Preface:

In early April, the AI infrastructure landscape witnessed a significant acquisition. d-Matrix, a trailblazer in generative AI inference computing, announced its acquisition of the data center division of Carlsbad, California-based GigaIO. This collaboration began in 2025 when d-Matrix integrated its Corsair inference platform into GigaIO's SuperNODE architecture, creating a hyperscale solution capable of supporting dozens of Corsair accelerators per single node.

Today, this transaction signifies the complete integration of GigaIO's FabreX PCIe memory fabric and SuperNODE platform into d-Matrix's product lineup. Founder and CEO Sid Sheth's vision is unequivocal: “Inference extends beyond any single chip—it's now a systems challenge.”

Author | Fang Wensan

Image Source | Network

From Single Chips to Rack-Level Infrastructure

What defines a 'full-system computing solution'? It signifies a shift in AI inference competition from focusing solely on the computational prowess of individual chips to encompassing end-to-end capabilities. This includes accelerators, networking, memory interconnects, software stacks, and even entire racks. This acquisition builds upon the collaboration initiated in 2025, aiming to bolster d-Matrix's capacity to deliver system-level AI infrastructure rather than isolated silicon components.

GigaIO's FabreX, a PCIe-based composable memory fabric, supports the creation of decoupled compute and memory pools across nodes, enabling dynamic configuration at the rack or cluster level. This technology seamlessly integrates with d-Matrix's existing Corsair inference accelerators, JetStream networking, Aviator software, and the SquadRack rack-level reference architecture, co-developed with Broadcom and Arista. From a broader industry standpoint, the full-system approach has gained consensus among leading enterprises. At the 2026 GTC Conference, NVIDIA's product form had already transitioned from standalone GPUs to integrated 'chip-rack-data center' systems, indicating a shift in computational power competition to data center-level platforms. d-Matrix's acquisition strategy aligns perfectly with this trend.

d-Matrix's Insightful Judgement: Memory Bandwidth as the Primary Bottleneck

d-Matrix has charted a technical path distinct from the GPU-centric approach. After NVIDIA established dominance in AI training in 2019, founder Sheth chose to focus on inference rather than training chips. “Venturing into that space without significant differentiation would be a futile endeavor,” he remarked.

d-Matrix's core insight is that for Transformer-based inference, the bottleneck has never been computation but rather the movement of weights. The primary source of latency arises from data shuttling between compute cores and memory. To mitigate this, they developed digital in-memory computing technology, where matrix multiplication occurs directly within memory cells, with memory blocks doubling as compute blocks. Summation operations are executed via embedded adder trees, offering a more efficient hardware solution for AI inference. The platform leverages SRAM instead of HBM, tailored specifically for Transformer workloads. Corsair integrates large-capacity SRAM and LPDDR5X within the chip, enabling matrix operations to occur as close to storage as possible, thereby reducing energy consumption and latency associated with data movement. Furthermore, d-Matrix plans to innovate in 3D DRAM memory stacking, expanding memory capacity into the third dimension and promising to enhance AI model running speeds by 10x while reducing energy consumption by up to 90% compared to the current industry-standard HBM4.

This foundational architectural reconstruction reflects a deep understanding of the core needs of inference scenarios. As d-Matrix articulates, they are addressing 'three major obstacles' to achieve swift, efficient, and high-performance AI inference, with memory bandwidth being the most critical. Sheth's statement clearly outlines the evolutionary logic behind the full-system approach: “We recognize the need for something unique, something more efficient—not just solving compute issues but also tackling compute, memory, memory bandwidth, memory capacity, and all related challenges.”

Market Indicators: Financing Patterns and Customer Targeting

d-Matrix's full-system strategy has garnered substantial capital support. In November 2025, the company secured a $275 million Series C funding round, reaching a valuation of $2 billion and accumulating $450 million in total financing. Participants included European tech investment firm Bullhound Capital, Singapore's sovereign wealth fund Temasek, Microsoft's venture capital fund M12, the Qatar Investment Authority, and EDBI. The involvement of these top-tier investment institutions serves as a strong endorsement of d-Matrix's technical roadmap and commercial prospects.

At the product level, the Corsair platform's performance metrics are impressive. It achieves a throughput of 30,000 Tokens/second with only 2ms latency per Token on the Llama 70B model; on the Llama 8B model, a single server can deliver extreme performance of 60,000 Tokens/second with 1ms latency per Token. Additionally, its solution is said to reduce interactive latency by up to 10x in performance mode compared to alternatives using HBM. Sheth claims that their solution outperforms GPUs by 2-3x in cost, 5-10x in energy efficiency, and nearly 10x in speed.

Target customers include hyperscale cloud providers, cutting-edge AI labs, and enterprise-level deployments. Partners such as supercomputers are actively bringing d-Matrix's solutions to market. Sheth anticipates that the acquisition will accelerate revenue growth and support new pricing models in rack configurations for heterogeneous systems.

The Turning Point for AI Inference and the Strategic Importance of the Full-System Approach

The significance of this acquisition lies in the structural transformation of demand within the AI industry. Deloitte projects that the global proportion of inference workloads in AI computational power will increase from about one-third in 2023 to approximately two-thirds by 2026. NVIDIA further highlights that global computational power demand has surged 1 million-fold over the past two years due to the rapid growth in inference tasks.

At this structural inflection point, full-system computing solutions offer distinct advantages. As inference workloads increasingly operate in a distributed and heterogeneous manner across CPUs, GPUs, and inference accelerators, data must move efficiently in real-time between chips, nodes, racks, and entire data centers. Enterprises with comprehensive system stacks can provide lower-latency, higher-energy-efficiency, and more cost-competitive solutions. Galaxy Securities explicitly states that the competition for computational power has shifted from the chip level to the data center-level platform. d-Matrix CEO Sheth succinctly puts it: “Inference transcends any single chip. It's now a systems challenge.”

Conclusion:

From the acquisition of GigaIO's data center business to the foundational breakthroughs in digital in-memory computing technology, and the structural surge in inference computational power demand, all indicators point to the future of AI inference lying in system-level holistic optimization. The 2026 acquisition merely marks the commencement of this systemic competition.

Online Sources:

Alibaba Cloud: 'Defining the 2026 Intelligent Computing Era: Deconstructing the Underlying Protocols for Enterprise-Level AI Applications Transitioning from 'Experimental' to 'Production' States'

Smart Finance: 'GF Securities: AI Inference Efficiency Innovation and Agent Resonance Unlock a Trillion-Dollar Market Space'

Sina Finance: 'Digital Economy Weekly: GTC2026 Highlights—AI Shifts from Chip Competition to System Competition'

China Science and Technology Network: 'The Year of Token Explosion! 2026 Zhongguancun Forum Annual Sub-Forum Discusses New Visions for Large-Scale AI Inference Services'