Etched, a rival to Nvidia in AI inference chips, says it has booked $1B in contracts

Etched, a company developing chips for artificial intelligence workloads, says it has already booked $1 billion in sales under contract for “inference systems” powered by its chip. Inference is the stage where trained AI models are put to work to produce outputs, such as recommendations, language responses, or vision detections, and it is a major driver of data-center chip demand.

The claim comes as the company also put a $5 billion valuation on its progress, according to a report circulated by Yahoo Finance on June 30. The article frames the $1 billion figure as booked revenue under an arrangement tied to customers building or deploying inference systems that use Etched’s technology.

Etched is being discussed as a competitor in the same broader AI compute market that has made Nvidia a dominant supplier of data-center graphics processing units and related acceleration platforms. Nvidia’s business model relies on selling compute hardware and software stacks that help customers train and run AI models efficiently, and the inference layer has become especially important as deployments move from experimentation to production.

However, the publicly reported information does not, in the available material, identify the customer, the contract term, the timing of revenue recognition, or how the booked figure maps to the capacity customers are planning to deploy. It also does not specify whether the $1 billion refers to total contract value, revenue expected over time, or an initial commercial booking with further milestones.

The report’s wording suggests Etched is emphasizing traction, not just technology readiness. For chip startups, getting paid for real inference deployments is a key milestone because it can reduce uncertainty about whether their hardware and software ecosystem can meet operational requirements like throughput, latency, power efficiency, and integration with existing AI software workflows.

For Nvidia, competition in inference chips matters because most large-scale AI rollouts increasingly emphasize cost per inference, total power consumption, and reliability in continuous service. Even if customers keep Nvidia as a fallback or primary platform, new competitors can pressure pricing and influence which architectures teams choose for new deployments.

Still, it is important to separate valuation statements and “booked” contract claims from audited financial results. The available material does not include financial statements, revenue breakdowns by customer, or any confirmation from filings that would allow outside observers to verify the $1 billion number independently. Until more details are disclosed, the timing, sustainability, and margins of these bookings remain unclear.