Etched, the startup betting that a chip purpose-built for transformer models can replace racks of general-purpose GPUs, is fielding funding offers that would value it at more than $40 billion, TechCrunch reported on Tuesday, citing people familiar with the discussions. The bids reportedly range from about $40 billion from top-tier investors to as much as $50 billion from lesser-known backers — a spread that reflects both how aggressively capital is chasing AI silicon and how much uncertainty remains about what the company is actually worth.
The offers arrive barely six weeks after Etched closed a $700 million Series D at a $21 billion valuation, a round led unusually by Jane Street, an existing customer of the company rather than a conventional venture investor. A move to $40 billion or more would roughly double the valuation before a single dollar of the new capital has been spent.
Etched's pitch is architectural. Its Sohu chip hard-wires the transformer computation that dominates modern AI, trading the flexibility of a GPU for efficiency on one narrow class of workload. If the bet pays off, the company argues, a single server can replace a cluster for large-model inference, cutting the power and capital costs that now dominate AI economics. If it does not — or if mainstream accelerator vendors simply keep improving fast enough — the same specialization that makes the design efficient becomes a liability.
No revenue figures have been disclosed publicly, and no investor has been named in the current talks, so the numbers should be read as reported terms under discussion rather than a closed round. The trajectory, though, is the story: a young chip startup moving from a $21 billion round to a reported $40-billion-plus conversation in weeks measures how much money is chasing any credible route to cheaper AI compute — and how little patience investors have for waiting on the next generation of general-purpose hardware.
