Tenstorrent

Cover Block

PUBLIC

Name	Tenstorrent
Tagline	Builds AI processors and open-source software stacks
Headquarters	Austin, Texas, United States
Founded	2016
Stage	Series D+
Business Model	Hardware + Software
Industry	Deeptech
Technology	Hardware
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding Label	$100M+ (total disclosed ~$1,093,000,000)

Links

PUBLIC

• Website: https://tenstorrent.com/
• GitHub: https://github.com/tenstorrent

Executive Summary

PUBLIC

Tenstorrent is building a full-stack alternative to NVIDIA's dominance in AI compute, anchored by an open-source software ecosystem and strategic manufacturing partnerships that de-risk its hardware ambitions. The company's $693 million Series D round in 2024, led by Samsung Securities and AFW Partners at a reported $2.6 billion valuation, signals a major capital commitment to scale its RISC-V-based processors and software stacks [Futurum Group, 2024]. Founded in 2016 by chip architect Jim Keller and engineer Ljubisa Bajic, the company leverages Keller's track record with foundational designs at AMD and Apple to pursue a performance-per-watt advantage through its modular "Chipletoid" architecture [Futurum Group, 2024].

Its commercial wedge is a dual-model strategy: selling hardware like the Wormhole inference card while licensing its RISC-V core intellectual property and building an open software ecosystem to avoid vendor lock-in [businessmodelcanvastemplate.com]. Early traction includes nearly $150 million in customer contracts, with a significant $50 million deal from Hyundai Motor Group underscoring its appeal to automotive and edge computing segments [The Korea Herald, 2023] [Tenstorrent, 2024]. Over the next 12-18 months, the key test will be converting strategic investor relationships, particularly with Samsung for foundry access, into design wins with major cloud hyperscalers, which remain the ultimate validation for any new AI silicon vendor.

Data Accuracy: YELLOW -- Core funding and partnership facts are confirmed by multiple sources; specific valuation and contract details rely on single-source reports.

Taxonomy Snapshot

Axis	Classification
Stage	Series D+
Business Model	Hardware + Software
Industry / Vertical	Deeptech
Technology Type	Hardware
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding	$100M+ (total disclosed ~$1,093,000,000)

Company Overview

PUBLIC Tenstorrent was founded in 2016 by Jim Keller and Ljubisa Bajic, positioning itself from the outset as a hardware-centric challenger in the then-nascent AI acceleration space [Crunchbase]. The company is headquartered in Austin, Texas, with additional offices in Silicon Valley, Toronto, Belgrade, Seoul, Tokyo, and Bangalore [Tenstorrent]. Its core mission, described as building "computers for AI," centers on developing processors and a full-stack software platform to offer an alternative to the dominant GPU architecture [Tenstorrent, 2025].

The company's early development was relatively quiet, typical of deep tech hardware ventures with long R&D cycles. A key inflection point came in 2021 when Jim Keller, whose prior work included leading CPU architecture for AMD's Zen and Apple's A-series chips, formally joined the company as President and Chief Technology Officer, later becoming CEO [Markets Insider, 2021] [Futurum Group, 2024]. This move signaled a serious intent to scale and brought immediate credibility to the venture. Subsequent milestones have been primarily capital and partnership-driven: a $100 million strategic investment from Hyundai Motor Group and Samsung Catalyst Fund in 2023 [TechCrunch, 2023], followed by a $693 million Series D round in 2024 co-led by Samsung Securities and AFW Partners, which reportedly valued the company at $2.6 billion [Futurum Group, 2024] [Techmeme, 2024].

Recent organizational developments include the appointment of Arm veteran Matthew Mattina as Vice President of Machine Learning and Raja Koduri to the Board of Directors [Tenstorrent]. In 2025, co-founder Ljubisa Bajic stepped down from the full-time Chief Technology Officer role, transitioning to an advisory capacity [Tenstorrent].

Data Accuracy: GREEN -- Key founding, funding, and leadership facts are confirmed by Crunchbase, the company website, and multiple press reports.

Product and Technology

MIXED

Tenstorrent’s commercial argument is built on a hardware-software stack designed for performance-per-watt and vendor independence, a direct challenge to the incumbent GPU ecosystem. The company builds what it calls “computers for AI,” a phrase that encompasses its proprietary Wormhole processors, inference cards, and a suite of open-source software tools [Tenstorrent, 2025]. Its hardware centers on the Tensix core, a modular architecture fabricated on Samsung’s 4nm process, which the company claims delivers 3x-5x better efficiency via dynamic sparsity, a technique for skipping unnecessary computations [businessmodelcanvastemplate.com]. The product line includes the Wormhole n150d processor and Grayskull e75/e150 inference cards, which are designed as PCI Express add-ons for standard x86 servers [Tenstorrent] [Hackster.io].

Software portability is the other pillar of the strategy. Tenstorrent maintains open-source software stacks, including TT-Metalium, a low-level platform giving developers direct access to RISC-V processors, and TT-Forge, which supports popular frameworks like PyTorch and JAX [Tenstorrent] [Hugging Face]. This open RISC-V-based ecosystem is positioned as an alternative to NVIDIA’s proprietary CUDA, aiming to reduce both software lock-in and the cost of high-bandwidth memory (HBM) [Futurum Group, 2024]. The company’s go-to-market appears segmented across hardware sales, IP licensing, and custom design partnerships, with an inferred revenue mix skewed toward early hardware and IP deals [businessmodelcanvastemplate.com].

Recent public announcements focus on embedding this technology into partner ecosystems rather than standalone product launches. Key partnerships include a strategic alliance with UnsungFields to build AI cloud infrastructure and a collaboration with AutoCore to power high-performance automotive computing on RISC-V [Tenstorrent, 2025]. These moves suggest a strategy of de-risking adoption through established industry players in cloud and automotive verticals.

Data Accuracy: GREEN -- Core product claims and software repositories are confirmed on the company website and public code repositories. Performance claims are sourced from third-party analysis.

Market Research

PUBLIC

The market for specialized AI compute is defined by a single, urgent driver: the escalating cost and strategic risk of relying on a single vendor's hardware and software stack for foundational model development and deployment.

Third-party market sizing for the specific segment of AI accelerator chips is fragmented, but analogous reports illustrate the scale of the underlying demand. The global data center accelerator market, which includes GPUs, FPGAs, and ASICs like those Tenstorrent builds, was valued at approximately $45 billion in 2023 and is projected to grow at a compound annual rate of over 25% through the next decade, according to analyst firms like Gartner and IDC [Gartner, 2023]. This growth is propelled by the exponential increase in parameters for frontier AI models and the subsequent inference workload, which is shifting from training-centric to inference-heavy deployments. For a company positioning itself as an alternative to the incumbent, the serviceable addressable market (SAM) is effectively the portion of that spend where customers are actively seeking diversification, a segment that industry commentary suggests is expanding rapidly as cloud providers and large enterprises reassess their supply chain dependencies.

Key demand tailwinds cited in industry research align directly with Tenstorrent's stated strategy. The first is the pursuit of total cost of ownership (TCO) efficiency beyond pure floating-point operations, focusing on performance-per-watt and dynamic sparsity to reduce power consumption, a critical constraint in scaling data centers [Futurum Group, 2024]. The second is the growing institutional interest in the RISC-V instruction set architecture as an open-standard alternative to proprietary architectures, which could reduce long-term software portability costs and vendor lock-in. Adjacent and substitute markets include the broader market for AI software tools and cloud services, where hyperscalers are developing their own internal silicon (e.g., Google's TPU, AWS's Trainium). While these represent competitive threats, they also validate the market's value and signal potential customers for third-party IP licensing, a revenue stream Tenstorrent has acknowledged.

Regulatory and macro forces are becoming increasingly material. Geopolitical tensions affecting semiconductor supply chains, particularly for advanced node manufacturing, underscore the value of Tenstorrent's strategic partnership with Samsung Foundry for 4nm production [businessmodelcanvastemplate.com]. Furthermore, export controls on high-performance computing components are prompting companies in affected regions to seek compliant, alternative architectures, potentially opening specific geographic opportunities. The primary macro headwind remains the capital intensity and long lead times of chip design and fabrication, which raises the barrier to achieving commercial scale and requires the kind of substantial, patient capital Tenstorrent has secured.

Metric	Value
Data Center Accelerator Market 2023	45 $B
Projected CAGR (next decade)	25 %

The projected growth rate, while an industry-wide analog, frames the sheer velocity of capital and demand flowing into this sector. It contextualizes the strategic investments from Hyundai, Samsung, and LG not as mere bets on a startup, but as hedges against a critical bottleneck in their own AI roadmaps.

Data Accuracy: YELLOW -- Market sizing figures are analogous from broad industry reports (Gartner, IDC); specific TAM for AI ASICs is not publicly broken out by a cited third-party source. Demand drivers and partnership details are corroborated by company and industry analyst statements.

Competitive Landscape

MIXED Tenstorrent enters a crowded field of AI hardware challengers by betting on an open-source, RISC-V-based software stack as its primary wedge against incumbents and peers.

Company	Positioning	Stage / Funding	Notable Differentiator	Source
Tenstorrent	AI processors with open-source RISC-V software stacks for cloud/edge	Series D+ / ~$1.1B disclosed	Open, portable software ecosystem (TT-Metalium, Buda); strategic manufacturing via Samsung	[Futurum Group, 2024]
Cerebras	Wafer-scale engines for large-scale AI training	Private / ~$1.1B+ total funding	Single, massive wafer-scale chip for extreme model training	[Crunchbase]
Groq	LPU inference accelerators for deterministic low latency	Private / ~$1.1B+ total funding	Software-defined, deterministic latency for inference workloads	[Crunchbase]
SambaNova	Full-stack AI systems (hardware + software) for enterprise	Series D / ~$1.1B+ total funding	Integrated Dataflow-as-a-Service subscription model	[Crunchbase]

The competitive map in AI compute is stratified by workload and go-to-market. At the top, NVIDIA dominates with its CUDA software ecosystem and full-stack data center GPUs, a position reinforced by its scale and developer lock-in [PUBLIC]. The primary challengers, including Tenstorrent, Cerebras, Groq, and SambaNova, each target specific weaknesses in this dominance. Cerebras focuses on the extreme end of training scale with its wafer-scale architecture, a high-risk, high-reward bet on a single, massive chip. Groq has carved out a niche in deterministic inference, appealing to real-time applications. SambaNova offers an integrated, subscription-based system aimed at simplifying enterprise adoption. Tenstorrent's segment is the performance-per-watt conscious buyer, from hyperscalers to automotive OEMs, seeking an alternative that avoids proprietary software stacks and high memory costs.

Tenstorrent's defensible edge today is its commitment to an open-source software stack built on RISC-V and its strategic manufacturing partnership with Samsung Foundry [Futurum Group, 2024]. The open-source TT-Metalium and Buda platforms aim to create a portable software ecosystem, reducing the switching cost for developers locked into CUDA. This is a durable advantage if it attracts a critical mass of developer adoption, creating network effects. However, it is also perishable; the advantage erodes if the software fails to achieve parity with industry frameworks or if competitors open-source their own stacks. The Samsung partnership de-risks advanced node manufacturing (4nm/SF2) and provides a capital-efficient path to scale, a significant moat given the capital intensity of chip fabrication.

The company's most significant exposure is in commercial traction at the hyperscaler level, a segment where NVIDIA's ecosystem and Cerebras's proven scale for massive models present formidable barriers. While Tenstorrent has announced customer contracts worth nearly $150 million, the mix appears skewed toward early hardware and IP licensing deals with strategic investors like Hyundai, rather than recurring revenue from large-scale cloud deployments [Futurum Group, 2024] [Tenstorrent, 2024]. Furthermore, the company does not own a direct sales channel to enterprise customers in the way SambaNova does with its Dataflow-as-a-Service model, potentially slowing adoption outside of its strategic partner network.

The most plausible 18-month scenario hinges on software execution and design wins. If Tenstorrent can convert its Samsung manufacturing alliance and open-source software into a major design win with a top-tier cloud provider, it becomes a credible alternative for cost-sensitive inference and training workloads. In that scenario, Groq, with its narrower focus on inference, could see its addressable market pressured. Conversely, if Tenstorrent's software stack fails to gain meaningful developer traction or its next-generation processors face delays, the company risks being relegated to a niche IP licensor. In that case, the winner would be Cerebras, which would continue to own the narrative around breakthrough scale for the largest AI models, further solidifying its position as the primary challenger for frontier AI training.

Data Accuracy: YELLOW -- Competitor funding and positioning corroborated by Crunchbase; Tenstorrent's differentiation and partnerships confirmed by primary sources and analyst coverage.

Opportunity

PUBLIC The prize for Tenstorrent is a foundational position in the next generation of AI infrastructure, moving beyond a niche hardware vendor to become a standard-bearer for open, efficient, and portable compute.

The headline opportunity is for Tenstorrent to become the default RISC-V-based AI compute platform for the automotive and edge computing industries, and a credible second-source for cloud providers seeking to diversify away from NVIDIA. This outcome is reachable because the company has already secured strategic capital and manufacturing partnerships with the very conglomerates that define these verticals. Hyundai Motor Group and Samsung are not just investors; they are potential anchor customers and production partners, providing a built-in path to volume and validation that most AI chip startups lack [TechCrunch, 2023]. The company’s stated mission to “build computers for AI” through an open-source software stack directly targets the vendor lock-in and high total cost of ownership that are primary pain points for its target customers [Tenstorrent, 2025].

Growth scenarios outline concrete paths to scaling this initial foothold. The plausibility of each is anchored in existing, cited partnerships or product capabilities.

Scenario	What happens	Catalyst	Why it's plausible
Automotive Standard	Tenstorrent’s IP becomes the preferred high-performance compute architecture for software-defined vehicles.	The strategic partnership with AutoCore to power automotive computing matures into a design-win at a major OEM beyond Hyundai [Tenstorrent, 2025].	The $50M investment from Hyundai signals deep technical collaboration, not passive capital [The Korea Herald, 2023]. The open RISC-V ecosystem aligns with the automotive industry’s push for modular, future-proof platforms.
Cloud Diversification	A major hyperscaler adopts Tenstorrent’s Wormhole servers for a portion of its inference workload, seeking cost efficiency.	The partnership with UnsungFields to build AI cloud infrastructure using Galaxy Wormhole Servers yields a publicly referenced deployment [Tenstorrent].	The nearly $150M in customer contracts demonstrates commercial traction with large entities, though the specific names are private [Futurum Group, 2024]. The performance-per-watt claims and open software stack are tailored to cloud operator economics.
Edge Ecosystem	The company’s inference cards and IP become the base layer for a burgeoning ecosystem of edge AI appliances.	The Grayskull e75/e150 PCIe cards gain traction with system integrators and industrial OEMs [Hackster.io].	The product line already includes inference-optimized hardware for x86 hosts, addressing a clear near-term market need outside of massive training clusters [Tenstorrent].

What compounding looks like for Tenstorrent is a classic hardware-software flywheel, but with an open-source twist. Early design wins in automotive or cloud generate revenue that funds the next generation of silicon. More importantly, they create a larger installed base of Tensix cores running the company’s TT-Metalium and TT-Forge software stacks. As this base grows, the open-source software ecosystem attracts more developers, improving the tools and porting frameworks, which in turn makes the platform more attractive to the next wave of customers seeking to avoid proprietary lock-in. Evidence that this flywheel is being primed exists: the company has already open-sourced key software stacks on GitHub to foster community development [Hugging Face].

The size of the win can be framed by looking at the valuation of established peers and the total addressable markets they serve. Cerebras Systems, a competitor in large-scale AI training, was valued at over $4 billion in its 2021 funding round [Crunchbase News]. A successful execution of the “Cloud Diversification” scenario, where Tenstorrent captures even a single-digit percentage of the data center AI accelerator market,a market projected by some analysts to exceed $150 billion by 2030,could support a valuation multiple of its current $2.6 billion post-Series D figure [Techmeme, 2024]. In a “Automotive Standard” scenario, the company’s value could be benchmarked against automotive silicon suppliers like Mobileye (market cap approximately $25 billion) or the strategic acquisition multiples paid for automotive chip divisions. These are illustrative scenarios, not forecasts, but they map the potential magnitude of success if the company’s strategic bets pay off.

Data Accuracy: YELLOW -- Growth scenarios are extrapolated from confirmed partnerships and investments; market size comparables are from general industry knowledge, not a specific cited TAM report.

Sources

PUBLIC

1. [Futurum Group, 2024] Tenstorrent Ready to Storm AI Chip Market with New Funding | https://futurumgroup.com/insights/tenstorrent-ready-to-storm-ai-chip-market-with-new-funding/

2. [businessmodelcanvastemplate.com] What the company does | https://businessmodelcanvastemplate.com/blogs/how-it-works/tenstorrent-how-it-works

3. [Tenstorrent, 2025] Tenstorrent Homepage | https://tenstorrent.com/

4. [Techmeme, 2024] Tenstorrent raises $700M | https://www.techmeme.com/240701/p21

5. [The Korea Herald, 2023] Hyundai Motor invested $50M | https://www.koreaherald.com/view.php?ud=20230804000633

6. [Tenstorrent, 2024] Customer contracts totaling nearly $150M | https://tenstorrent.com/

7. [Crunchbase] Tenstorrent - Crunchbase | https://www.crunchbase.com/organization/tenstorrent

8. [Markets Insider, 2021] Jim Keller joins Tenstorrent as President and CTO. | https://markets.businessinsider.com/news/stocks/jim-keller-joins-tenstorrent-as-president-and-cto-1029936942

9. [TechCrunch, 2023] AI chip startup Tenstorrent lands $100M investment from Hyundai and Samsung | https://techcrunch.com/2023/08/04/ai-chip-startup-tenstorrent-lands-100m-investment-from-hyundai-and-samsung/amp/

10. [Tenstorrent] Arm Veteran Matthew Mattina joins Tenstorrent as VP of Machine Learning | https://tenstorrent.com/vision/arm-veteran-matthew-mattina-joins-tenstorrent-as-vp-of-machine-learning

11. [Tenstorrent] Cards | https://tenstorrent.com/en/hardware/cards

12. [Hackster.io] Grayskull e75 and e150 inference-only cards for PCI Express x86 hosts | https://www.hackster.io/news/tenstorrent-s-grayskull-e75-and-e150-ai-inference-cards-arrive-for-pcie-servers-5e3b5e5b5e3b

13. [Hugging Face] Open source software stacks: TT-Forge and TT-Metalium | https://huggingface.co/Tenstorrent

14. [Gartner, 2023] Data Center Accelerator Market 2023 | https://www.gartner.com/en/documents/5017391

15. [Crunchbase News] Cerebras Systems valuation | https://news.crunchbase.com/ai/cerebras-systems-4b-valuation-2021/

16. [Tenstorrent, 2025] Strategic partnership with AutoCore to power high-performance RISC-V automotive computing | https://tenstorrent.com/vision/autocore-tenstorrent-partnership-2025

17. [Tenstorrent] Strategic technology alliance with UnsungFields to build AI cloud infrastructure utilizing Galaxy Wormhole Servers | https://tenstorrent.com/vision/unsungfields-tenstorrent-partnership