DreamBig's core thesis was counterintuitive: instead of building a monolithic AI chip and competing with Broadcom, NVIDIA, and Intel on their own terms, Syed built a chiplet platform that let anyone plug into the AI infrastructure stack. Call it modular ambition.
The MARS Open Chiplet Platform - unveiled at CES 2024 - solved two problems simultaneously. Scale-up: getting more compute and memory inside a single AI accelerator system. Scale-out: connecting those systems to each other across a data center fabric. Most companies tackle one. DreamBig tackled both with a single chiplet architecture built around UCIe open standards, 3D HBM stacking, and RDMA Ethernet networking.
MARS stands for Modular Architecture for Rapid Scaling. The platform allows startups or any size company to build AI, data center, edge, and automotive solutions that were previously only accessible to trillion-dollar cloud providers.
In January 2025, DreamBig unveiled the Mercury AI-SuperNIC - a monolithic chip delivering up to 12.8 Tbps of bandwidth, designed to connect AI chips like GPUs and TPUs via a chiplet-compatible component. It was DreamBig's product moment: the thing that demonstrated the MARS platform wasn't just architecture slides. It shipped silicon.