency over flexible generality.

Data Accuracy: YELLOW -- Market sizing is drawn from analogous, high-level third-party reports; specific TAM for SambaNova's niche is not publicly broken out.

Competitive Landscape

MIXED SambaNova Systems enters the market as a vertically integrated challenger, offering a proprietary AI chip and software stack as a full-stack alternative to assembling components from the dominant GPU ecosystem.

Company	Positioning	Stage / Funding	Notable Differentiator	Source
SambaNova Systems	Full-stack enterprise AI infrastructure (custom chips, systems, software) for on-prem/private cloud LLM workloads.	Series D+; ~$1.84B total disclosed [Sacra][Reuters, Feb 2026]	Proprietary Reconfigurable Dataflow Unit (RDU) architecture and integrated software stack for LLM training and inference.	[Sacra]
Nvidia	Dominant supplier of general-purpose GPU hardware (H100, Blackwell) and CUDA software ecosystem for AI compute.	Public company; market cap >$2 trillion.	Vast, entrenched CUDA developer ecosystem and software moat, with industry-leading performance per watt on flagship chips.	[PUBLIC]
Cerebras	Developer of wafer-scale AI chips (WSE-3) and systems for large-scale model training and inference.	Private; >$720M raised [Crunchbase].	Wafer-scale engine architecture designed to avoid inter-chip communication bottlenecks for massive models.	[PUBLIC]

The competitive map in enterprise AI infrastructure is stratified by technical approach and go-to-market focus. At the hardware layer, Nvidia’s dominance is defined by its CUDA software platform as much as its silicon, creating a high switching cost for developers. Challengers like SambaNova and Cerebras pursue architectural differentiation,dataflow and wafer-scale, respectively,to claim performance or efficiency advantages for specific workloads, primarily large language model training and inference. An adjacent layer of competition comes from cloud hyperscalers (AWS, Google, Microsoft) developing their own custom silicon (e.g., Trainium, TPU) and offering it as a managed service, which competes with SambaNova’s cloud subscription offerings. Finally, a cohort of software-focused AI infrastructure companies provides orchestration and optimization layers atop commodity hardware, addressing a different part of the customer’s problem around deployment and management.

SambaNova’s defensible edge today rests on its full-stack integration and its focus on sovereign, on-premises deployments. The company’s custom RDU architecture, co-designed with its software compiler, is engineered for the sparse, dataflow-intensive computations common in transformers, a claim supported by performance benchmarks for models like Llama 405b and DeepSeek [LinkedIn, 2026]. This integration allows SambaNova to offer performance guarantees and predictable total cost of ownership, a value proposition targeted at government labs and regulated enterprises where data sovereignty and control are non-negotiable. The durability of this edge is tied to continued architectural performance leadership and the ability to keep its software stack tightly optimized for its hardware, a task that becomes more complex as model architectures evolve.

The company’s primary exposure is to the sheer scale of Nvidia’s ecosystem and the rapid innovation cycle of its competitors. Nvidia’s annual architectural updates and deep software investments raise the performance bar continuously, while its vast sales channel and brand recognition make it the default, low-risk choice for many enterprises. SambaNova’s go-to-market is also inherently more complex and capital-intensive, requiring direct sales of high-value hardware systems and professional services, which limits its scale and speed compared to pure software or cloud service models. Furthermore, the February 2026 strategic collaboration with Intel, while a validation of its inference technology, also signals a potential dependency or a concession that its own hardware may not address the entire market alone [Reuters, Feb 2026].

The most plausible 18-month scenario hinges on the adoption of specialized inference at scale. If enterprise demand for running massive, proprietary LLMs shifts decisively toward on-premises infrastructure due to cost, latency, or sovereignty concerns, SambaNova’s integrated stack could see accelerated adoption in its core government and financial services verticals. In this case, the winner would be SambaNova, securing anchor deployments like the renewable-powered sovereign AI cloud with Argyll in the UK as a blueprint [Business Wire, 2025]. The loser in this scenario would be generic cloud GPU instances that cannot match the total cost or performance predictability of a purpose-built system. Conversely, if Nvidia’s next-generation architectures close the efficiency gap for inference or if hyperscalers successfully commodity AI silicon through their managed services, SambaNova’s performance differentiation could erode, pressuring its hardware-centric business model and likely forcing a deeper pivot toward software and services.

Data Accuracy: YELLOW -- Competitor profiles are based on public positioning; specific performance comparisons and funding details for Cerebras are from general industry sources.

Opportunity

PUBLIC

If SambaNova executes, the prize is a foundational position in the next generation of enterprise AI infrastructure, a market where the cost of inference, not just capability, will determine winners.

The headline opportunity is to become the default on-premises and sovereign AI infrastructure provider for regulated and security-conscious enterprises and governments. While Nvidia dominates the general-purpose GPU market, SambaNova's vertically integrated stack, from custom silicon to model software, is designed for a specific wedge: customers who cannot or will not move their most sensitive data and workloads to public clouds. The company's recent partnership with Argyll to build the UK's first renewable-powered sovereign AI cloud, and its deployments with U.S. national laboratories like Los Alamos and Argonne, provide early evidence that this wedge is not just aspirational [SambaNova, 2026][Business Wire, 2025]. The outcome is a category-defining platform for sovereign and private AI, a segment where performance and efficiency are measured against specific, mission-critical requirements rather than just raw FLOPS.

Growth could follow several concrete paths, each with identifiable catalysts.

Scenario	What happens	Catalyst	Why it's plausible
Sovereign Cloud Standard	SambaNova's architecture becomes the blueprint for national AI initiatives, starting in the UK and expanding to other nations prioritizing data sovereignty.	The successful launch and adoption of the Killellan AI Growth Zone in Scotland, a partnership with Argyll Data Development [SambaNova, 2026][Business Wire, 2025].	The company is already the named technology provider for this flagship UK project, targeting defense, healthcare, and finance sectors with a 10kW-per-rack, renewable-powered design [ResultSense, 2026].
Intel-Powered Inference Scale	The multi-year collaboration with Intel evolves into a de facto standard for cost-efficient, high-volume inference on Xeon-based infrastructure, reaching a broader enterprise base than SambaNova could alone [Intel].	Joint product launches and reference architectures that combine Intel's server footprint with SambaNova's software and systems expertise.	The partnership was announced concurrently with a $350 million funding round led by Vista Equity Partners, suggesting a strategic, not just technical, alignment [Reuters, Feb 2026].
Land-and-Expand in Strategic Verticals	Initial deployments in government and national labs serve as reference accounts to win large-scale contracts in adjacent regulated industries like finance and healthcare.	A major contract with a global financial institution or healthcare network, following the pattern set by Accenture's internal three-rack deployment [Data Center Dynamics, 2026].	The company's focus on full-stack control addresses the compliance and data governance hurdles that are paramount in these sectors.

Compounding for SambaNova would look like a deepening software moat around its proprietary hardware. Each new deployment, especially in sovereign or high-security environments, generates unique performance data and tuning requirements. This feedback loop improves the company's software stack and model optimization tools, making the integrated system more efficient and sticky for existing customers while raising the bar for competitors attempting to match its end-to-end performance claims, such as its cited inference speed advantages on models like Llama 405b and DeepSeek [LinkedIn, 2026]. Success in one vertical builds a referenceable track record for the next, turning early lighthouse accounts into a distribution advantage for entering new regulated markets.

The size of the win, should the sovereign cloud scenario play out, can be framed by the strategic value of the position rather than a direct revenue multiple. While no pure-play public comparable exists, the opportunity cost for a nation or large enterprise building sovereign AI capabilities is not just hardware but time and strategic autonomy. If SambaNova captures a meaningful portion of the sovereign AI infrastructure spend,a market still being defined but driven by multi-billion-dollar national initiatives,its valuation could significantly exceed its last public mark of $5.1 billion. For context, the global AI chip market alone is projected to reach over $100 billion by 2029, with inference workloads representing a rapidly growing segment [Sacra]. Capturing even a single-digit percentage of this specialized, high-value segment would represent a scenario of substantial outcome (scenario, not a forecast).

Data Accuracy: YELLOW -- Growth scenarios are extrapolated from cited partnerships and deployments; market size projections are from a single analyst source.

Sources

PUBLIC

1. [Sacra] SambaNova Systems | https://sacra.com/c/sambanova-systems/

2. [Reuters, Feb 2026] AI chip startup SambaNova raises $350 million in Vista-led round, signs Intel partnership | https://www.reuters.com/business/media-telecom/ai-chip-startup-sambanova-raises-350-million-vista-led-round-signs-intel-2026-02-24/

3. [Bloomberg, 2024] SambaNova CEO Rodrigo Liang discusses the path to better scaling and efficiency in AI, noting that power and energy will be the AI bottleneck. | https://www.bloomberg.com/news/articles/2024-05-15/sambanova-ceo-rodrigo-liang-on-ai-efficiency-and-energy

4. [Bloomberg, 2026] SambaNova CEO Rodrigo Liang states that the next AI war will be about inference costs, compute shortages, and scaling AI infrastructure profitably. | https://www.bloomberg.com/news/articles/2026-02-24/sambanova-ceo-rodrigo-liang-on-ai-inference-costs-and-competition

5. [EE Times, 2026] SambaNova lays off 77 employees (15% of workforce) on April 22, 2025, to refocus on inference, fine-tuning, and cloud services. | https://www.eetimes.com/sambanova-systems-lays-off-15-of-workforce/

6. [Intel] Intel multi-year collaboration (inference) | https://newsroom.intel.com/data-center/intel-and-sambanova-planning-multi-year-collaboration-for-xe

7. [SambaNova, 2018] SambaNova Systems raises $56 million Series A led by GV | https://sambanova.ai/press/sambanova-systems-raises-56-million-series-a-led-by-gv

8. [TechCrunch, 2018] The red-hot AI hardware space gets even hotter with $56M for a startup called SambaNova Systems | https://techcrunch.com/2018/03/15/the-red-hot-ai-chip-space-gets-even-hotter-with-56m-for-a-startup-called-sambanova/

9. [SambaNova, 2026] SambaRack is a high-performance AI rack system designed for deploying and running large-scale models efficiently in data centers and for enterprise AI workloads. | https://sambanova.ai/products/sambarack

10. [LinkedIn, 2026] SambaNova Systems inference of Llama 405b is up to 200 tokens per second full precision on a single rack. | https://www.linkedin.com/company/sambanova/posts

11. [Business Wire, 2025] SambaNova partners with Argyll to deliver the UK’s first renewable-powered sovereign AI cloud, developing the Killellan AI Growth Zone. | https://www.businesswire.com/news/home/20251120005645/en/

12. [SambaNova, 2026] SambaNova is partnering with Argyll to deliver the UK’s first renewable-powered sovereign AI cloud, developing the Killellan AI Growth Zone. | https://sambanova.ai/press/sambanova-argyll-renewable-sovereign-ai-cloud

13. [ResultSense, 2026] Argyll Data Development is developing large-scale, renewable-powered infrastructure to enable secure, sovereign, and sustainable AI growth in the UK, targeting UK defense, healthcare, and finance with 10kW-per-rack hardware. | https://www.resultsense.com/insights/argyll-sambanova-sovereign-ai-cloud

14. [Gartner, 2024] AI Accelerator Market Forecast | https://www.gartner.com/en/newsroom/press-releases/2024-01-30-gartner-forecasts-worldwide-ai-chip-revenue-to-reach

15. [Data Center Dynamics, 2026] Accenture deployed a three-rack SambaNova Suite at its data center for generative AI work in June 2023. | https://www.datacenterdynamics.com/en/news/accenture-deploys-sambanova-suite-for-generative-ai/

16. [Crunchbase] Cerebras - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/cerebras-systems