GigaIO's PCIe Fabric Has Landed a $21 Million Bet on Edge AI

The 13-year-old infrastructure startup is shipping suitcase-sized supercomputers to the U.S. Department of Defense and scaling its composable data center racks.

About GigaIO

Published

The AI boom has turned data centers into a frantic game of musical chairs. GPUs are bolted into fixed servers, storage is siloed, and networking is a proprietary tangle. Utilization rates for expensive accelerators can dip below 30 percent. GigaIO, a 13-year-old startup based in Carlsbad, California, built its entire business on the premise that this architecture is obsolete. Its answer is a software-defined fabric that treats an entire data center rack as a single, fluid pool of resources.

The Bet on PCIe

GigaIO’s core technology is FabreX, a fabric network that uses the industry-standard PCI Express protocol to dynamically connect compute, GPUs, storage, and networking. The company calls it a universal AI memory fabric. The pitch is straightforward: by using PCIe instead of proprietary interconnects, data centers can compose and recompose hardware resources on the fly, eliminating I/O bottlenecks and boosting utilization. It is a bet on open standards over vendor lock-in, and on composability over fixed infrastructure.

This approach has yielded two flagship products. SuperNODE is a 32-GPU single-node AI supercomputer designed for large-scale training and inference in the data center. Gryf is a portable, suitcase-sized AI supercomputer built for edge deployments, co-designed with SourceCode [GigaIO, April 2025]. The latter has secured significant orders from the U.S. Department of Defense and the intelligence community [GigaIO, April 2025].

Traction in a $120 Billion Market

GigaIO operates in the $120 billion AI infrastructure market, a figure the company cites [GigaIO, Unknown]. Its traction comes in two forms: benchmark performance and strategic customer wins. In MLPerf Inference benchmarks, a SuperNODE with two MI300X accelerators demonstrated near-linear scaling, hitting 46,755 tokens per second, the highest result for a single node [GigaIO, November 2024]. The company also claims its fabric delivers 83.5x faster response times under load for large vision-language models.

On the customer side, the public record points to a mix of commercial and government clients. Beyond the DoD orders for Gryf, GigaIO lists MITRE, VISA, IBM, and Dell as key customers. The company also recently sold its SuperNODE platform and patented FabreX fabric to AI chip company d-Matrix [HPCwire, April 2026].

The Funding Path

GigaIO’s funding history shows a steady, if somewhat opaque, climb. Disclosed totals vary by source, but the most recent and concrete round is a $21 million Series B first close in July 2025, led by returning investor Impact Venture Capital [GigaIO, July 2025]. This followed a $14.7 million Series B in September 2021, also led by Impact [Business Wire, September 2021]. Other backers include SK hynix, Madrona Venture Group, and Four Palms Ventures.

2021 Series B | 14.7 | M USD
2025 Series B | 21 | M USD

Where the Wheels Could Come Off

For all its technical ambition, GigaIO faces a crowded field and an execution-heavy path. The competitive landscape includes established giants and well-funded specialists.

  • The Incumbent Wall. Hewlett Packard Enterprise, Dell, and Lenovo dominate enterprise data center sales. Displacing their integrated stacks requires not just superior technology, but a compelling total-cost-of-ownership story and proven enterprise support.
  • The Specialist Threat. Companies like Liqid also offer composable infrastructure solutions. GigaIO’s differentiation rests on its PCIe-native fabric architecture, but it must continually prove performance and cost advantages against similar approaches.
  • The Hardware Grind. Being a hardware and software company in a capital-intensive sector is a double-edged sword. It creates deeper moats but also demands significant ongoing R&D and manufacturing scale. The recent sale of its fabric technology to d-Matrix could be read as a strategic partnership or a necessary monetization of IP.

The company’s answer to these pressures is its edge focus with Gryf and its performance benchmarks. By carving out a niche in portable, high-performance edge computing for defense and intelligence, GigaIO may have found a wedge that larger players are slower to address.

The Next Twelve Months

With $21 million in fresh capital, the immediate roadmap is about scaling. The July 2025 funding announcement specifically cited scaling AI inferencing infrastructure solutions [GigaIO, July 2025]. Expect GigaIO to push deeper into its defense and intelligence beachhead with Gryf while expanding SuperNODE deployments in commercial data centers for AI training workloads. A key milestone to watch will be the naming of additional enterprise customers beyond the initial references, providing clearer evidence of market penetration against the incumbent wall.

Impact Venture Capital has now led two consecutive rounds, a vote of confidence in a long-game bet. The strategic involvement of SK hynix, a memory giant, hints at potential future collaborations at the hardware layer. For a company founded in 2012, the AI acceleration of the last few years has finally brought its core thesis into sharp, urgent focus. The question for 2026 is whether GigaIO can convert its architectural elegance and defense contracts into a durable, venture-scale business in the heart of the data center.

Sources

  1. [GigaIO, April 2025] GigaIO Announces General Availability of Gryf, the World’s First Portable AI Supercomputer | https://gigaio.com/2025/04/general-availability-gryf-portable-ai-supercomputer/
  2. [GigaIO, November 2024] GigaIO SuperNODE Achieves Highest Single-Node Performance in MLPerf Inference v4.0 | https://gigaio.com/2024/11/gigaio-supernode-achieves-highest-single-node-performance-in-mlperf-inference-v4-0/
  3. [GigaIO, July 2025] GigaIO Secures $21M to Scale AI Inferencing Infrastructure Solutions in Series B First Close | https://gigaio.com/2025/07/gigaio-secures-21m-to-scale-ai-inferencing-infrastructure-solutions-in-series-b-first-close/
  4. [Business Wire, September 2021] GigaIO Completes Oversubscribed Series B Round Totaling $14.7 Million | https://www.businesswire.com/news/home/20210921005315/en/GigaIO-Completes-Oversubscribed-Series-B-Round-Totaling-14.7-Million
  5. [HPCwire, April 2026] GigaIO Sells SuperNODE Platform and FabreX Fabric to d-Matrix | https://www.hpcwire.com/2026/04/15/gigaio-sells-supernode-platform-and-fabrex-fabric-to-d-matrix/

Read on Startuply.vc