GigaIO
Composable, software-defined rack-scale infrastructure for AI and HPC using PCIe-based FabreX fabric.
Website: https://gigaio.com/
PUBLIC
| Name | GigaIO |
| Tagline | Composable, software-defined rack-scale infrastructure for AI and HPC using PCIe-based FabreX fabric. |
| Headquarters | Carlsbad, United States |
| Founded | 2012 |
| Stage | Series B |
| Business Model | Hardware + Software |
| Industry | Deeptech |
| Technology | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Solo Founder |
| Funding Label | Series B (total disclosed ~$45,880,000) |
Links
PUBLIC
- Website: https://gigaio.com/
- LinkedIn: https://www.linkedin.com/company/gigaio
- X / Twitter: https://x.com/gigaioinc
Data Accuracy: GREEN -- Confirmed by company press releases and public profiles.
Executive Summary
PUBLIC GigaIO builds composable, software-defined rack-scale infrastructure for AI and high-performance computing workloads, a bet that the future of data center architecture lies in dynamic, PCIe-based resource pools rather than fixed server clusters [GigaIO]. Founded in 2012 by Joey Maitra, the company's core wedge is its FabreX fabric, a proprietary network that uses the industry-standard PCI Express protocol to disaggregate and recombine GPUs, storage, and networking on demand, aiming to reduce I/O bottlenecks and improve utilization [Business Wire, September 2021]. The product portfolio includes the SuperNODE, a 32-GPU data center appliance, and the Gryf, a portable, suitcase-sized AI supercomputer for edge deployments that has secured orders from U.S. defense and intelligence agencies [GigaIO, April 2025]. The founding team is led by Maitra, the technical architect, with CEO Alan Benjamin and a bench of sales and engineering executives from enterprise hardware backgrounds steering commercial execution [Crunchbase].
Capitalization shows a Series B extension closed in July 2025, bringing total disclosed funding to at least $45.9 million from a mix of venture and strategic investors, including SK hynix and Impact Venture Capital [HPCwire, July 2025]. The business model combines hardware and software sales, targeting enterprises in life sciences, financial services, and manufacturing that require flexible, high-performance infrastructure. Over the next 12-18 months, the key watchpoints are the commercial scaling of the Gryf platform in the defense sector and the expansion of SuperNODE deployments against incumbent offerings from HPE and Dell, which will test the company's ability to convert its technical differentiation into sustained revenue growth.
Data Accuracy: YELLOW -- Core product claims and recent funding are confirmed by company releases and industry press; total funding history and some team details rely on single sources or unverified databases.
Taxonomy Snapshot
| Axis | Classification |
|---|---|
| Stage | Series B |
| Business Model | Hardware + Software |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Solo Founder |
| Funding | Series B (total disclosed ~$45,880,000) |
Company Overview
PUBLIC
GigaIO operates as a hardware and software company from Carlsbad, California, with a founding date consistently cited as 2012 [TheCompanyCheck]. The company was founded by Joey Maitra, who is listed as the founder and holds the title of Chief Architect [TheCompanyCheck, LinkedIn]. The founding narrative positions GigaIO as aiming to bring cloud-like elasticity and composability to on-premises data centers, a concept the company likens to being "to Hardware What VMware is to Software" [GigaIO].
Key milestones trace the evolution from its founding to its current focus on AI infrastructure. The company announced an oversubscribed Series B round of $14.7 million in September 2021, led by Impact Venture Capital [Business Wire, September 2021]. A significant product milestone was reached in April 2025 with the general availability announcement of Gryf, a portable AI supercomputer developed for edge environments, which the company stated had already secured significant orders from the U.S. Department of Defense and intelligence community [GigaIO, April 2025]. Most recently, in July 2025, GigaIO secured a $21 million first close of a Series B extension, again led by Impact Venture Capital, to scale its AI inferencing infrastructure solutions [GigaIO, July 2025].
Data Accuracy: YELLOW -- Founding year and founder confirmed by one source; executive team and funding events corroborated by multiple press releases.
Product and Technology
MIXED
GigaIO's product architecture is built on a single, foundational premise: that the PCI Express (PCIe) bus, the standard for connecting high-speed peripherals inside a server, can be extended across an entire rack to create a dynamic, software-defined fabric. The company's core technology, FabreX, is a patented PCIe-based fabric network that allows data center operators to disaggregate and recombine compute, GPU accelerators, storage, and networking resources on demand [GigaIO]. This approach, which the company likens to what VMware did for software, aims to eliminate the I/O bottlenecks that plague traditional fixed-server AI and HPC clusters, promising higher resource utilization and reduced total cost of ownership [GigaIO].
Two flagship products demonstrate the application of this fabric. For the data center, the SuperNODE is a 32-GPU single-node AI supercomputer designed for large-scale training and inference workloads [GigaIO]. Public benchmarks show strong technical performance, with the system achieving 99.7% of ideal theoretical scaling in HPL-MxP tests and, in a two-accelerator (16-GPU) configuration, demonstrating near-linear scaling to 46,755 tokens per second in MLPerf Inference benchmarks, which the company claims is the highest result for a single node [GigaIO, November 2024], [HPCwire]. For edge and field deployments, GigaIO offers Gryf, a portable, suitcase-sized AI supercomputer co-designed with SourceCode [GigaIO, April 2025]. Gryf extends the FabreX composable architecture to rugged environments, can be stacked up to five units, and has secured what the company describes as significant orders from the U.S. Department of Defense and intelligence community [GigaIO, April 2025]. The system offers six different sled types for GPU, compute, storage, and networking components [GigaIO].
A notable recent development is the sale of GigaIO's SuperNODE platform and its FabreX fabric intellectual property to d-Matrix, a deal announced in April 2026 [HPCwire, April 2026], [BigDATAwire, April 2026]. This transaction suggests a strategic pivot or partnership that could reshape GigaIO's future product focus, potentially toward its edge-oriented Gryf platform or next-generation fabric development. The company's public materials consistently emphasize its dual identity as both a hardware and software provider, with the software layer enabling the dynamic composition and management of the underlying PCIe fabric resources [GigaIO].
Data Accuracy: GREEN -- Product details, specifications, and performance claims are sourced directly from company press releases and website. The d-Matrix asset sale is reported by multiple industry publications.
Market Research
PUBLIC
The demand for infrastructure that can efficiently compose and scale AI accelerators is no longer a niche concern but a core operational challenge for any enterprise building or deploying large models.
GigaIO's primary addressable market is the broader AI infrastructure sector, which the company cites as a $120 billion opportunity [GigaIO]. This figure aligns with third-party market analyses, such as those from Grand View Research and MarketsandMarkets, which project the global AI infrastructure market to reach between $120 billion and $150 billion by the end of the decade, growing at a compound annual growth rate of approximately 25% [Grand View Research, 2024], [MarketsandMarkets, 2024]. The composable infrastructure segment within this, which allows dynamic pooling of compute, storage, and networking, is a critical and faster-growing subset as enterprises seek to maximize utilization of expensive GPU resources.
Demand is driven by several converging tailwinds. The primary driver is the exponential increase in model size and complexity, which strains traditional, fixed-server architectures and creates severe I/O bottlenecks [HPCwire]. This is compounded by the rapid diversification of accelerator types (GPUs, FPGAs, custom ASICs) from multiple vendors, creating a need for a vendor-agnostic fabric to manage heterogeneity. Furthermore, the shift of AI workloads from centralized cloud regions to on-premises and edge locations for latency, cost, and data sovereignty reasons creates a need for the portable, rack-scale solutions GigaIO offers [GigaIO, April 2025].
Adjacent and substitute markets present both opportunities and competitive pressure. The company's technology touches the high-performance computing (HPC) market, valued at over $40 billion, where scientific and engineering simulations have long demanded low-latency interconnects [Hyperion Research, 2024]. The traditional server and storage networking market, dominated by Ethernet and InfiniBand, represents the incumbent substitute; GigaIO's bet is that a PCIe-native fabric offers a performance and flexibility advantage for accelerator-centric workloads. The edge AI inference market, projected to exceed $20 billion by 2028, is a direct target for products like the Gryf portable supercomputer [ABI Research, 2024].
Regulatory and macro forces are largely supportive but introduce complexity. Increased scrutiny of large cloud providers in certain jurisdictions and growing data localization laws incentivize on-premises and sovereign AI deployments, a tailwind for GigaIO's data center and edge offerings. However, the cyclical nature of semiconductor supply, particularly for advanced GPUs, can impact the timing of capital expenditures for its customers. The strategic investment from SK hynix suggests an alignment with broader industry efforts to develop high-bandwidth memory solutions optimized for AI, which could become a regulatory focal point in global tech competition [Business Wire, September 2021].
Total AI Infrastructure Market (2030) | 120 | $B
Edge AI Inference Market (2028) | 20 | $B
HPC Market (Current) | 40 | $B
The cited market sizes, while from different sources and timeframes, illustrate the layered opportunity: a massive total addressable market in AI infrastructure, with substantial sub-segments in HPC and edge inference where GigaIO's composable architecture is positioned as a solution.
Data Accuracy: YELLOW -- The $120B TAM is company-cited. Adjacent market figures are drawn from analogous, named third-party reports.
Competitive Landscape
MIXED GigaIO's competitive position is defined by its attempt to carve out a space between large-scale, integrated system vendors and software-only composability solutions, using a hardware-accelerated, PCIe-based fabric as its core wedge.
| Company | Positioning | Stage / Funding | Notable Differentiator | Source |
|---|---|---|---|---|
| GigaIO | Composable rack-scale infrastructure via proprietary PCIe fabric (FabreX). | Series B (~$45.9M total disclosed). | PCIe-native hardware fabric for low-latency, high-bandwidth composability across heterogeneous accelerators. | [Business Wire, September 2021], [HPCwire, July 2025] |
| Liqid | Composable disaggregated infrastructure (CDI) platform. | Acquired by Liqid Inc. (private). | Software-defined composability across PCIe, Ethernet, and InfiniBand, with a focus on data center orchestration. | [PUBLIC] |
| HPE | Integrated server and hyperconverged infrastructure with composable software (Synergy). | Public company. | Full-stack offering from hardware to management software, with deep enterprise sales channels and global service. | [PUBLIC] |
| Dell EMC | Integrated server (PowerEdge) and hyperconverged (VxRail) solutions. | Public company. | Broad portfolio, entrenched customer relationships, and scale advantages in procurement and support. | [PUBLIC] |
| Rockport Networks | High-performance, switchless network fabric for HPC/AI. | Venture-backed. | Optical network fabric architecture designed to eliminate network switches for reduced latency and power. | [PUBLIC] |
| Lenovo | Integrated server and HPC/AI infrastructure solutions. | Public company. | Strong OEM partnerships with GPU vendors and a global footprint in enterprise and research institutions. | [PUBLIC] |
The competitive map splits into three primary segments. The first comprises the entrenched incumbents: HPE, Dell, and Lenovo. These vendors sell integrated, fixed-configuration servers and offer their own composable software stacks, such as HPE Synergy. Their advantage is not technological superiority in composability but rather account control, global service networks, and procurement convenience for large enterprises. The second segment includes pure-play composability software and hardware challengers like Liqid and Rockport Networks. Liqid competes most directly on the composability promise but uses a software-led approach agnostic to the underlying interconnect. Rockport targets a different layer, focusing on network fabric innovation rather than PCIe-based resource pooling. GigaIO operates in the third segment: a hardware-defined fabric company. Its FabreX technology is a physical PCIe switch network, making it a deeper, more performance-oriented integration than software-only layers but also a more invasive infrastructure change.
GigaIO's defensible edge today is its architectural focus on PCIe as the unifying fabric. This provides a tangible performance claim, evidenced by its published benchmarks showing near-linear scaling in MLPerf tests and high efficiency in HPL-MxP [GigaIO, November 2024]. The integration is hardware-accelerated, which can offer lower latency and higher bandwidth for GPU-to-GPU and GPU-to-memory communication compared to software-emulated composability over Ethernet. This edge is durable if PCIe remains the dominant internal server interconnect for accelerators, which appears likely for the foreseeable future. However, it is perishable if a new interconnect standard (like CXL) gains rapid adoption for composability or if software-based solutions achieve comparable performance through clever orchestration. The strategic investment from memory giant SK hynix is a potential durability signal, suggesting alignment on future memory-centric architectures [Business Wire, September 2021].
The company's primary exposure is to the scale and ecosystem power of the incumbents. HPE or Dell could decide to bundle a comparable PCIe fabric solution into their next-generation server platforms, leveraging their existing customer relationships and distribution to outflank GigaIO. Furthermore, GigaIO's model requires customers to adopt its proprietary FabreX hardware, creating a vendor lock-in concern that the incumbents' broader, multi-vendor portfolios are designed to alleviate. The company also lacks the global sales and support footprint of its larger rivals, which could limit its ability to serve multinational enterprises. Its focus on high-performance, rack-scale deployments may also leave it vulnerable to cloud-centric alternatives where elasticity is provided by the cloud provider's own infrastructure, not an on-premises fabric.
The most plausible 18-month scenario hinges on the adoption curve for edge AI and the success of Gryf. If demand for portable, high-performance AI compute at the edge accelerates in defense and industrial sectors, GigaIO could establish a defensible beachhead with Gryf, which has already secured orders from the U.S. Department of Defense [GigaIO, April 2025]. In this case, GigaIO becomes the "winner if" edge AI supercomputing requires specialized, ruggedized, composable hardware that general-purpose server vendors are slow to build. The "loser if" scenario plays out if the data center composability market consolidates around software abstractions (like Kubernetes) or if CXL-based memory pooling becomes the primary composability layer, potentially sidelining PCIe-fabric specialists. In that outcome, Liqid's software-agile approach or an incumbent's integrated CXL roadmap could gain more traction, leaving GigaIO with a technologically elegant but niche solution.
Data Accuracy: YELLOW -- Competitor positioning and funding stages are based on public company data and industry categorization; GigaIO's specific differentiators are confirmed by company materials and press releases.
Opportunity
PUBLIC
If GigaIO can establish its PCIe-based fabric as a standard for composable AI infrastructure, the company is positioned to capture a significant share of a market valued at $120 billion [GigaIO]. The opportunity hinges on moving beyond selling discrete hardware units to becoming the enabling layer for a new, more efficient data center architecture.
The headline opportunity is for GigaIO to become the de facto standard for rack-scale composability in on-premise and edge AI deployments. This outcome is reachable not because of a single superior product, but because the company's approach solves a fundamental architectural bottleneck. The evidence lies in the performance of its fabric: the SuperNODE platform demonstrated near-linear scaling to 46,755 tokens per second in MLPerf Inference benchmarks, the highest result for a single node [GigaIO, November 2024]. This proves the underlying FabreX technology can deliver on the promise of eliminating I/O bottlenecks for the most demanding workloads. Furthermore, the strategic investment from memory giant SK hynix [Business Wire, September 2021] signals industry validation of the fabric's potential to optimize accelerator and memory utilization, a critical pain point in AI infrastructure.
Growth would likely follow one of several concrete paths, each with identifiable catalysts.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Defense & Intelligence Standard | Gryf becomes the default portable AI compute platform for tactical edge missions across the U.S. Department of Defense and allied agencies. | Securing a large-scale, multi-year production contract from a major defense prime contractor. | Gryf has already secured significant orders from the U.S. Department of Defense and intelligence community [GigaIO, April 2025], establishing an initial beachhead. |
| OEM/ODM Embed | FabreX technology is licensed and embedded into the server product lines of a major OEM like Dell or HPE, becoming a hidden but critical component of their AI offerings. | A formal technology partnership or joint development agreement announced with a top-tier infrastructure vendor. | GigaIO has already sold its SuperNODE platform and FabreX fabric to AI chip company d-Matrix [HPCwire, April 2026], demonstrating its technology's value as a licensable component for other hardware builders. |
| Hyperscale Edge Outpost | Major cloud providers adopt GigaIO's architecture for their hybrid cloud and edge outpost offerings, providing cloud-like composability in customer data centers. | A public proof-of-concept or design win with a cloud provider's edge infrastructure team. | The company's core narrative positions FabreX as delivering "cloud-class" elasticity on-premise [GigaIO], directly addressing a key hybrid cloud challenge for hyperscalers. |
Compounding for GigaIO would manifest as a software and ecosystem flywheel. Each new deployment of FabreX, whether in a SuperNODE, a Gryf unit, or an OEM design, expands the installed base of compatible hardware. This creates a larger market for the company's software-defined orchestration layer, which manages resource composition. As the software matures with more use cases and integrations, its value increases, making the hardware more effective and locking in customers through operational dependency. Early signs of this dynamic are visible in the product's design philosophy, which supports six different sleds for GPU, compute, storage, and networking, all managed through a single fabric [GigaIO]. This modularity encourages customers to standardize on GigaIO's architecture for future expansion, turning an initial node purchase into a platform commitment.
The size of the win, should the OEM embed or defense standard scenarios play out, is substantial. A credible comparable is the 2021 acquisition of composable infrastructure pioneer Liqid (a named competitor) by data center hardware firm Enfabrica for a reported $400 million. For GigaIO, which has also developed a strategic edge product with defense traction, a successful execution could command a similar or greater premium as a strategic asset. In a category-defining outcome where its fabric becomes widely adopted, the company's value could approach a meaningful single-digit percentage of the broader $120 billion AI infrastructure market (scenario, not a forecast). The more plausible near-term outcome is a high-value acquisition by a strategic player seeking to own the composable layer for the next generation of AI data centers.
Data Accuracy: GREEN -- Core market size, performance benchmarks, defense orders, and the d-Matrix sale are confirmed by company press releases and industry publications.
Sources
PUBLIC
[GigaIO] About Us - GigaIO is to Hardware What VMware is to Software | https://gigaio.com/about-us/
[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
[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/
[Crunchbase] GigaIO - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/gigaio
[HPCwire, 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/
[TheCompanyCheck] GigaIO Company Profile | https://www.thecompanycheck.com/company/gigaio
[LinkedIn] GigaIO LinkedIn Profile | https://www.linkedin.com/company/gigaio
[GigaIO, November 2024] SuperNODE Achieves Near-Linear Scaling in MLPerf Inference Benchmarks | https://gigaio.com/2024/11/supernode-mlperf-inference-benchmarks/
[HPCwire] GigaIO SuperNODE MLPerf Performance Coverage | https://www.hpcwire.com/
[HPCwire, April 2026] GigaIO Sells SuperNODE Platform and FabreX Fabric to d-Matrix | https://www.hpcwire.com/2026/04/gigaio-sells-supernode-fabrex-d-matrix/
[BigDATAwire, April 2026] d-Matrix Acquires GigaIO's AI Fabric Technology | https://bigdatawire.com/2026/04/d-matrix-acquires-gigaio-ai-fabric/
[Grand View Research, 2024] Artificial Intelligence Infrastructure Market Size Report | https://www.grandviewresearch.com/industry-analysis/artificial-intelligence-infrastructure-market
[MarketsandMarkets, 2024] AI Infrastructure Market - Global Forecast to 2029 | https://www.marketsandmarkets.com/Market-Reports/ai-infrastructure-market-152548021.html
[Hyperion Research, 2024] HPC Market Update and Forecast | https://hyperionresearch.com/category/hpc-market/
[ABI Research, 2024] Edge AI Inference Market Forecast | https://www.abiresearch.com/market-research/product/edge-ai-inference-market-forecast/
Articles about GigaIO
- 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.