Reading this on Krawl? Register for free.
Unlock listen-aloud, reading history and personalised feeds — at zero cost.
Free registration unlocks the full Finance Desk
$21 Billion for a Four-Month-Old Product: How the Market Is Learning to Price Young Chip Companies
A startup founded by three Harvard dropouts is now worth more than Nvidia paid for Groq. We use Etched to explain how investors value young technology companies — and where that arithmetic can break.
Four months ago, the product did not exist outside a design document. Today, Etched — a semiconductor startup co-founded by three Harvard dropouts, all of them 24 — has booked more than $1 billion in orders, started shipping, and is closing a $700 million round led by Jane Street that values the company at $21 billion. Total capital raised is approaching $2 billion.
For anyone studying valuation, this is not merely a Silicon Valley story. It is a clean illustration of how private markets price companies when the traditional inputs are missing, and of a genuine structural shift in where global technology capital is flowing.
Why this deal breaks an old rule
Venture capital has spent two decades getting rich on software, where the marginal cost of a new customer is near zero and a product can be rebuilt in a weekend. Semiconductors are the opposite: enormous upfront capital, multi-year development cycles, and physical supply chains. The prevailing wisdom among investors, as Sequoia's Sonya Huang has put it, was to avoid backing young founders in chips — inexperience is survivable on the internet, far less so in silicon.
Etched inverted that logic deliberately. It bought tens of millions of dollars of diagnostic equipment and designed its own server racks before its first wafers were even fabricated. It built a two-megawatt data centre inside its own San Jose offices so customers could test remotely. It keeps a team of twenty in Taiwan and owns a factory producing server components — vertical integration that most chip startups outsource. The payoff was speed: 44 days from receiving test chips from TSMC to running live inference workloads, against an industry norm of six months or more.
The valuation puzzle
Here is the question a CFA candidate should be able to frame. How does $1 billion of bookings support a $21 billion valuation — a multiple of roughly 21 times orders that have barely converted to revenue?
A discounted cash flow model is close to useless here. DCF requires forecastable free cash flows and a defensible terminal value; Etched has neither a revenue history nor a stable competitive position. So investors fall back on three other approaches.
The market approach. You look for a comparable transaction. Conveniently, one exists: in December, Nvidia paid roughly $20 billion to license technology from and hire the leadership of Groq, a similar inference-focused startup. That transaction effectively established a market-clearing price for this asset class, and Etched has now been marked just above it. This is exactly how comparable-company and comparable-transaction analysis works — and exactly why it is fragile. One data point is not a market.
The venture capital method. An investor estimates a plausible exit value at some horizon, discounts it at a very high required return (often 40–60% annually to compensate for failure risk and illiquidity), and works backwards to today's price. At a $21 billion post-money valuation, the buyer is implicitly underwriting an exit in the hundreds of billions — which only makes sense if Etched captures a meaningful slice of AI inference compute, a market Nvidia currently dominates.
Probability-weighted scenarios. Rather than one forecast, you assign outcomes: a small chance of an enormous win, a moderate chance of an acquisition, and a substantial chance of near-total loss. Venture returns follow a power law — a handful of investments carry an entire fund — so a rational investor can pay a price that looks absurd against the base case if the tail case is large enough.
Talent as an unrecorded asset
Roughly 15% of Etched's 400 employees came from Nvidia, including a chief silicon architect who worked on multi-GPU parallelisation there, and a systems engineer with nearly 23 years at the company who has since brought across about a dozen colleagues. A former Nvidia data-centre engineering VP is an investor.
None of this appears on a balance sheet. Accounting recognises internally generated intangibles only in narrow circumstances, which means the gap between book value and market value for such firms is enormous and largely unexplained by reported financials. When you see a price-to-book ratio in the hundreds for a technology company, this is usually what you are looking at.
The trend behind the headline
Three shifts are visible in this single transaction.
First, capital is rotating from AI software toward AI hardware and inference — the computation that happens every time a model answers a query, as opposed to the one-time cost of training it. Etched's claim is architectural: a design allowing chips inside a server to communicate as one processor, cutting certain latencies from about 4,000 nanoseconds on an Nvidia Blackwell processor to roughly 700.
Second, the investor base has broadened well beyond classic venture. This round includes Kleiner Perkins, Andreessen Horowitz, Tiger Global, Bain Capital Ventures and Blackstone. Crossover and private-equity money now competes for pre-profitability assets, which structurally pushes private valuations higher.
Third, and most interesting, the customer is the investor. Etched targeted quantitative trading firms as both buyers and backers, and even hired a former Two Sigma quant to run finance. Jane Street is simultaneously its first customer and its lead investor. That alignment validates demand — but it also means the anchor customer has an interest in a rising mark, a conflict worth naming.
The risks to hold in mind
Concentration: one dominant customer. Competition: Nvidia can respond with price, roadmap or acquisition. Key-person dependency in a company whose value is largely its engineers. And valuation risk itself — private marks are model-based, not market-tested, and adjust violently when sentiment turns.
The lesson is not that $21 billion is right or wrong. It is that in early-stage technology, price reflects the distribution of possible futures, not the present cash flows. Understanding which assumption is doing the heavy lifting is the whole job.
Found this useful? Share it!
Interested in Finance Education?
Explore our CFA and investing courses — built for serious learners.
