Nvidia rival AI chip startup Etched raises $500m

Etched, a San Jose-based AI chip startup, has raised about US$500 million in a new funding round, according to sources familiar with the matter.

The funding round was led by investment firm Stripes and included participation from Peter Thiel, Positive Sum, and Ribbit Capital, the sources said.

This deal reportedly values Etched at US$5 billion and brings its total fundraising to nearly US$1 billion.

Etched is developing an AI chip called Sohu and is working with Taiwan Semiconductor Manufacturing Co.’s Emerging Businesses Group for production.

Etched, Stripes, Positive Sum, and Ribbit Capital declined to comment on the funding round.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

A $5B price tag needs near-term delivery. Tapeout status (final sign-off that sends a chip design to manufacturing), production timing, and third-party benchmarks versus Nvidia’s B200 are undisclosed 1.
A claim that Sohu delivers 20x faster inference than Nvidia H100 data center GPUs, and that one server replaces 160 H100s, lacks third-party or standardized tests 1.
Etched works with Taiwan Semiconductor Manufacturing Co. (TSMC) on 4nm and has High Bandwidth Memory (HBM), but no sampling dates or customer deployments are public, so investors face multi-year R&D risk over near-term revenue 2.
If transformer architectures shift in a big way, which CEO Gavin Uberti has acknowledged could make Etched obsolete 3, defending a $5B value gets harder given long semiconductor build cycles 2.

MLOps (Machine Learning Operations) plus inference platform vendors should integrate Etched’s Software Development Kit (SDK) with its compilers to support customers seeking options beyond Nvidia’s 70-95% share 1.
Specialized clouds such as CoreWeave or Lambda Labs could differentiate by offering Etched-optimized infrastructure 4.
Tooling firms that push model portability with abstraction layers can win as customers hedge GPU reliance, which opens support for Nvidia and AMD plus chips like Sohu 4.
Early adopters in inference-as-a-service platforms (cloud services that run AI models on demand) like Together AI or Fireworks.ai could gain pricing power if Sohu’s claims hold, which may let them undercut rivals tied to Nvidia 4.