Etched Raises $300M with $1B in Pre-Orders

2026-07-23 · Source: Big Data & AI News - EE Times · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cloud Computing & IT Infrastructure · Depth: Intermediate, medium

Summary

AI chip startup Etched has secured \$300 million in a Series C funding round, achieving a \$10 billion pre-money valuation and bringing its total raised capital to \$1.1 billion. The round was led by Sequoia, with participation from A16Z, Jane Street, SK Hynix, and Diffusion Capital. Etched already holds \$1 billion in pre-orders for its AI racks, which are slated to begin shipping this summer. The company targets major AI firms and clusters, claiming its hardware delivers "significantly better" performance for workloads like coding, long context, and long horizon agents. Etched's technical approach includes a vertically integrated design, encompassing everything from ASIC to cooling, and a novel low-voltage inference (LVI) scheme. LVI runs math engines at under half the voltage of competitors, enabling over 80% utilization for trillion-parameter MoE models without thermal throttling, potentially doubling or tripling global inference capacity. Its full-reticle TSMC N4P chip features six HBM stacks and a custom cluster-scale shared memory pool.

Key takeaway

For AI Architects evaluating next-generation inference hardware, Etched's \$1 billion in pre-orders and claimed "significantly better" performance warrant attention. Your teams should investigate their low-voltage inference (LVI) technology, which promises over 80% utilization for large MoE models without thermal throttling. This could double or triple your inference capacity per watt. Consider piloting Etched's racks, shipping this summer, especially for long context and agentic AI workloads, to assess their vertical integration benefits and cluster-scale memory.

Key insights

Etched's low-voltage inference and vertical integration aim to significantly boost AI inference capacity and utilization.

Principles

• Vertical integration optimizes performance across the stack.
• Low-voltage inference enhances FLOPS per Watt and utilization.
• Cluster-scale shared memory improves data access efficiency.

Method

Etched employs a vertically integrated approach, designing ASICs, packaging, boards, cooling, and servers in-house. It uses low-voltage inference for math engines and a custom interconnect for cluster-wide shared memory.

In practice

• Deploy Etched racks for high-utilization MoE inference.
• Consider disaggregated setups with Etched hardware.
• Evaluate LVI for energy-efficient, high-throughput AI.

Topics

• AI Hardware
• Inference Accelerators
• Low-Voltage Inference
• Vertical Integration
• Cluster Computing
• MoE Models

Best for: MLOps Engineer, NLP Engineer, CTO, AI Hardware Engineer, AI Architect, Investor

Related on AIssential

• Etched - Building AI Hardware to Make Inference Faster and Cheaper - [Invest Like the Best, EP.480] — Etched's architecture leverages low voltage inference and cluster scale memory to achieve significantly faster and cheaper AI model inference.
• Nvidia competitor Etched hits $5B valuation, $1B in sales for AI chip — Etched redefines AI chip design by optimizing for inference through low voltage and cluster-scale memory, enabling massive throughput.
• Partnering with Etched: Building the Inference Machine — Etched is building production-ready, cluster-scale custom silicon for AI inference, achieving superior throughput and interactivity.
• Etched Defies Skeptics With $10.3B Valuation As AI Chip Startup Closes $300M Series C - Bitcoin World — Specialized AI inference hardware for transformer models is gaining significant market validation.
• Running 26B and 35B LLMs at Full Speed on €990 of Used Hardware — No Cloud Required — Cost-effective local LLM inference is achievable with dual secondhand GPUs, outperforming single high-end cards and revealing speculative decoding's limitations on PCIe.
• One Sohu Chip Replaces 160 Nvidia GPUs. That Math Just Got a $5 Billion Valuation. — Specialized ASICs for transformer inference offer significant cost and power efficiency over general-purpose GPUs, albeit with architectural lock-in.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.