The first rung of the ladder
Here is what exists today: the company's job board carries a listing for a silicon engineer and one for a technical program manager on the silicon side, and a spokesperson confirmed the plan to Business Insider, which spotted the listings, and to TechCrunch. In a chip programme that is the rung below architectural decisions. There is no named process node, no design sent to manufacturing, no wafer ordered and no dated delivery.[1]
The spokesperson's phrase about a multi-chip approach is the load-bearing part of the sentence. If hardware from other companies keeps working alongside its own designs as the company scales, the first function of the new team is to bring in-house the ability to design hardware and models together, rather than to replace a supplier. A second reading holds: these listings could be an attempt to create an option that improves the terms of externally bought hardware in negotiation. What separates the two readings is who owns the first concrete design decision.[1]
The companies walking the same road
The comparison only works if the rungs are kept apart. Its competitor OpenAI announced a chip called Jalapeño, developed with Broadcom for large language model inference in data centres; Google has been running its models on its own hardware for a while and Meta has designed and deployed its own chips as well. On the Anthropic side there is an assessment reported by The Information: Samsung as a manufacturing partner. That belongs on the shelf as a reported discussion rather than a confirmed contract.[1]
Because key team members are still being hired, it will be some time before the company or its users see any benefit. Two measurements would move this file up a rung: a foundry agreement with a named partner and process node, and a co-design result measured on a stated workload with its boundaries given. The plainest form of the second is throughput per watt. Until those measurements are published, the file stays a hiring plan.[1]