Rivos, a chip startup based in Santa Clara, is seeking to raise between US$400 million and US$500 million at a valuation above US$2 billion, according to sources.

Founded in 2021, Rivos is developing a graphics processing unit (GPU) designed for AI inference and completed the physical design of its chip earlier this year.

If the new funding round closes, Rivos’ total funding would exceed US$870 million, making it one of the best-funded chip startups yet to mass-produce a product.

Currently, the company has raised over US$370 million so far.

Rivos employs more than 450 people and is working on software to help customers run Nvidia CUDA-based machine-learning models on its chips, sources said.

🔗 Source: The Information

🧠 Food for thought

1️⃣ AI chip startups require unprecedented capital levels before generating revenue

Rivos’ pursuit of $500 million demonstrates the extreme capital intensity facing AI chip companies, with funding requirements far exceeding most other startup categories.

The company has already raised over $370 million since 2021 and still needs hundreds of millions more before mass-producing a single chip, highlighting the enormous upfront investment required in this sector.

This trend extends across the industry: Cerebras Systems raised $25 million and reached a $900 million valuation by 2017 despite having no product on the market, while Graphcore secured $80 million in funding that same year under similar circumstances 1.

The capital requirements reflect the fundamental realities of chip development. Rivos employs over 450 people and faces expenses for specialized engineering teams, expensive design software tools, and costly manufacturing partnerships with companies like TSMC.

Even after completing their chip design milestone called “tape-out” earlier this year, Rivos still needs substantial funding to scale manufacturing and compete effectively, illustrating how chip startups must sustain massive burn rates for years before seeing meaningful revenue.

2️⃣ Breaking Nvidia’s software moat remains the critical challenge for chip competitors

Rivos is developing software to translate Nvidia’s CUDA code to run on its own chips, tackling the same fundamental challenge that has stymied many AI chip startups attempting to compete with Nvidia’s integrated hardware-software approach.

The article notes that many AI chip startups have “made similar promises, only to fall short of building software capable of competing effectively with Nvidia’s tightly integrated hardware and software.”