NextSilicon's Reconfigurable Chip Lands a Supercomputer at Sandia National Labs

The promise is written in the first line of code you don't have to write. A researcher at a national lab, running a complex graph analytics job, loads their existing C++ program. The system, outfitted with a NextSilicon Maverick-2 accelerator, reads the workload and reconfigures its own hardware on the fly. No rewrite, no recompilation, just the quiet hum of silicon adapting itself to the problem. This is the foundational fiction of NextSilicon, a six-year-old Israeli deeptech company: that the most intelligent compute architecture isn't the one that runs the fastest code, but the one that makes the code you already have run faster.

A wedge into the fortress

NextSilicon's bet is not to build a better general-purpose GPU. It is to circumvent the need for one altogether in the high-performance computing (HPC) fortress. The company's Intelligent Compute Architecture (ICA) is a software-defined hardware accelerator designed to dynamically reconfigure for specific HPC and AI workloads [NextSilicon.com]. The flagship Maverick-2 chip claims performance leaps that read like science fiction next to established giants: 4x the performance-per-watt of an Nvidia Blackwell GPU, and up to 10x the raw performance of leading GPUs while drawing 60% less power [HPCwire, Oct 2025] [InsideHPC, Oct 2025]. For a field where power consumption is a primary constraint, these are not incremental claims. They are the kind of numbers that get you a meeting at a place like Sandia National Laboratories.

The proof is in the supercomputer

That meeting appears to have gone well. The most concrete validation of NextSilicon's technology to date is its deployment inside Spectra, a new supercomputer launched by Sandia National Labs. The system uses Maverick-2 chips and is employed by Sandia, Lawrence Livermore, and Los Alamos National Laboratories under the National Nuclear Security Administration's Advanced Simulation and Computing (ASC) program [DCD]. This is not a pilot or a proof-of-concept; it is production hardware solving some of the world's most demanding computational problems. The partnership with Sandia stretches back over three years, involving joint work on both hardware and software, and was facilitated through a collaboration with Penguin Solutions [Sandia.gov]. For a startup positioning itself as a challenger to an industry titan, a named, high-stakes customer in a national lab is the ultimate credibility signal.

The company's trajectory and capacity are underscored by its funding and scale.

2019 Series A | 26.6 | M USD
2021 Series C | 120 | M USD

Led by repeat founder Elad Raz, who sold his previous startup to Mellanox in 2014, NextSilicon raised a $120 million Series C in June 2021 at a reported $1.5 billion valuation, with Third Point Ventures leading the round [Dealroom.co, Jun 2021] [CTech]. Public estimates place its headcount between 222 and 330 employees, suggesting a substantial engineering operation is underway [ZoomInfo] [trueup.io].

The long road from wedge to wall

Yet, taking a wedge into a fortress is different from storming the walls. NextSilicon's path is paved with immense technical and commercial challenges. The HPC market, while lucrative, is a fraction of the total addressable market dominated by Nvidia's CUDA ecosystem. NextSilicon's success hinges on convincing developers and procurement officers at other labs and commercial entities to adopt a novel architecture. The risks are not merely about performance, but about the entire software stack, developer tools, and long-term vendor stability.

• The ecosystem moat. Nvidia's dominance is protected not just by hardware, but by CUDA, a vast software ecosystem that has become the de facto standard for accelerated computing. NextSilicon's promise of "no code rewrites" is a direct attack on this moat, but displacing an entrenched standard requires more than a faster chip.
• The scaling challenge. A design win at one national lab, even a prestigious one, must be replicated. The sales cycle in government and enterprise HPC is measured in years, not quarters. NextSilicon's estimated revenue of $9.2 million, while not confirmed, suggests the commercial ramp is still in its early stages [ZoomInfo].
• The capital intensity. Building hardware is a capital-intensive endeavor with long R&D cycles. The $120 million war chest is substantial, but it must fund both continued chip development and a global go-to-market push against a competitor with nearly limitless resources.

What to watch in the next cycle

For NextSilicon, the next twelve months will be about moving from a singular, prestigious reference customer to a repeatable commercial pattern. The open roles on its website, like Director of Partner Enablement, signal a focus on building the channel and support infrastructure required for broader adoption [NextSilicon.com]. The key metrics to watch will be announcements of follow-on deployments at other national labs or large commercial HPC clusters, and any expansion of its partnership with Penguin Solutions or similar system integrators.

Ultimately, NextSilicon is answering a cultural question that has simmered beneath the AI boom's insatiable demand for compute: what if the answer isn't just more, bigger, hotter chips, but smarter, more adaptable ones? The company is betting that in the specialized corridors of high-performance computing, where every watt and every millisecond is scrutinized, efficiency and flexibility will trump raw, general-purpose scale. The code is already written. The question is whether the hardware will learn to read it.

Sources

1. [NextSilicon.com] Company Website | https://www.nextsilicon.com/
2. [HPCwire, Oct 2025] NextSilicon Maverick-2 Performance Claims | https://www.hpcwire.com
3. [InsideHPC, Oct 2025] NextSilicon Maverick-2 Performance and Power Analysis | https://insidehpc.com
4. [DCD] Sandia National Labs Spectra Supercomputer Deployment | https://www.datacenterdynamics.com
5. [Sandia.gov] NextSilicon and Penguin Solutions Partnership Announcement | https://www.sandia.gov
6. [Dealroom.co, Jun 2021] NextSilicon Series C Funding Round | https://app.dealroom.co/companies/nextsilicon
7. [CTech] NextSilicon Funding and Valuation Report | https://www.calcalistech.com/ctech/articles/0,7340,L-3909860,00.html
8. [ZoomInfo] NextSilicon Employee and Revenue Estimates | https://www.zoominfo.com/c/nextsilicon/547299897
9. [trueup.io] NextSilicon Employee Count | https://trueup.io