Etched: Meet the 2 Harvard dropouts that just raised $120M to compete with Nvidia

In partnership with

(4 minutes)

Hi 👋, this is the Today in AI Newsletter: The weekly newsletter bringing you one step closer to building your own startup.

We analyze a cool, industry-shaping AI startup every week, with a full breakdown of what they do, how they make money, how much they’ve raised, and the opportunity ahead. 

Let’s get to the good stuff in this email: 

• This AI startup is betting that Transformer models will be better than Nvidia’s GPUs

• Their chips are promised to have a 20x higher throughput than Nvidia’s H100 Chips

• They’ve just raised a $120M Series A led by Positive Sum and Primary Venture Partners

So what’s the startup and who are the founders behind it? Here’s the story of Etched 📈

AI models are often trained on graphics processing units (GPUs), which can be programmed to be used in complex use cases, but they can consume a lot of power, exponentially raising costs. 💰

This is what Etched is trying to solve. 💡 

Etched was founded in June 2022 by Gavin Uberti and Chris Zu to transform AI chips with a single type of architecture: Transformers. 🤖 

This is the same type of architecture that powers a lot of today’s biggest models including OpenAI’s ChatGPT and Sora.

Unlike any other company on the market, Etched chips can only run on transformers.

This means that if transformers change dramatically or go away completely, Etched will get wiped out completely. But if they’re right and transformers keep being the dominant way that AI models work, their chip will be the most performant on the market by an order of magnitude.

Gavin says that if this works out in their favor, Etched can be the world’s biggest company…

How does it work? 🧐

Etched’s flagship chip: Sohu, is able to burn the transformer algorithm into the chip. They’re building both their own silicon and their own servers, which enables them to gain a 20x higher throughput for AI models like ChatGPT or Meta’s Llama. 🚀

This chip will also be able to get a much better latency than current chips like Nvidia’s H100s - it’ll be able to cut time to the first token by a factor of 2. ☄️

The transformer models Etched is using can not only be used for LLMs, they work better than other models for image generation and videos too. 🎞️

Since GPT-2 in 2019, transformer models have pretty much stayed exactly the same, but the scale it’s done at is completely different.

OpenAI’s GPT-2 model used 1.5B parameters, while the latest GPT4 uses 1.8T parameters - Etched is betting that this will only get bigger. 📈

Their chip is promised to deliver on:

1. Only one core: Transformers

2. Being Expansible to 100T+ Parameters

3. 144 GB HBM3E per chip

4. Fully open-source stack

So far, Etched has partnered with TSMC - the world’s largest semiconductor foundry and has been working on their 4 nano-meter nodes (the same ones used in Nvidia’s Blackwell GPUs). 🦾

According to Gavin, Etched already has pre-orders for tens of millions of their chips. 💸 

Backstory 👀

Back in high school, Gavin dipped his toes in lots of extracurriculars including being on the math team and participating in robotics challenges.

His math team ended up placing 1st worldwide in Purple Comet math contest, and he had the world record for the highest score at the FTC Robotics challenge. 🏆 

This set him up pretty nicely to end up at Harvard in 2020, where he chose to study Math and Computer Science. 👨‍🎓 This is where he met Chris Zu. 

Chris had also been studying Math and CS, which the 2 quickly bonded over, as well as a passion for building cool things.

They had been taking classes on how to operate models at scale and how to network computers efficiently, and even did some reading on their own. They got hooked.

This is when they came up with Etched, designing more specialized, less power-intensive chips for running Gen-AI models. 💪

Up until March 2023, Etched had been a dorm room project where Chris was on campus and Zoom called every other day with Gavin, who’d been finishing up his gap year.

They couldn’t balance both school and working on Etched, so decided to double down on their startup and knew they’d regret it if they didn’t try. 💯 

Just one month later, Etched raised a $5.36M seed round led by Primary Venture Partners, dropped out of school and moved to the Bay Area to go all in on this.

Fast forward to March 2024, Gavin, Chris and their COO Robert Watchen, were awarded the Thiel Fellowship. 🎖️ They were, and still are one of the only teams in history where an entire founding team received the fellowship.

💰 On June 25th 2024, Etched raised a $120M Series A led by Positive Sum and Primary Venture Partners with participation from Peter Thiel and Amjad Masad (Co-founder of Replit), among many others.

Before you go 👋

Use AI as Your Personal Assistant

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI technology into your processes, allowing you to optimize resource allocation and maximize output with precision and ease.

Download Now!

Thank you for reading.

Know someone who'd love this newsletter? Share the love. 🌟