d-Matrix launched JetStream, a new I/O accelerator designed to reduce latency in large-scale AI inference deployments. The Santa Clara–based company positioned JetStream as a key enabler for scaling AI services that are increasingly moving from training toward real-time reasoning, multi-modal, and agentic workloads.
JetStream integrates with d-Matrix’s Corsair compute accelerators and Aviator software, providing a platform that spans compute, software, and networking. The company claims performance gains of up to 10x speed, 3x better cost-performance, and 3x greater energy efficiency compared to GPU-based solutions when running models exceeding 100 billion parameters. Packaged as a PCIe Gen5 NIC delivering 400Gbps, JetStream is compatible with standard Ethernet switches, allowing deployment without wholesale data center upgrades.
With samples available now and full production expected by year-end, d-Matrix joins a small set of AI infrastructure providers offering end-to-end compute, I/O, and software solutions. The JetStream launch underscores the increasing importance of networking efficiency as AI workloads scale in size and interactivity.
- JetStream NICs deliver up to 400Gbps bandwidth per card over PCIe Gen5
- Optimized for d-Matrix Corsair compute accelerators and Aviator software stack
- Transparent NIC and streaming solution designed specifically for AI inference
- Works with standard Ethernet switches for seamless data center integration
- Enables ultra-low latency inference on models with 100B+ parameters
- Full-height PCIe form factor; samples shipping now, production in late 2025
“JetStream networking comes at a time when AI is going multimodal, and users are demanding hyper-fast levels of interactivity,” said Sid Sheth, co-founder and CEO of d-Matrix.
🌐 Analysis: JetStream extends d-Matrix’s “full-stack” approach by addressing a key bottleneck in AI inference — data movement between accelerators and the network. The NIC is purpose-built for AI inference scaling, rather than adapted from general-purpose networking silicon. Packaged as a PCIe Gen5 card delivering 400Gbps, JetStream offers both throughput and ultra-low latency tailored to inference workloads where response time directly impacts user experience. By remaining compatible with off-the-shelf Ethernet switches, it lowers barriers to deployment compared to proprietary interconnects.
At the architectural level, JetStream integrates tightly with Corsair accelerators and the Aviator software layer, enabling inference pipelines that can handle models exceeding 100B parameters without the inefficiencies common to GPU-based deployments. Its streaming-focused design supports multimodal, reasoning-heavy AI services, which require rapid back-and-forth between compute, storage, and network. In this way, JetStream acts as the connective tissue that prevents compute advances from being stranded by I/O bottlenecks.
Strategically, this positions d-Matrix as one of the few companies delivering a complete inference platform — compute, software, and networking — optimized for the unique demands of the inference regime. Competitors such as Groq and Cerebras emphasize raw compute or compiler-driven optimization, while d-Matrix is betting on a vertically integrated stack where networking is not an afterthought. With samples already shipping and production targeted for year-end, d-Matrix is signaling to hyperscalers that it can be a credible alternative to GPU-based systems for latency-sensitive, multi-user AI services.
Founded in 2019 and headquartered in Santa Clara, d-Matrix is led by co-founder and CEO Sid Sheth, alongside co-founder and CTO Sudeep Bhoja. The company has raised over $200 million from investors including Playground Global, Microsoft’s M12, SK hynix, and Temasek. Key milestones include unveiling its digital in-memory compute (DIMC) technology for inference acceleration, the launch of its Corsair accelerator platform in 2023, and the rollout of its Aviator software in 2024. JetStream represents the latest step in its strategy to offer a fully integrated platform for AI inference at scale.
🌐 We’re tracking the latest developments in AI infrastructure. Follow our ongoing coverage at: https://convergedigest.com/category/ai-infrastructure/