AI Hardware Startup Etched Reaches $21 Billion Valuation Following $700 Million Funding Round
Artificial intelligence chipmaker Etched announced on Tuesday that it secured $700 million in a fresh funding round, lifting the company’s valuation to $21 billion. The rapid capital infusion was led by quantitative trading firm Jane Street, following evaluations and purchases of the startup’s specialised hardware.
What Happened
The latest funding marks a steep rise in Etched’s corporate valuation over a short period. The company held a $5 billion valuation in December before closing a $300 million Series C round at a $10.3 billion valuation in July. With the addition of the $700 million round, its valuation effectively doubled within a single month, rising by nearly $11 billion.
The financing round was anchored by Jane Street, which conducted tests on Etched’s hardware before deploying it within its own data centres. Other backing investors include prominent firms and individuals such as Sequoia Capital, Kleiner Perkins, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone.
Key Highlights
- Massive Valuation Growth: The startup’s market valuation climbed from $5 billion in December to $10.3 billion in July, reaching $21 billion following the latest $700 million raise.
- Proprietary Architecture: Etched designs complete systems designated as “frontier inference clusters,” competing with full-system architectures such as Nvidia’s AI factories.
- Two-Stage Inference Architecture: Co-founder and Chief Operating Officer Robert Wachen explained that the hardware breaks inference down into separate “prefill” and “decode” phases.
- Prefill Hardware: Etched developed a low-voltage prefill chip designed to integrate more transistors while managing heat constraints, processing input tokens more rapidly.
- Cluster-Scale Memory: For the output token decode phase, the company engineered high-speed interconnects and a shared memory pool designed to maintain low latency across multiple chips.
- Model Flexibility: Etched clarified that its hardware is not locked to a single specific artificial intelligence architecture and can execute any frontier model.
Why This Matters
Inference refers to the computing stage that processes user prompts and returns answers. The process requires substantial computational resources during prompt interpretation (prefill) and heavy memory bandwidth during response generation (decode). Etched designed separate hardware solutions for each stage from scratch, promising customers increased execution speeds alongside reduced operational costs.
Addressing misconceptions from the startup’s early development, Etched noted that its current systems are no longer limited to running one custom-etched model. The systems can instead serve any frontier model, allowing enterprises to adapt the infrastructure to varied workloads.
What to Watch Next
Following the funding and product testing, customers such as Jane Street are actively deploying the technology. Jane Street reported receiving positive early test results and confirmed running an Etched rack directly inside its data centre to handle demanding computational workloads. The wider adoption and operational performance of Etched’s frontier inference clusters across customer facilities will remain a central development to observe.
Frequently Asked Questions
Who led the latest funding round in Etched?
The $700 million round was led by quantitative trading and investment firm Jane Street, alongside participation from several venture capital and private equity investors.
How does Etched handle artificial intelligence inference?
Etched addresses inference by dividing it into two steps: a compute-heavy prefill phase that uses a specialised low-voltage chip to read prompts, and a memory-heavy decode phase supported by cluster-scale memory and low-latency interconnects to generate outputs.
Can Etched’s chips run different artificial intelligence models?
Yes. While Etched originally aimed to build chips tailored to a single model, the company clarified that its current hardware can run any frontier model.
Source: TechCrunch report on Etched company announcements.