Etched raises $800M, hits $5B valuation with $1B in sales for AI inference chips

AI Chip Startup Etched Emerges With $800 Million in Funding

AI chip startup Etched has officially emerged from stealth mode with a striking announcement: the company has raised $800 million in total funding and secured $1 billion in sales contracts for its specialized inference hardware 1

1

. The bulk of this capital came from a $500 million round led by Stripes that closed in December, valuing Etched at $5 billion post-money 2

2

. Founded in 2022 by Harvard dropouts Gavin Uberti and Robert Wachen—both Thiel fellows—the company has attracted an exceptional roster of investors including Jane Street, which led a previously unannounced round and has invested more than $100 million in total 4

4

.

The investor list reads like a who's who of AI and finance. VentureTech Alliance, a venture firm with a strategic partnership with TSMC, participated alongside trading firms Hudson River Trading and Two Sigma, plus Ribbit Capital 2

2

. Angel investors include AI heavyweights Geoffrey Hinton, who won a Nobel Prize for foundational work in modern AI, computer vision pioneer Fei-Fei Li, Andrej Karpathy, Arthur Mensch, and Scott Wu. Billionaires Stanley Druckenmiller and Peter Thiel also back the venture 1

1

.

TSMC Manufactured Chips Target AI Inference Bottleneck

Etched positions itself as a Nvidia competitor by focusing exclusively on AI inference rather than training. The company successfully completed its first A0 silicon tapeout earlier this year using TSMC's N4P process technology 3

3

. This specialization matters because inference—what happens after a user submits a prompt—is currently the biggest bottleneck and cost center for AI companies trying to serve customers at scale 1

1

.

The company's chip, called Sohu, embeds the transformer architecture directly into silicon rather than relying on general-purpose GPUs 2

2

. Etched has developed what it calls low-voltage inference technology, running chips at half the voltage of most AI chips to prevent overheating and maintain sustained performance 4

4

. This approach solves a critical problem: most AI chips throttle down as they draw more power at full-voltage mode, delivering under half their peak FLOPs. Etched's VLI processor can run trillion-parameter sparse MoEs at 80% peak FLOPs without thermal throttling 3

3

.

Frontier Inference Clusters Offer Full-Stack Solution

What sets Etched apart from other specialized AI chips startups is its full-stack approach. The company sells what it calls frontier inference clusters—complete systems that bundle the chips with custom-designed server rack hardware including circuit boards, cooling plates, and networking connections 4

4

. No other chip startup has designed the entire rack system, according to co-founder and president Robert Wachen 2

2

.

Etched has also developed a Cluster Scale Memory system that combines High-Bandwidth Memory and SRAM to tackle both memory capacity and memory latency challenges 3

3

. This hybrid approach addresses the tradeoffs between massive SRAM blocks that offer faster decode speeds but poor throughput, and HBM that provides capacity but higher latency. The company promises best-in-class throughput, latency, cost, and power efficiency for both prefill and decode workloads across multi-trillion-parameter models, long-context, and agentic AI applications 3

3

.

AI Hardware Innovation Race Intensifies

The timing of Etched's announcement reflects a dramatic shift in the AI hardware landscape. Back in 2023, the founders struggled to attract investor interest despite a 30-page memo arguing that AI would eventually need specialized AI chips beyond general-purpose GPUs. Every major investor they pitched passed, and the company reportedly operated month-to-month, close to running out of cash 1

1

. Today's funding environment looks completely different, with investors chasing everything AI-related, especially chip technology that speeds up inference.

The company has assembled a team of over 400 engineers from leading firms including Nvidia, Google, Broadcom, TSMC, and SK Hynix 3

3

. More than half are based near its San Jose headquarters 2

2

. Etched has built a 2MW datacenter in its offices and opened a factory in Taiwan for 24/7 engineering operations 3

3

.

The inference chip market is attracting enormous capital as the industry shifts from training models to running them at scale. Competitor Cerebras had the first breakout IPO of the year, while Groq raised $650 million. Hyperscalers Amazon, Google, and Microsoft all build their own in-house AI chips, and OpenAI recently announced its first custom chip built by Broadcom 1

1

. London-based Fractile raised $220 million for inference chips that put compute and memory on the same die 2

2

.

Etched is currently testing its first product with customers and plans to start shipping this summer 2

2

. Whether the company can deliver on its $1 billion in contracts depends on whether its chips perform as promised under production workloads. As Positive Sum CEO Patrick O'Shaughnessy noted, "If you have compute now, people will buy it" 2

2

. The startup's bet is that being early with a full-stack rack solution gives it an edge that matters more than benchmark numbers alone.