Etched Melonjak ke $21B dalam Lonjakan Sepekan yang Panjang

WHALE WATCHER · 2026-08-20T08:03:25.000Z
Diukir, perusahaan chip AI yang didukung Jane Street, mengumpulkan $700 juta dalam putaran pendanaan baru pada 18 Agustus, yang lebih dari dua kali lipat valuasinya menjadi $21 miliar dalam waktu kurang dari sebulan. Valuasi sebelumnya berada di bawah $10 miliar, artinya seluruh pelipatan terjadi dalam hitungan minggu—tidak biasa bahkan menurut standar lingkungan pendanaan AI 2025 dan 2026, di mana valuasi bergerak cepat. Para investor bertaruh bahwa silikon inferensi yang dibuat khusus akan meraih porsi yang jauh lebih besar dari pembangunan AI, dan startup ini kini termasuk salah satu perusahaan perangkat keras dengan valuasi tertinggi di dunia.
Etched, the AI chip startup backed by Jane Street, raised $700 million in a new funding round on August 18 that more than doubled its valuation to $21 billion in under a month. The previous valuation stood below $10 billion, meaning the entire doubling compressed into weeks, unusual even by the standards of the 2025, 2026 AI funding environment, where valuations have moved fast.
Investors are betting that purpose-built inference silicon will capture an outsized share of the AI buildout, and the startup is now among the most richly valued hardware companies in the world.
Key Takeaways
Etched raised $700 million on August 18, more than doubling its valuation to $21 billion in under a month
The company was founded in 2022 by Gavin Uberti and Chris Zhu, both of whom left Harvard to start it
Etched’s Sohu chip hardwires transformer computation into silicon as an application-specific integrated circuit built for one task
Jane Street, a quantitative trading firm, is a notable backer and does not typically invest in early-stage technology companies
Reuters reported the company confirmed the raise on Tuesday, citing the valuation directly.
The Physical Bet Etched Has Hardwired Into Silicon
The company is built around a single idea: the transformer architecture, which underpins virtually every frontier AI model from OpenAI‘s GPT series to Google‘s Gemini, is here to stay. Most semiconductor companies build general-purpose chips that can run many kinds of software. Etched took the opposite approach and hardwired transformer computation into silicon, producing an application-specific integrated circuit, or ASIC, a chip built for one task.
Unlike a graphics processing unit, originally designed for video game pixels and later repurposed for AI training, an ASIC cannot be reprogrammed.
If the transformer architecture were ever replaced by something different, the chip would become obsolete. That is the core risk.
The core upside is raw efficiency: a chip that does only one thing can do that one thing far faster and at lower energy cost than a chip designed for many tasks.
The company argues the transformer bet is safe. Every frontier model released in the last four years has used the transformer as its backbone, including GPT-4, Gemini Ultra, and Claude 3, each of which shipped on transformer foundations with no announced successor architecture.
Every watt and every unit of die area on an Etched chip is allocated to transformer operations, which is precisely why the company claims its Sohu chip can run the largest language models at speeds Nvidia‘s H100 GPU cannot match on pure inference workloads. Locking the architecture into hardware before competitors catch up creates a durable cost advantage in AI inference, the phase where a trained model generates answers for users, which now represents the majority of AI compute spending as deployed applications multiply.
From Cambridge Dorm Room To $21 Billion Valuation
Etched was founded in 2022 by Gavin Uberti and Chris Zhu, both of whom left Harvard to start the company.
The startup emerged from stealth in mid-2024 with its first chip, called Sohu.
The prior funding round, which set the valuation at roughly $10 billion, closed just weeks before this $700 million raise. The speed reflects a specific dynamic in the AI hardware market: demand for inference compute has grown faster than most analysts projected, and purpose-built inference chips are now a credible alternative to the GPU-dominant supply chain.
That demand is physical, more queries, more tokens, more power draw, more rack space consumed, and it is that physical constraint, not a slide deck, that is pulling capital toward Etched.
Jane Street, the quantitative trading firm known for systematic capital allocation, is a notable backer. Jane Street does not typically take large positions in early-stage technology companies.
Its presence on the cap table signals that sophisticated capital believes the inference-ASIC thesis carries institutional-grade merit.
Also Read: Nvidia’s $105B Ohio Guarantee Corrects the $250B Myth
Why The $21 Billion Number Matters Beyond The Headline
The valuation milestone is not just a funding story. It reflects a structural shift in how capital is pricing AI infrastructure risk.
For the first three years of the current AI investment cycle, nearly all chip capital flowed to Nvidia and to GPU cloud providers building on Nvidia hardware. Etched is among a small group, alongside Cerebras and Groq, making the case that specialized silicon will eventually fragment Nvidia‘s near-monopoly on AI compute.
The counter-argument is that Nvidia has not stood still.
Its newest architectures incorporate increasingly specialized transformer acceleration, blurring the line between general-purpose GPU and ASIC. Nvidia also controls the CUDA software ecosystem, which most AI developers use to write and optimize their models. Any ASIC challenger must offer hardware performance advantages large enough to justify the cost of porting software away from CUDA.
A $21 billion valuation implies investors believe Etched will capture significant revenue at scale.
That figure is larger than the entire market capitalization of most established semiconductor companies outside the top ten. The company has not publicly disclosed revenue figures or customer names.
What The Raise Signals For The Broader Market
Long-term bond yields hit multi-decade highs on August 18, Bloomberg noted, tightening the environment for growth-stage valuations.
Higher rates typically compress valuations for growth-stage companies by raising the discount rate applied to future cash flows. The fact that Etched commanded a doubling in valuation under those conditions signals that investors view AI hardware as sufficiently near-term in its revenue potential to override macro pressure.
When capital spending and public strategy disagree, believe the spending, and $700 million in a rising-rate environment is unambiguous.
The next test is commercial deployment at scale. A chip that performs well in controlled benchmarks must also prove reliability across diverse enterprise workloads and integrate with existing AI software stacks.
It must also ship in sufficient volume to support data center-scale orders, where power budgets, thermal envelopes, and supply chain continuity determine whether a promising ASIC becomes a real business or a well-funded benchmark result.
Etched has not yet published independent third-party validation of its Sohu performance figures. That gap between stated performance and verified production results is the key question the $700 million will need to answer.
Read Next: iPhone 17 Pro Users Get Exclusive Breakthrough iOS 27 AI Features