Rivos
Last reviewed
Jun 8, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,670 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
11 citations
Review status
Source-backed
Revision
v1 · 1,670 words
Add missing citations, update stale details, or suggest a clearer explanation.
Rivos, Inc. was a fabless semiconductor startup based in Santa Clara, California, that designed server-class system-on-chip (SoC) processors pairing high-performance RISC-V CPU cores with a data-parallel accelerator aimed at large language model and data-analytics workloads. Founded in 2021 by a group of veteran chip engineers, the company positioned its hardware as an open-instruction-set alternative to the prevailing combination of x86 or Arm processors paired with NVIDIA GPUs. Rivos drew early prominence from a 2022 trade secret lawsuit brought by Apple, which it settled in 2024, and from its 2025 acquisition by Meta Platforms, which folded the team into Meta's in-house AI chip program.[1][2][3]
Rivos developed power-optimized chips that combined server-grade RISC-V CPUs with what the company called a Data Parallel Accelerator, a general-purpose GPU-style engine tuned for generative AI and data analytics. The defining design choice was tight integration of the CPU and the parallel-compute engine over a shared, coherent memory system, an approach broadly comparable to NVIDIA's Grace-Hopper and Grace-Blackwell superchips but built on the royalty-free RISC-V instruction set rather than proprietary architectures.[1][2] The company marketed its platform as a way for cloud and enterprise customers to run AI and analytics in data centers without rewriting existing applications.[4]
Rivos remained in stealth for much of its existence, emerging publicly in 2024 when it announced a large funding round and confirmed it had taped out silicon. In late 2025, Meta acquired the company, ending Rivos's independent existence after roughly four years.[3][5]
Rivos was founded in September 2021 in Santa Clara, California. Its co-founders were a cohort of experienced processor architects, several of whom had worked together across multiple chip companies over the prior two decades.[2] Key figures included:
The startup was backed from inception by industry figures including Lip-Bu Tan, founder of Walden International (and later chief executive of Intel), and Amarjit Gill, a co-founder of SiByte and P.A. Semi.[2] Reflecting the seniority of its team, Rivos reportedly began operations with more than one hundred employees on its first day.[2]
Rivos's product strategy centered on a single SoC that united a CPU complex and a data-parallel accelerator on a coherent memory fabric supporting both DDR DRAM and high-bandwidth memory (HBM).[1] The intent was to let the same chip handle general-purpose serving, database and analytics work, and AI inference and training, narrowing the data-movement penalties that arise when a separate CPU and discrete GPU communicate over a peripheral bus.[1][4]
The accelerator implemented the RISC-V vector extensions and targeted the modern RVA23 application profile, the standardized feature baseline that brings RISC-V closer to feature parity with Arm and x86 for application software.[5][6] To lower the barrier to adoption, Rivos built a CUDA-compatible software stack, allowing code written for NVIDIA's ecosystem to run on Rivos hardware, and reportedly taped out a processor running at roughly 3.1 GHz.[1] The company also pursued a PCIe-based AI accelerator card alongside its integrated server SoC.[4]
| Attribute | Detail |
|---|---|
| Company type | Fabless semiconductor startup |
| Founded | September 2021 |
| Headquarters | Santa Clara, California |
| CEO and co-founder | Puneet Kumar |
| Architecture | RISC-V CPU plus Data Parallel Accelerator (GPGPU) |
| Memory | Coherent DDR DRAM and HBM |
| Software | CUDA-compatible stack |
| Target workloads | Large language models, generative AI, data analytics |
| Status | Acquired by Meta, 2025 |
In April 2022, Apple filed suit against Rivos in the U.S. District Court for the Northern District of California, alleging that the startup had systematically poached engineers from Apple's silicon group and, in the process, misappropriated trade secrets.[7][8] Apple stated that Rivos had hired more than 40 of its former employees and accused at least two departing engineers of taking gigabytes of sensitive SoC specifications and design files in their final days at the company, allegedly using USB drives, AirDrop, and personal backups to exfiltrate material related to unreleased chips.[7][8] Apple characterized Rivos as building SoCs that would compete with its own A-series and M-series processors.[7]
Rivos contested the allegations. In April 2023, the court dismissed Apple's trade-secret claims against the company, though Apple was granted leave to file an amended complaint.[8] Rivos also countersued, arguing that Apple used overly restrictive employment agreements to discourage staff from joining competitors and to hinder emerging chip startups.[8]
The dispute was resolved in February 2024, when the two companies notified the court they had signed an agreement to settle the case.[3][9] The settlement included provisions for the remediation of Apple confidential information based on a forensic examination of Rivos's systems; the specific financial terms, if any, were not publicly disclosed.[9] The case became a frequently cited example in trade-secret and employee-mobility law, illustrating the tension between protecting proprietary designs and California's strong policy favoring worker movement between employers.[8]
With the Apple litigation behind it, Rivos publicly announced its first major funding round in April 2024, disclosing that it had raised more than $250 million in a Series A round to bring its first silicon product to market.[5][10] The round was led by Matrix Capital Management, whose representative Romit Shah joined the Rivos board.[5][10] Other participants included Intel Capital, Dell Technologies Capital, Koch Disruptive Technologies, Walden Catalyst Ventures, Cambium Capital, Hotung Venture Group, and MediaTek.[5][10] Rivos said it would use the proceeds to finalize its first chip and to expand its manufacturing and software engineering efforts.[10]
The company subsequently raised additional capital, with reports placing its cumulative funding in the range of $370 million.[2] In August 2025, shortly before its acquisition, Rivos was reported to be in talks for a further round of roughly $300 million to $400 million that would have valued the company at more than $2 billion.[11][3] That financing did not close as a standalone round; instead the discussions gave way to an acquisition by Meta.[3]
In late September and early October 2025, multiple outlets reported, and parties close to the deal confirmed, that Meta had agreed to acquire Rivos.[3][11] The acquisition was first reported around September 30, 2025, and widely covered the following day.[11][3] Terms of the transaction were not disclosed, and reporting described only that the price was likely in the billion-dollar range, consistent with the company's recent valuation discussions; no official figure was released, and the deal should not be assumed to have matched the reported $2 billion target.[3][11]
For Meta, the purchase was framed as a way to accelerate its Meta Training and Inference Accelerator (MTIA) program, the company's line of custom data-center silicon, and to reduce its heavy dependence on NVIDIA GPUs for AI training and inference.[3][1] Reporting indicated that Rivos engineers had already contributed to Meta's MTIA roadmap before the formal acquisition, and that chief executive Mark Zuckerberg had pressed for faster progress on in-house chips amid a multi-year, multi-hundred-billion-dollar data-center buildout.[1][11] Acquiring Rivos gave Meta an experienced, intact processor-design team rather than a finished product, an "acqui-hire" pattern common in the semiconductor industry.[1]
Observers compared the move to Amazon's 2015 acquisition of Annapurna Labs, which became the foundation of AWS custom chips such as Graviton and Trainium, suggesting Rivos could play a similar long-term role for Meta.[1]
Rivos's trajectory touched several themes shaping AI infrastructure in the mid-2020s. First, it was one of the more ambitious bets that the open RISC-V instruction set could move beyond embedded and microcontroller niches into the high end of the data center, competing directly with x86, Arm, and NVIDIA's CUDA-based GPU ecosystem.[6] Meta's willingness to acquire a RISC-V design house was read by some analysts as a notable vote of confidence in the architecture for large-scale AI, even though Meta did not publicly commit to shipping RISC-V products.[6][1]
Second, the deal exemplified the vertical-integration strategy pursued by hyperscale cloud and platform companies, which increasingly design their own silicon to control cost, performance, and supply. Meta's MTIA effort sits alongside Google's TPU, Amazon's Trainium and Inferentia, and Microsoft's Maia accelerators, all of which aim to lessen reliance on NVIDIA and to tailor hardware to each company's workloads.[3][1] Acquiring an outside team to accelerate that work, rather than building it organically, signaled the urgency big-tech firms felt to expand AI compute capacity.[11]
Finally, the earlier Apple litigation made Rivos a reference point in debates over engineer mobility and trade-secret protection in the semiconductor industry, where small teams of senior architects carry deep, sometimes proprietary, knowledge between competing employers.[8]