/Tech7h ago

AI startup Etched exits stealth with $800M in funding for its Sohu chip, which hardcodes LLM attention into silicon

Story Overview

Etched has stepped out of stealth with $800 million raised across multiple rounds to advance its Sohu ASIC, a chip built specifically for transformer-based LLM inference, while also reporting more than $1 billion in customer contracts and plans to ship its first full inference racks this summer after a completed A0 tapeout.

53344326649.9K

Original post

Patrick OShaughnessy@patrick_oshag#571inTech

Three years ago, two Harvard dropouts set out to build a better AI chip than the largest companies in the world.

Almost everyone I called at the time said it was impossible.

Today, Etched (@Etched) comes out of stealth with $800M total raised, $1B in signed customer contracts, and a working next-gen AI chip.

This was my excuse to ask the two founders, @UbertiGavin and @robertwachen, every question I have about compute and inference.

We discuss: - Why they built an entire rack and not just a chip - The two technical bets behind their architecture no one else has tried - How two founders in their twenties recruited industry legends - The night they nearly ran out of money - Why whoever produces the most tokens wins

If you care about the future of compute, Gavin and Rob are two people to know. I think you will find the story of what they have built hard to forget.

Enjoy!

TIMESTAMPS 0:00 Intro 1:00 Why Nobody Believed Etched Would Work 14:06 Why Inference Is the Bottleneck 22:27 Gavin and Rob’s Origin Stories 33:24 Taking Huge Risks to Move Faster 49:43 Kernels, Compilers, and the AI Stack 1:02:08 Raising $100M to Survive 1:16:00 The Future of Models, Agents, and Intelligence

8:00 AM · Jun 30, 2026 · 2.7K Views

Contract Momentum

Early contracts already exceed total funding

The company states it has secured over $1 billion in deals and kicked off production to fulfill them, with early customer tests highlighting throughput, latency, and power metrics on inference workloads.

Architecture Focus

Specialized hardware bets on transformer dominance

Sohu targets only transformer models through architectural choices like low-voltage inference and cluster-scale memory, though independent benchmarks confirming claimed advantages over general-purpose GPUs remain unavailable in public sources.

Sentiment

Many users praised Etched's fast AI inference chip development and $1B contracts as a strong hardware bet, while some called the claims snake oil or insulted associated figures.

Pos

80.0%

Neg

20.0%

7 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS9.9KBOOKMARKS20LIKES72REPLIES15

Bryan Johnson@bryan_johnson

Breaking news for people who want to look hot, be young and not die.   A few years ago, two college dropouts told me they could accelerate longevity by building a faster AI chip.   I invested, and they just pulled it off.  What it means:  > 10x more throughput (tokens per second) for the same power footprint

> Dramatically lower operational costs for executing today’s frontier models

> Run far larger, more capable AI models within the same power and thermal budget, because a transformer-specific chip spends a fraction of the energy per token that a general-purpose GPU does

Rob and Gavin's approach resonated with me because solving aging is a gigantic combinatorial search problem. The chemical space of small, drug-like molecules has around 10^60 possibilities.

These compounds need to be mapped against a human proteome derived from 20,000 genes, including 1,600 transcription factors, and a dense web of interactions among them.   The size of the combinatorial space is problematic. You need to identify which targets to modulate, within specific cellular lineages, at exact dosages, and in optimal temporal sequences.  Traditional high precision physics simulations are too slow to brute force the problem. You can shortcut it with AI inference, using frontier neural networks as hyper fast surrogate models to predict biological interactions instantly. By hardwiring transformer logic into silicon, Etched offers the infrastructure needed to run these massive biological foundation models at scale.  I'm surprised and impressed they were able to pull this off, and so quickly.

They already have $1B in orders

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

6h9.9K7220

RETWEETS7

Cory Levy@cory

congrats to the Etched team on coming out of stealth with a working chip and over $1 billion in signed customer contracts!

and $800M raised at a $5B valuation

proud to be their very first believer and investor in 2022 while they were still students at Harvard

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

3h13.2K19331

Matt Shumer@mattshumer_

Nvidia is worth $5T because the world realized inference is the new oil.

If Etched’s chips are meaningfully better than Nvidia’s for inference…

What is Etched worth?

This is a worldview-shattering event.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h8.4K245

Matt Shumer@mattshumer_

Nvidia has some serious competition coming their way.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h5.8K185

Chubby♨️@kimmonismus

Etched is coming out of stealth with $800M raised, $1B+ in customer contracts, first racks shipping this summer, and claims of SOTA inference throughput, latency, and power efficiency.

But holy, look who backed the funding. The who-is-who if AI reseracher and VC.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h1.3K113

Patrick OShaughnessy@patrick_oshag

Inference will never be the same: Etched invented two new ways to massively improve compute cost, speed, and per watt efficiency: low voltage inference (more FLOPs) and cluster scale memory (memory/bandwidth)

The combination runs trillion-parameter models at over 80% MFU, where today's GPUs deliver 20 to 50%.

"Everyone said you can't run at voltages lower than GPUs. That was dissatisfying, because plenty of other chips already do.

Bitcoin miners run at under a quarter of the voltage of GPUs. We found a new mechanism to run at much lower voltages, and we think all AI chips in the future will be low voltage chips.

People ask how much memory bandwidth is on your chip. You should be asking how much is on your full scale-up cluster.

We added far more bandwidth at much lower latency from chip to chip."

Patrick OShaughnessy@patrick_oshag

Three years ago, two Harvard dropouts set out to build a better AI chip than the largest companies in the world.

Almost everyone I called at the time said it was impossible.

Today, Etched (@Etched) comes out of stealth with $800M total raised, $1B in signed customer contracts, and a working next-gen AI chip.

This was my excuse to ask the two founders, @UbertiGavin and @robertwachen, every question I have about compute and inference.

If you care about the future of compute, Gavin and Rob are two people to know. I think you will find the story of what they have built hard to forget.

Enjoy!

3h1.5K82

Beff (e/acc)@beffjezos

Etched had a very clear thesis about the landscape of algorithms years ago and they made a bold bet, executed relentlessly while pushing the boundaries of what is possible in the digital substrate.

Visited their operations months ago and was thoroughly impressed. Kudos to the team and congrats on the huge milestone!

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

6h1.8K152

Tim Sweeney@TimSweeneyEpic

It's exciting to watch new computing architectures emerge from the high-memory-size, high-memory-bandwidth, high-compute, high-communication needs of AI. We've been in somewhat in the innovation doldrums with commodity PC architectures eating everything.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

3h2.2K140

Tri Dao@tri_dao

It's wild how quickly Etched designed and got the chips out, all within 2 years. They went deep, hardcoding attention into silicon and getting very high MFU. This kind of hardware tailored made for LLM inference is soon gonna bring cost of intelligence down 10x

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h560161

"Leigh Marie" Braswell@LM_Braswell

Congrats to my friends & the team at @Etched! Awesome progress solving one of the most important bottlenecks in AI.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h653181

Iain Dunning@iaindunning

Congrats to Etched! One of our @HRTVentures investments, more to come.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

7h555121

elie@eliebakouch

what is the pareto frontier they are optimizing for? my understanding is that they both work on latency (new shared memory) and throughput (more flops per joule?)

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

5h41341

Matt Shumer@mattshumer_

I’m a very early investor in Etched, as I’ve believed for years that inference will be the biggest market we’ve ever seen.

Incredibly impressed by the team’s execution and so excited to use models powered by their hardware!

Matt Shumer@mattshumer_

Nvidia is worth $5T because the world realized inference is the new oil.

If Etched’s chips are meaningfully better than Nvidia’s for inference…

What is Etched worth?

This is a worldview-shattering event.

7h2.2K60

Packy McCormick@packyM

same person who said this about Intel ~30% ago said that etched was going to successfully complete its first tapeout and would become immensely valuable when it did.

spooky.

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

4h71251

Altcoinist@Altcoinist

@bryan_johnson Bryan Johnson & Ribbit bullish confluence

6h2149

Sigil Wen@0xSigil

blazingly fast execution (and inference)

congratulations @robertwachen, @UbertiGavin and the @Etched team on taping out the first chip!

honored to be an early investor :)

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

5h1.2K110

Clive Chan@itsclivetime

impressively fast execution from the etched team after a rocky start. hopefully it works well. congratulations!

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

6h35270

Chubby♨️@kimmonismus

*of AI reseracher, typo

Chubby♨️@kimmonismus

Etched is coming out of stealth with $800M raised, $1B+ in customer contracts, first racks shipping this summer, and claims of SOTA inference throughput, latency, and power efficiency.

But holy, look who backed the funding. The who-is-who if AI reseracher and VC.

7h1.8K130

Anjney Midha@AnjneyMidha

very cool

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

5h56741

andrew gao@itsandrewgao

Cooking

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

6h15511