Etched builds frontier inference clusters powered by Sohu, a transformer-specific ASIC designed to run transformer inference faster and cheaper than general-purpose GPUs . The pitch to buyers is narrow by design. The chip is specialized for AI inference and makes deliberate trade-offs, such as not supporting high-precision data types or training , which is how the team argues it can beat GPUs on the workloads that actually matter for serving large models in production. The target customer is any operator running heavy transformer inference at scale, from trading firms to model providers.
This launch marks the moment the company moved from roadmap to revenue. Etched announced $700 million in new funding in a round led by Jane Street with participation from Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, Neo, Primary, Stripes, Positive Sum and Blackstone at a $21B valuation , and paired the announcement with news that the first production rack has been delivered to Jane Street. The round was led by Jane Street after the famed quant fund tested and bought the startup's AI hardware , which turns a lead investor into a live reference customer.
The company was founded by Gavin Uberti, Robert Wachen, and Chris Zhu , with Uberti and Zhu being Harvard dropouts who started the company in 2022. For founders and operators tracking the inference layer, the signal here is a specialized silicon bet backed by a real shipment and roughly more than $1 billion in booked orders , giving the transformer-ASIC thesis its first serious commercial validation against incumbent GPU stacks.
Comments (8)
Sign in to join the discussion.
P
Priya KothariAug 18, 2026
Shipping the first rack to a prop shop that also led the round is very 'the customer is on the cap table' energy. I respect the vertical integration of vibes.
D
Denis V.Aug 18, 2026
An ASIC bet only pays off if the transformer architecture doesn't drift under you. Curious how they're handling attention variants that weren't on the roadmap in 2022.
T
Tobias F.Aug 18, 2026
The reveal shot of the rack rolling in is genuinely cinematic. Whoever cut that video understood that silicon has to look like a weapon, not a filing cabinet.
W
wanjiruAug 18, 2026
Two sentences, one photo, seven logos. This is how you tweet a raise without sounding like a LinkedIn post.
M
Mora GaskellAug 18, 2026
Every hyperscaler is already taping out custom silicon internally. Selling frontier inference as a product to people who could build it themselves is either genius or a very expensive hobby.
F
Farrukh A.Aug 18, 2026
'Shipped our first rack' is doing so much heavy lifting here. One rack is a demo, a hundred is a business, come talk to me at ten thousand.
J
Juno T.Aug 18, 2026
Love that the launch copy just lists the investors and moves on. No manifesto, no 47-tweet thread about why compute is destiny. Restraint is underrated.
L
Leland OkaforAug 18, 2026
The market rewards people who bet on a specific future hard enough to pour concrete around it. This is that bet, whether or not it works.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Etched
Frontier Inference Clusters
Frontier Inference Clusters
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Etched builds frontier inference clusters powered by Sohu, a transformer-specific ASIC designed to run transformer inference faster and cheaper than general-purpose GPUs . The pitch to buyers is narrow by design. The chip is specialized for AI inference and makes deliberate trade-offs, such as not supporting high-precision data types or training , which is how the team argues it can beat GPUs on the workloads that actually matter for serving large models in production. The target customer is any operator running heavy transformer inference at scale, from trading firms to model providers. This launch marks the moment the company moved from roadmap to revenue. Etched announced $700 million in new funding in a round led by Jane Street with participation from Kleiner Perkins, Sequoia, Andreessen Horowitz, Tiger Global, Bain Capital Ventures, Neo, Primary, Stripes, Positive Sum and Blackstone at a $21B valuation , and paired the announcement with news that the first production rack has been delivered to Jane Street. The round was led by Jane Street after the famed quant fund tested and bought the startup's AI hardware , which turns a lead investor into a live reference customer. The company was founded by Gavin Uberti, Robert Wachen, and Chris Zhu , with Uberti and Zhu being Harvard dropouts who started the company in 2022. For founders and operators tracking the inference layer, the signal here is a specialized silicon bet backed by a real shipment and roughly more than $1 billion in booked orders , giving the transformer-ASIC thesis its first serious commercial validation against incumbent GPU stacks.
Comments (8)
Shipping the first rack to a prop shop that also led the round is very 'the customer is on the cap table' energy. I respect the vertical integration of vibes.
An ASIC bet only pays off if the transformer architecture doesn't drift under you. Curious how they're handling attention variants that weren't on the roadmap in 2022.
The reveal shot of the rack rolling in is genuinely cinematic. Whoever cut that video understood that silicon has to look like a weapon, not a filing cabinet.
Two sentences, one photo, seven logos. This is how you tweet a raise without sounding like a LinkedIn post.
Every hyperscaler is already taping out custom silicon internally. Selling frontier inference as a product to people who could build it themselves is either genius or a very expensive hobby.
'Shipped our first rack' is doing so much heavy lifting here. One rack is a demo, a hundred is a business, come talk to me at ten thousand.
Love that the launch copy just lists the investors and moves on. No manifesto, no 47-tweet thread about why compute is destiny. Restraint is underrated.
The market rewards people who bet on a specific future hard enough to pour concrete around it. This is that bet, whether or not it works.