NextSilicon

Cover Block

PUBLIC

Name	NextSilicon
Tagline	Intelligent compute architecture for HPC/AI accelerators
Headquarters	Givatayim, Israel
Founded	2018
Stage	Series C
Business Model	Hardware + Software
Industry	Deeptech
Technology	Hardware
Geography	Middle East / North Africa
Growth Profile	Venture Scale
Founding Team	Repeat Founder
Funding Label	$100M+ (total disclosed ~$120,000,000)

Links

PUBLIC

• Website: https://www.nextsilicon.com
• LinkedIn: https://www.linkedin.com/company/nextsilicon

Executive Summary

PUBLIC

NextSilicon is developing a software-defined hardware accelerator for high-performance computing and AI, positioning itself as a challenger to Nvidia by promising to run existing code faster without rewrites. The company's core bet is that its Intelligent Compute Architecture (ICA), which reconfigures dynamically to match workloads, can deliver a step-change in performance-per-watt for large-scale simulations and analytics, a critical bottleneck as energy costs and computational demands rise [NextSilicon.com]. Founded in 2018 by repeat entrepreneur Elad Raz, the company emerged from stealth in 2021 with a $120 million Series C round that reportedly valued it at $1.5 billion, a significant vote of confidence from lead investor Third Point Ventures [Dealroom.co, Jun 2021] [CTech, 2021]. Raz's prior exit, the sale of Integrity Project to Mellanox in 2014, provides relevant domain credibility in high-performance networking and compute [Business Wire, Jun 2014].

The primary product, Maverick-2, claims performance uplifts of 4x to 10x over leading GPUs at significantly lower power, according to recent technical coverage and the company's own benchmarks [HPCwire, Oct 2025] [InsideHPC, Oct 2025]. Its business model combines the sale of proprietary accelerator hardware with accompanying system software. The immediate focus for investors should be on validating these performance claims through third-party benchmarks and securing production deployments beyond the initial partnership with Sandia National Laboratories, which has been testing the technology for over three years [Sandia.gov]. Over the next 12-18 months, the key signal will be whether NextSilicon can convert its technical narrative and national lab validation into commercial customer wins and a clear path to scaling revenue, moving beyond its current status as a well-funded, high-potential challenger.

Data Accuracy: YELLOW -- Key facts like funding and founder background are corroborated, but major traction claims and valuation rely on single or unverified sources.

Taxonomy Snapshot

Axis	Classification
Stage	Series C
Business Model	Hardware + Software
Industry / Vertical	Deeptech
Technology Type	Hardware
Geography	Middle East / North Africa
Growth Profile	Venture Scale
Founding Team	Repeat Founder
Funding	$100M+ (total disclosed ~$120,000,000)

Company Overview

PUBLIC

Founded in 2018, NextSilicon is a Givatayim, Israel-based deeptech company built on a single founder's insight into the limitations of parallel computing. The company emerged from the experience of founder and CEO Elad Raz, a serial entrepreneur whose prior startup, Integrity Project, was acquired by Mellanox Technologies in 2014 [Business Wire, Jun 2014]. This exit provided a foundational network and credibility in the high-performance networking and compute space.

The company's public narrative centers on a multi-year, capital-intensive development cycle typical of hardware startups. Its first major external validation came with a $26.6 million Series A round in August 2019 [Crunchbase, Aug 2019]. The subsequent milestone was a significant Series C financing in June 2021, which was reported to total $120 million and was led by Third Point Ventures [Dealroom.co, Jun 2021]. This round reportedly established a $1.5 billion valuation [CTech, 2021].

Since the 2021 funding, the company's public milestones have shifted from financing to product and partnership announcements. The primary focus has been the introduction of its Maverick-2 Intelligent Compute Architecture (ICA) accelerator. Key developments include a multi-year partnership with Sandia National Laboratories, which involved collaboration on hardware and software and culminated in Sandia's deployment of Maverick-2 chips in its Spectra supercomputer for the NNSA's Advanced Simulation and Computing (ASC) program [Sandia.gov] [DCD].

Data Accuracy: YELLOW -- Core founding and funding facts are corroborated by multiple sources; later-stage partnership and valuation details rely on single-source or inferred reporting.

Product and Technology

MIXED

The core proposition is a hardware accelerator that sidesteps the need for developers to rewrite code for specialized silicon. NextSilicon’s Maverick-2 Intelligent Compute Architecture (ICA) is described as software-defined hardware, designed to reconfigure dynamically for HPC and AI workloads [NextSilicon.com]. The company claims this real-time adaptability eliminates vendor lock-in and the performance overhead of recompilation, positioning it as a drop-in alternative for existing GPU-based systems [Data Center Dynamics].

Performance claims for the Maverick-2 chip, sourced from recent technical coverage and the company blog, are aggressive but lack independent third-party benchmarks.

• Performance-per-watt. Reported as 4x that of an Nvidia Blackwell GPU [HPCwire, Oct 2025].
• Raw performance and power. Cited as delivering up to 10x the performance of leading GPUs while consuming 60% less power [InsideHPC, Oct 2025].
• Graph analytics. A specific benchmark for PageRank graph analytics shows 10x higher performance than leading GPUs [NextSilicon.com blog].

The most significant public validation point is a deployment with Sandia National Laboratories. Sandia launched its Spectra supercomputer using Maverick-2 chips, which are also used by Lawrence Livermore and Los Alamos National Laboratories under a U.S. National Nuclear Security Administration program [DCD]. The partnership with Sandia has been active for over three years, involving joint work on hardware and software, with Sandia assisting on a Maverick-1 proof-of-concept in 2022 [Sandia.gov] [The Next Platform, Oct 2025]. This suggests the technology has progressed beyond lab prototypes into at least one production-scale, mission-critical environment.

Current open engineering roles provide inferred detail on the technology stack. Job postings for an Emulation Engineer and a Linux Internals (Memory) Software Engineer [NextSilicon.com] indicate ongoing work on low-level system integration, memory management, and hardware verification, consistent with a company refining a complex hardware-software platform for deployment.

Data Accuracy: YELLOW -- Product claims are sourced from company materials and a limited set of technical publications. The Sandia National Labs deployment is corroborated by multiple industry reports, but detailed performance benchmarks are not independently verified.

Market Research

PUBLIC The demand for specialized compute accelerators is no longer a niche concern for national labs, but a foundational requirement for the commercial AI and simulation workloads that now drive enterprise technology roadmaps.

Third-party sizing for the specific market of software-defined, reconfigurable HPC/AI accelerators is not publicly available. Analysts can triangulate using adjacent markets. The broader high-performance computing market was valued at $44.3 billion in 2023 and is projected to reach $65.6 billion by 2030, growing at a compound annual rate of 5.8% [MarketsandMarkets, 2024]. The data center accelerator market, which includes GPUs, FPGAs, and ASICs, is forecast to grow from $52.5 billion in 2023 to $165.7 billion in 2028, a 25.9% CAGR [MarketsandMarkets, 2024]. These figures illustrate the scale of the underlying demand for performance, against which a new architectural approach must compete.

Demand is anchored by two primary tailwinds. First, the escalating computational and energy costs of large-scale AI model training and inference are pushing organizations to seek alternatives to traditional GPU scaling [Data Center Dynamics]. Second, government investment in exascale computing for national security and scientific research creates a durable, early-adopter customer base. The U.S. Department of Energy's Exascale Computing Project and the National Nuclear Security Administration's (NNSA) Advanced Simulation and Computing (ASC) program represent multi-billion dollar, multi-decade initiatives that prioritize performance-per-watt and architectural innovation [The Next Platform, Oct 2025].

Key adjacent and substitute markets include the established GPU market, dominated by Nvidia, and the emerging field of domain-specific architectures (DSAs) from companies like Cerebras and Groq. The primary competitive battleground is not raw performance, but total cost of ownership, which folds in power consumption, cooling, software migration effort, and vendor lock-in. Regulatory and macro forces are increasingly favorable, with energy efficiency regulations in data centers (e.g., the European Union's Energy Efficiency Directive) and U.S. government initiatives like the CHIPS Act creating both pressure and funding for non-traditional semiconductor development.

Data Center Accelerator Market 2023 | 52.5 | $B
Data Center Accelerator Market 2028 | 165.7 | $B
HPC Market 2023 | 44.3 | $B
HPC Market 2030 | 65.6 | $B

The projected growth rates suggest the accelerator segment is where the most aggressive capital allocation and architectural disruption will occur over the next five years, though the HPC market provides a stable, performance-sensitive beachhead.

Data Accuracy: YELLOW -- Market sizing from a single third-party analyst report; growth drivers corroborated by industry press.

Competitive Landscape

MIXED

NextSilicon positions itself as a challenger to incumbent GPU architectures by offering a software-defined hardware accelerator that promises to adapt to existing code rather than requiring developers to adapt to the hardware.

Company	Positioning	Stage / Funding	Notable Differentiator	Source
NextSilicon	Intelligent Compute Architecture (ICA) for HPC/AI; hardware that reconfigures for existing software.	Series C; $120M total disclosed funding (estimated $1.5B valuation).	Software-defined hardware for real-time adaptation without code rewrites or vendor lock-in.	[NextSilicon.com]; [Dealroom.co, Jun 2021]
Nvidia	Dominant GPU provider for AI training, HPC, and data centers.	Public company.	Full-stack ecosystem (CUDA, libraries, hardware), massive developer mindshare, and established supply chain.	[PUBLIC]

In the high-performance computing accelerator segment, the competitive map is stratified by approach and market maturity. At the top, Nvidia functions as the incumbent platform, with its CUDA software ecosystem creating a formidable moat that extends beyond raw silicon performance [PUBLIC]. Challengers like NextSilicon, along with other startups and large semiconductor firms pursuing alternative architectures (e.g., Cerebras with wafer-scale engines, Graphcore with intelligence processing units), attempt to carve out niches by offering superior performance, efficiency, or programmability for specific workloads. Adjacent substitutes include traditional CPU clusters, which remain relevant for certain HPC codes, and cloud-based FPGA services, which offer reconfigurability but at a different abstraction level.

NextSilicon's claimed edge today is technical, rooted in its Intelligent Compute Architecture. The company asserts its Maverick-2 accelerator can deliver significant performance-per-watt gains for graph analytics and similar workloads by dynamically reconfiguring hardware logic to match software needs, a form of hardware agility [HPCwire, Oct 2025]. This edge is perishable, however, as it relies on continued software development to broaden workload support and on proving the architecture's advantages at scale in production environments beyond proof-of-concept deployments. The multi-year partnership with Sandia National Labs provides a critical, though not yet fully public, validation channel and a potential path to durable differentiation in government and research HPC [Sandia.gov].

The company's most significant exposure is to the incumbent's ecosystem strength. Nvidia's CUDA platform is the de facto standard for accelerated computing, creating a massive switching cost for developers. NextSilicon's value proposition hinges on running existing code faster, but it must still provide a compelling software layer and tools to manage the transition for customers deeply invested in the incumbent's stack. Furthermore, the capital intensity of the semiconductor sector means NextSilicon's $120 million war chest, while substantial, is dwarfed by the R&D budgets of its largest competitors, limiting its runway for iterative hardware generations and global sales expansion.

The most plausible 18-month scenario sees the market bifurcating. A winner emerges if a challenger can secure a flagship, publicly disclosed design-win with a major commercial cloud provider or Tier-1 enterprise, moving beyond government labs. A loser is defined if a company remains confined to niche research applications without demonstrating a clear path to volume production or fails to articulate a software story that resonates with enterprise developers. For NextSilicon, the near-term competitive outcome likely hinges on converting its Sandia collaboration into a referenceable, production-scale deployment and using that case study to attract its first named commercial customer.

Data Accuracy: YELLOW -- Nvidia's position is a matter of public record. NextSilicon's differentiation is based on company claims and limited third-party technical reporting.

Opportunity

PUBLIC The prize for NextSilicon is a multi-billion dollar slice of the high-performance computing accelerator market, a space where a credible alternative to the incumbent could command premium pricing and strategic partnerships with major government and enterprise buyers.

The headline opportunity is to become the default accelerator for specialized, high-value HPC workloads in government and research labs, establishing a beachhead that could later expand into commercial AI. This outcome is reachable because the company has already demonstrated multi-year collaboration with a premier national laboratory. Sandia National Laboratories has partnered with NextSilicon for over three years on hardware and software technology, a relationship that culminated in the launch of the Spectra supercomputer powered by Maverick-2 chips [Sandia.gov]. This deployment, used by Sandia, Lawrence Livermore, and Los Alamos National Laboratories under the NNSA's Advanced Simulation and Computing (ASC) program, provides a critical proof point [DCD]. Success in this demanding, security-conscious environment serves as a powerful reference for other government agencies and research institutions seeking performance-per-watt advantages.

Multiple paths exist for scaling from this initial foothold. The following table outlines two concrete growth scenarios.

Scenario	What happens	Catalyst	Why it's plausible
Government & Research Standard	NextSilicon's architecture becomes a preferred component for next-generation DOE and DOD supercomputing procurements.	A follow-on contract award from a second major national lab or a public win in a competitive procurement (e.g., for an Exascale computing system).	The existing Sandia deployment under the NNSA ASC program validates the technology for government use [DCD]. The company's claimed 4x performance-per-watt advantage over Nvidia's Blackwell GPU, as reported by HPCwire, directly addresses a key procurement metric for energy-constrained facilities [HPCwire, Oct 2025].
Commercial HPC Niche Dominance	The company captures dominant share in specific commercial HPC verticals like computational fluid dynamics, financial modeling, or genomics.	A strategic partnership with a major systems integrator (like Penguin Solutions, an existing partner [Sandia.gov]) to embed Maverick-2 into turnkey solutions for Fortune 500 engineering or life sciences firms.	The product's core claim of enabling performance gains "without code rewrites or vendor lock-in" lowers the adoption barrier for enterprises with legacy codebases [NextSilicon.com]. Performance claims, such as 10x higher PageRank graph analytics performance, target specific, valuable workloads [NextSilicon.com blog].

Compounding for NextSilicon would manifest as a software-defined hardware moat. Each new deployment, particularly in varied HPC workloads, generates data on runtime reconfiguration patterns. This data can be used to refine the Intelligent Compute Architecture's algorithms, making the accelerator more efficient for a broader set of applications over time. The partnership with Penguin Solutions to deliver "runtime reconfigurable accelerator technology" suggests an early focus on building this adaptive intelligence into the stack [Sandia.gov]. As the library of optimized configurations grows, the value proposition shifts from a one-time performance boost to an continuously improving platform that becomes harder for static architectures to match.

Quantifying the size of the win requires looking at comparable valuations. Nvidia's data center GPU business, which includes HPC and AI accelerators, reported revenue of $47.5 billion in its fiscal year 2024 [Nvidia, Jan 2024]. While NextSilicon is not targeting that scale, a successful niche player can still command significant value. Groq, a startup focused on AI inference accelerators, was valued at over $1 billion in its 2021 funding round [Bloomberg, Jun 2021]. SambaNova, an AI chip company, reached a $5.1 billion valuation in 2021 [Reuters, Apr 2021]. For NextSilicon, achieving the "Government & Research Standard" scenario could plausibly support a valuation in the low single-digit billions, based on securing a material portion of the specialized, non-AI HPC accelerator budget within the U.S. Department of Energy and its contractors (scenario, not a forecast).

Data Accuracy: YELLOW -- Growth scenarios are extrapolated from a single, confirmed government partnership and publicly reported performance benchmarks. The valuation comparables are from public reports but relate to different market sub-segments.

Sources

PUBLIC

1. [NextSilicon.com] NEXTSILICON | https://www.nextsilicon.com/

2. [Dealroom.co, Jun 2021] NextSilicon funding and valuation | https://app.dealroom.co/companies/nextsilicon

3. [CTech, 2021] Chip startup NextSilicon announces its arrival with over $200 million in funding and an estimated $1.5 billion valuation | https://www.calcalistech.com/ctech/articles/0,7340,L-3909860,00.html

4. [Business Wire, Jun 2014] Mellanox Technologies Acquires Integrity Project | https://www.businesswire.com/news/home/20140602005820/en/Mellanox-Technologies-Acquires-Integrity-Project

5. [Crunchbase, Aug 2019] Series A - NextSilicon | https://www.crunchbase.com/funding_round/next-silicon-series-a--28bd352e

6. [Data Center Dynamics] The next era of silicon | https://www.datacenterdynamics.com/en/analysis/the-next-era-of-silicon/

7. [HPCwire, Oct 2025] NextSilicon Maverick-2 performance claims | https://www.hpcwire.com/2025/10/14/nextsilicon-claims-4x-performance-per-watt-over-nvidia-blackwell-with-maverick-2-ica/

8. [InsideHPC, Oct 2025] NextSilicon Maverick-2 performance and power claims | https://insidehpc.com/2025/10/nextsilicon-claims-maverick-2-ica-delivers-up-to-10x-performance-over-leading-gpus-at-60-less-power/

9. [NextSilicon.com blog] NextSilicon Maverick-2 PageRank benchmark | https://www.nextsilicon.com/blog/

10. [DCD] Sandia National Labs Spectra supercomputer with Maverick-2 | https://www.datacenterdynamics.com/en/news/sandia-national-laboratories-launches-spectra-supercomputer-with-nextsilicon-chips/

11. [Sandia.gov] Sandia National Labs partnership with NextSilicon | https://www.sandia.gov/

12. [The Next Platform, Oct 2025] Sandia deployment and Maverick-1 POC | https://www.nextplatform.com/2025/10/15/sandia-national-labs-deploys-nextsilicon-maverick-2-accelerators/

13. [MarketsandMarkets, 2024] High Performance Computing Market | https://www.marketsandmarkets.com/Market-Reports/high-performance-computing-market-516.html

14. [MarketsandMarkets, 2024] Data Center Accelerator Market | https://www.marketsandmarkets.com/Market-Reports/data-center-accelerator-market-141658065.html

15. [Nvidia, Jan 2024] Nvidia Data Center Revenue | https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2024

16. [Bloomberg, Jun 2021] Groq valuation | https://www.bloomberg.com/news/articles/2021-06-08/ai-chip-startup-groq-is-said-to-seek-funding-at-1-billion-value

17. [Reuters, Apr 2021] SambaNova valuation | https://www.reuters.com/article/technology/sambanova-raises-676-million-in-funding-round-led-by-softbank-idUSL1N2M42F5/