Descriptions:
AI chip startup Etched emerged from two years of stealth operation with an $800 million funding round at a reported $5 billion valuation, backed by quantitative trading giant Jane Street and TSMC-linked Venture Tech Alliance. Bloomberg Technology’s Ed Ludlow interviews CEO Gavin Uberti about the company’s rack-scale inference system — a pre-assembled 32-chip rack built specifically for AI inference rather than training, targeting what Etched believes is the dominant workload as the industry shifts from building models to running them at scale.
The system’s two core technologies are low voltage inference and cluster-scale memory. Low voltage inference runs chips at under half the voltage of typical NVIDIA GPUs, dramatically cutting power consumption and enabling more concurrent users on the same hardware. Cluster-scale memory allows each chip to access the HBM and SRAM of neighboring chips in the rack, reducing memory bottlenecks for large models and improving per-token economics. Uberti frames the combined effect as a step-change improvement in cost-per-user versus existing GPU infrastructure, particularly as model sizes continue to grow.
While acknowledging that NVIDIA still holds what he terms a technical monopoly, Uberti says Etched is seeing best-in-class benchmark results in lab conditions and has begun onboarding early customers for real-world testing. Jane Street and Venture Tech Alliance were cited as backers capable of evaluating the technology at a deep first-principles level, lending credibility to the company’s technical claims at a historically early stage.
📺 Source: Bloomberg Technology · Published June 30, 2026
🏷️ Format: Interview
