← Back to Startups List
Etched logo

Etched

etched.comadded 2026-09-02
Valuation pace is unusually fast: $5B to $21B in 8 months — investors are pricing the 'bet on the architecture' narrative far ahead of disclosed shipping or revenue milestones● Hiring

Worth-Joining Index · Scorecard

v2.1 ALPHA · 2026-08

Career Score

/100

Coverage

51%

Verified

17%

Stage cohort

Early

Vibe Check

🎪 community voting coming soon

Momentum
4/4
Efficiency
unknown
Backing
8/8
Market
7.7/11
Product
6.6/11
Upside
2/8

Latest valuation

$21B (2026-08 round)

Funding round · 2026-08

Total raised

~$1.5B total raised (as of 2026-08)

as of 2026-08

Team size

~400 employees

figures vary
AI芯片半导体推理硬件Founded · 2022Stage · Growth (Series C)HQ · San Jose, CASize · ~400 employees

Traction · Key numbers

Valuation$5B (Dec 2025) → $10.3B (Jul 2026) → $21B (Aug 2026) in 8 months Publicly verifiableas of 2026-08
Total raised~$1.5B total raised (as of Aug 2026) Publicly verifiableas of 2026-08
Headcount~400 employees~ 3rd-party estimateas of 2026-08

Funding

Total raised

~$1.5B total raised (as of 2026-08)

Rounds

4

Valuation pace

$5B to $21B in 8 months

$700M

Funding round · 2026-08 · valuation $21B (2026-08 round) · led by Jane Street

Valuation doubled again within one month

C

$300M

Series C · 2026-07 · valuation $10.3B post · led by Sequoia

With: Andreessen Horowitz, SK Hynix, Jane Street, Diffusion Capital · Valuation doubled in ~7 months; the highest valuation ever for a Sequoia-led Series C

$500M

Funding round · 2025-12 · valuation 约$5B · led by Stripes

With: Peter Thiel, Positive Sum, Ribbit Capital · Not explicitly labeled Series B in coverage

S

$120M

Seed/Early · 2024-06 · led by Positive Sum, Primary Venture Partners

Valuation pace is unusually fast: $5B to $21B in 8 months — investors are pricing the 'bet on the architecture' narrative far ahead of disclosed shipping or revenue milestones

Lead investors escalated from early VCs (Positive Sum/Primary) to Sequoia and now Jane Street — a quant giant writing a check signals broadening consensus on inference-compute scarcity, but also less tolerance for missed delivery milestones going forward

~$1.5B raised for a chip company still ramping tape-out/manufacturing is extreme capital intensity — if the Transformer-only bet breaks or scale-up slips, downside is concentrated

Founders

Gavin Uberti CEO

Harvard Thiel Fellow dropout with a background in AI compilers

Robert Wachen President

Harvard Thiel Fellow dropout, previously founded Prod

Chris Zhu Co-Founder

Harvard Thiel Fellow dropout with a mathematics research background

Reality · Working here

H1B / PERMUnknown
Remote policyUnknown
Comp modelUnknown
GlassdoorUnknown — work here? Help us fill this in →
Equity liquidityUnknown
Hiring statusCareers page lists 11+ open roles across hardware and software teams Company-claimedas of 2026-09

Signals & Risks

CompetitorsNvidiaGroqCerebras
ValuationValuation rose from $5B to $21B in about 8 months, far outpacing disclosed production and revenue progress, raising overvaluation risk
Architecture betThe chip only supports transformer architectures; a major shift in dominant model architecture would sharply erode its value
CompetitionCompetes directly against Nvidia's ecosystem moat and other inference-ASIC players like Groq and Cerebras; mass production and software ecosystem are the key tests

Editorial ✎

Good fit for

· Chip architects and ASIC/RTL engineers — a rare shot at building maximally specialized hardware for a single architecture, one of few teams besides Nvidia betting transformer-only

· Systems software/compiler engineers — the software stack for dedicated hardware is being built near from scratch, with room to define the whole interface

· People comfortable with a high-conviction, unproven bet — the narrative is clean but manufacturing scale and revenue are still unproven

Not for

· People who need near-term, verified cash-flow stability — valuation is running far ahead of verifiable manufacturing/revenue progress

· People uncomfortable with architecture-lock-in risk who'd rather bet on general-purpose hardware than a single-architecture ASIC

Ask in the interview

1. What is Sohu's actual production/shipment scale and customer list today, and does it match the pace of the valuation?

2. If a non-Transformer architecture (e.g. state-space models) becomes dominant, what's the contingency plan?

3. Compared to Nvidia's CUDA moat and inference-specialized rivals like Groq/Cerebras, how mature is Sohu's software ecosystem and developer lock-in?

Data changelog

Community corrections → editorial review → applied, fully on the record

2026-09-02
最近一轮领投机构 filled in as Jane Street

data as of 2026-08 · Source: MarketScale

Publicly verifiable
2026-09-02
最近一轮估值 filled in as $21B (2026-08 round)

data as of 2026-08 · Source: MarketScale

Publicly verifiable
2026-09-02
累计融资额 filled in as ~$1.5B total raised (as of 2026-08)

data as of 2026-08 · Source: TechCrunch + MarketScale

Publicly verifiable
2026-09-02
团队规模 filled in as ~400 employees

data as of 2026-08 · Source: MarketScale

Publicly verifiable
Spot an error? Submit a correction → — you can correct facts, never negotiate scores.

Want to talk about this company?

Community members may work here or have interviewed — corrections and additions welcome too

Join the discussion