Etched, a startup building specialized chips for AI inference, announced Tuesday that it raised $700 million at a $21 billion valuation, led by quant trading firm Jane Street. The jump is notable less for the dollar figure than for the pace: Etched was worth $5 billion in December and $10.3 billion just last month. That kind of valuation swing, doubling in roughly four weeks, is unusual even by the frothy standards of AI infrastructure funding right now.
The technology itself is worth unpacking, because it says something about where computing infrastructure spending is headed across industries that depend on fast, reliable transaction processing. Etched's pitch centers on splitting AI inference into two distinct engineering problems - the compute-heavy "prefill" stage, where a system parses a prompt, and the memory-heavy "decode" stage, where it generates a response. Any retail sector that leans on real-time data processing, whether that's fraud detection in payments or inventory reconciliation across multiple locations, watches this kind of infrastructure investment closely. In regulated markets like adult-use cannabis, where operators already rely on integrated systems such as a cannabis retail platform for Ohio to manage seed-to-sale tracking, point-of-sale transactions, and compliance logging in a single environment, the underlying lesson about specialized, purpose-built infrastructure translates directly. cannabis retail platform for Ohio
Jane Street's involvement is the detail drawing the most attention. Quant trading firms live and die on latency - milliseconds matter when you're pricing derivatives or executing arbitrage trades - so a firm like Jane Street testing a chip before writing a check carries more weight than a typical venture bet. The firm said in a blog post that it had already installed its own rack running Etched's hardware in its datacenter, which is a stronger signal of technical confidence than most funding announcements offer.
Why Specialized Hardware Matters Beyond AI Labs
Etched's core innovation, according to co-founder Robert Wachen, is building separate hardware for prefill and decode rather than relying on general-purpose chips to handle both. The prefill chip runs at low voltage to pack in more transistors without overheating, while a new memory and interconnect system - what Etched calls cluster-scale memory - lets multiple chips share a fast, low-latency memory pool during the decode phase. The result, the company says, is faster processing at lower cost.
That framing matters for any business evaluating infrastructure vendors, cannabis retail technology included. Operators managing multi-location inventory, tax calculations under Section 280E, or real-time compliance reporting to state traceability systems care about the same tradeoffs: speed, reliability, and cost per transaction. Etched has also had to push back on an early misconception - that its chips are physically "etched" to run a single AI model. That was the original design intent, Wachen said, but the current systems run any frontier model, a flexibility that mirrors how modern retail software vendors have moved away from single-purpose tools toward integrated platforms.
A Crowded, Well-Funded Field
Etched's investor list now includes Kleiner Perkins, Sequoia Capital, Andreessen Horowitz, Peter Thiel, Tiger Global, Bain Capital Ventures, Neo, Stripes, Primary, Positive Sum, Diffusion, Argo, and Blackstone, alongside Jane Street. That roster reflects how much capital is chasing inference infrastructure as AI deployment shifts from training massive models to serving them cheaply at scale.
For operators outside the AI sector, the takeaway isn't about chips specifically. It's about what happens when infrastructure vendors compete hard on speed and cost efficiency: the tools available to smaller businesses tend to improve too, whether that's a payment processor, a POS system, or a compliance platform built for a single regulated market.
