Etched Draws $40-50 Billion Bids: Inside the Inference Chip Startup Challenging Nvidia

Posted on 9th Oct 2026 12:05:40 in Artificial Intelligence, Machine Learning

Tagged as: Etched, AI chips, inference, Nvidia, AI funding

Etched, the four-year-old chip startup that shipped its first AI rack this summer, is fielding funding offers that value the company at $40 billion to $50 billion — roughly double the $21 billion valuation it announced just weeks earlier, according to people familiar with the discussions. The talks are early, no deal has been signed, and Etched declined to comment. But the numbers tell a story that runs against the week’s mood music: even as investors openly debate an AI bubble, the bidding for one of Nvidia’s most credible challengers keeps climbing.

The spread between the offers is telling. Top-tier investors clustered at about $40 billion, while lesser-known backers pushed as high as $50 billion, one person said — a $10 billion gap that reflects how badly funds without brand-name access want a position. If Etched raises a round comparable to its last, the company would buy itself roughly three and a half years of runway, according to the person familiar with the offers — enough to fund the expensive work of building complete AI hardware systems around its own silicon.

A Valuation That Has Kept Doubling: $5 Billion to $40 Billion in Ten Months

Etched’s valuation has compounded faster than almost anything else in AI hardware. In December 2025, the San Jose-based company was valued at $5 billion. On July 23, it announced a $300 million round at a $10.3 billion valuation led by Sequoia Capital, with Jane Street, Andreessen Horowitz and SK Hynix among the participants. On August 18, it announced another $700 million at $21 billion — a round led by Jane Street after the quantitative trading firm tested the hardware and took delivery of Etched’s first shipped rack.

The investor list now spans Kleiner Perkins, Sequoia, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Blackstone and others. TechCrunch notes that these back-to-back financings are sometimes best understood as a single large round split into two tranches with separate valuations — a structure that has become increasingly common among the buzziest startups, and one that critics call a “dual pricing” trick that flatters momentum by re-pricing the same company every few weeks.

If the reported bids become a signed round, Etched’s valuation will have roughly quadrupled since July and grown around eight-fold since December — without the company having publicly disclosed a single dollar of revenue.

What Etched Actually Sells: Inference, Not Training

Etched is not another GPU company. It builds specialized systems for inference — the computing that happens after a user submits a prompt, when a trained model generates its answer. Inference runs behind every chatbot reply, code completion and agent action, and it is where the recurring cost of AI lives.

Co-founder and COO Robert Wachen has walked through the company’s approach: inference splits into two stages. The compute-heavy “prefill” stage reads and understands the prompt; the memory-heavy “decode” stage generates the output tokens a user actually sees. Etched designed a prefill chip that runs at low voltage, which lets it pack in more transistors without the overheating that constrains other high-end AI chips, plus a new memory system and interconnect for decoding that the company calls cluster-scale memory — many chips sharing one very fast, low-latency memory pool. The pitch to customers: more tokens, processed faster, at lower cost than Nvidia’s systems.

The credibility signal investors point to is commercial rather than technical. Etched said in July that it had secured more than $1 billion in customer orders after manufacturing its test chip, called A0, at a TSMC factory this summer. Its lead customer is also its biggest new backer. “We tested the chip and are pleased with the early results,” Jane Street wrote when the $21 billion round was announced. “Etched’s unique approach to inference delivers the precision we will need to support our most demanding workloads. We’re excited to now have our own rack running in our datacenter.”

The company has since opened a 10-megawatt data center in Silicon Valley, set up a facility in Taiwan to coordinate production near TSMC, and grown to about 400 employees. It has also moved past an early perception problem: the original idea of etching one specific model into silicon is gone, and its systems now run any frontier model.

Three Harvard Dropouts and a Team Full of Nvidia Alumni

Etched was founded by Gavin Uberti, Chris Zhu and Robert Wachen, who left Harvard to build the company — Uberti and Zhu met in an advanced mathematics course, and Wachen was Uberti’s roommate. The talent pipeline has become part of the story: roughly 15% of Etched’s workforce previously worked at Nvidia, according to a Wall Street Journal report, at a time when the industry’s biggest constraint is often not capital but people who know how to build AI systems at scale.

The ambitions match the payroll. In its August funding announcement, the company said it is working toward gigawatt-scale deployments, which will require new factories, global supply chains, fleet software and what it calls “self-improving kernel agents.” Its mission statement is blunt about the market it sees: “Under 1% of the world has access to frontier models. Scaling intelligence requires a new kind of inference hardware.”

Why Inference Is the Battlefield — and What the Bet Means

For most of the AI boom, headlines went to training: the giant runs that produce frontier models. But as those models move into daily use, the economics shift to inference. Every AI product that succeeds inherits a growing inference bill, and the companies serving those products pay it every second of every day. Cheaper, faster inference has therefore become one of the industry’s most valuable levers — and the reason a trading firm like Jane Street cares. In markets where microseconds move millions, a system that processes tokens faster is not a novelty; it is an edge.

Nvidia still dominates the compute market that powers all of this, and that dominance is exactly why so much capital is chasing alternatives. Etched is one of several challengers, alongside custom silicon efforts inside the big cloud providers themselves, trying to carve out the inference niche. What makes its case unusual is the combination of a shipped product, a paying customer in Jane Street, a $1 billion-plus order backlog and a founding team investors believe can execute.

For businesses watching from the sidelines, the significance is indirect but real. Competition in inference hardware is one of the forces pushing the cost of running AI downward — the same cost curve that shows up in the prices of the AI tools and API calls that companies use every day. If challengers like Etched deliver on their performance claims at scale, the practical effect is more AI capability for the same money.

The Caveats: Offers Are Not Deals

It is worth being precise about what has actually happened. Etched is reviewing preliminary offers; no financing has been completed, the terms could change, and a deal could still fall apart. The valuation figures are reported bids, not a signed term sheet, and the company has not commented. There is also no public revenue figure to anchor any of it — the $21 billion and $40 billion numbers rest on orders, tests and the belief that AI inference demand will keep compounding for years.

The execution list is long, too. Chip manufacturing, rack-scale production, data-center buildout and global supply chains are capital-hungry and unforgiving, and the company itself acknowledges the challenges ahead. Nvidia, meanwhile, ships at a scale no startup has matched, with a software ecosystem that has taken years to build.

Still, the trajectory says something about where the AI money believes the next decade is going. A few years ago, Etched was a contrarian bet that inference-specific silicon could matter. Today, investors are bidding up to $50 billion for a piece of that thesis — a price that assumes the answer is yes.

Sources

• TechCrunch — Etched fields funding offers at $40B+ valuation, sources say
• TechCrunch — Etched’s valuation doubles to $21B in a month
• The American Bazaar — Etched draws funding offers at over $40 billion valuation
• Etched — From Zero to One
• MLQ.ai — Etched Fields Funding Offers at $40B-$50B Valuation After $21B Round