Etched raises $700M at $21B valuation for AI inference chips
Subscribe to our free newsletter today to keep up to date with the latest business news.
AI chip startup Etched has raised $700 million at a $21 billion valuation as it begins delivering its first commercial systems.
The round was led by Jane Street, which is also Etched’s first announced customer. The trading firm received the first Etched rack in July and is deploying the hardware in its computing environment.
The deal gives Etched more than fresh capital. It also gives the company an early customer test as investors put large sums into alternatives to the dominant suppliers of AI computing hardware.
Etched says it has secured more than $1 billion in contracts with AI companies and cloud providers. The company has now raised about $1.9 billion.
Its valuation has risen even faster. Etched was valued at $10.3 billion after raising $300 million in July. The latest round roughly doubles that figure within weeks. The company was valued at about $5 billion in December 2025.
That rise puts considerable pressure on Etched to prove it can turn early customer demand into a large and durable business.
The company is entering commercial deployment at a valuation more often associated with businesses that already have large sales operations and mature supply chains. Investors are betting that demand for specialized AI inference chips will grow quickly enough to support a new group of hardware suppliers.
Why inference economics are changing the AI chip market
The case for Etched rests partly on a change in how AI infrastructure is being used.
Much of the first wave of AI investment focused on training. Training is the process of building a model by processing very large amounts of data. It can require thousands of chips working for weeks or months.
Inference happens after training. It is the computing work required each time someone sends a prompt to a model, generates an image or asks an AI system to complete another task.
As AI use grows, inference becomes a recurring operating cost.
That changes the economics of the market. Companies running AI services must consider how much each query costs, how quickly it can be processed and how much electricity the supporting hardware uses.
Those concerns are becoming more important as spending rises.
McKinsey estimates Amazon, Google, Meta and Microsoft could collectively commit more than $700 billion in capital spending during 2026, with AI infrastructure accounting for a large share of the investment.
Power supply is another constraint. The International Energy Agency expects global data center electricity consumption to reach about 945 terawatt-hours by 2030, roughly twice current levels. It also projects electricity use from accelerated servers, driven mainly by AI, to grow about 30% a year in its base case.
The scale of that demand helps explain investor interest in chips designed for narrower tasks.
General-purpose AI accelerators have an important advantage because they can support many types of workloads. Their software ecosystems are also well established.
Specialized chips take a different approach. They give up some flexibility in an effort to process specific workloads more efficiently.
For inference providers, small savings can become significant when spread across billions of requests.
That is the opportunity Etched is pursuing. Better specifications alone, however, will not determine whether the company succeeds. Customers also need reliable systems, software support and enough hardware to run production workloads at scale.
Jane Street offers validation, but the bigger test is scale
Jane Street’s role in the funding round gives Etched an unusual early proof point.
The firm is an investor, the lead investor in the latest round and the company’s first announced customer. Etched says Jane Street tested its hardware before leading the investment.
That gives Etched more evidence of demand than a funding round based only on prototypes or projected performance. A customer willing to deploy a new chip architecture is making a practical decision about cost, performance and reliability.
The relationship still requires some caution.
Jane Street has a financial interest in Etched as well as a customer relationship with the company. Its deployment is an important commercial milestone, but it should not be treated as an independent industry benchmark.
One customer deployment also does not answer the hardest question facing a chip startup: Can it scale?
Semiconductor companies depend on manufacturing capacity, advanced packaging, memory, networking equipment and complex global supply chains. Producing a limited number of systems is very different from supplying data centers with thousands of units.
Etched will also have to compete with Nvidia’s large installed base and mature software ecosystem. Customers moving to a different architecture must see enough financial benefit to justify the cost and risk of changing hardware.
That makes the next stage more important than the funding round itself.
Etched now needs to move from a first delivery to repeat orders. It also has to show that its systems can deliver consistent performance in production environments and that it can supply enough hardware to meet customer demand.
The next phase of AI infrastructure will be judged on efficiency
The amount of capital moving into AI infrastructure gives new chip companies room to compete.
McKinsey estimates that almost $7 trillion could be spent on data centers globally through 2030. More than $4 trillion of that could go toward computing hardware.
Even a small share of that market could support a large semiconductor business.
But heavy infrastructure spending also brings greater scrutiny. Companies operating AI models face pressure to reduce the cost of serving each user while managing limits on power, cooling and data center capacity.
That creates an opening for specialized hardware if it can lower operating costs without making systems less reliable or harder to use.
Etched’s $21 billion valuation suggests investors believe the company has a credible chance of doing that.
Its first delivery to Jane Street moves the company from development into commercial use. The next milestones will matter more: repeat deployments, broader customer adoption and performance at data center scale.
The AI chip market is no longer defined only by how quickly companies can train larger models. As those models reach more users, the cost of running them will carry more weight.
Etched is betting that specialized hardware can bring those costs down. Its valuation shows how much investors think that opportunity could be worth.
Source
