The company, co-founded by Harvard dropouts Robert Wachen, Gavin Uberti and Chris Zhu — all in their early 20s — has raised nearly $2 billion on the strength of a product that existed only as an idea four months ago. The founders have been dubbed the “kids in chips,” a nod to both their age and their ambition to disrupt the AI hardware market that Nvidia currently dominates.
Why Nvidia Engineers Are Jumping Ship
Etched’s pitch to Nvidia talent rests on a simple premise: equity. Early employees at a $21 billion startup stand to gain far more on a percentage basis than latecomers to a $3 trillion incumbent. That math, combined with the chance to build a chip purpose-built for transformer models rather than general-purpose GPUs, has given Etched rare leverage in one of the tightest labor markets in technology.
Nvidia has lost senior engineers to younger rivals before, but rarely to a company this young or this small. The talent drain signals that the moat around Nvidia’s workforce — long considered as valuable as its CUDA software stack — is starting to thin at the edges.
A Product That Was Just an Idea Months Ago
Four months ago, Etched’s chip was a concept on a whiteboard. Today, the startup has billions in committed capital, a valuation that puts it alongside established AI infrastructure players, and a hiring pipeline pulling directly from the industry’s incumbent. That trajectory is unusual even by the standards of the current AI funding boom, where nine-figure rounds have become routine.
Etched is building a fixed-function ASIC optimized for transformer inference rather than the flexible GPU architecture Nvidia sells. The bet is that as inference workloads overwhelm general-purpose accelerators, specialized silicon will win on cost and performance per watt.
The Risk Behind the Valuation
A $21 billion valuation for a company without a shipping product invites scrutiny. Etched’s entire thesis depends on transformers remaining the dominant model architecture long enough for its chip to matter. If the industry pivots to a new class of model — as it has before, from RNNs to attention mechanisms — fixed-function silicon becomes a very expensive mistake.
There is also the execution question. Designing a chip is one challenge; taping it out, securing fab capacity at TSMC, building the software stack, and convincing hyperscalers to deploy it at scale is another. Nvidia spent over a decade building CUDA into the moat it is today. Etched has months.
What Happens Next
Watch for three signals in the coming quarters. First, whether Etched can actually tape out its chip on schedule — any slip undermines the valuation narrative. Second, whether Nvidia responds with retention packages or counter-offers to slow the bleeding; a public talent war would confirm that Etched is being taken seriously inside the incumbent. Third, whether a hyperscaler commits to deploying Etched’s silicon at scale, which would be the real validation of the fixed-function thesis.
The broader implication is that the AI hardware market is fragmenting. Nvidia’s dominance is not in question today, but the existence of a $21 billion startup built specifically to attack its inference business suggests investors see room for specialized challengers. If Etched ships, expect more “kids in chips” clones. If it stumbles, the fixed-function bet loses its most visible test case.
— David Kim, technology desk, AXO News