Forget teaching a man to fish. Anthropic would rather teach the AI to fish, then eat well forever off custom chips it never has to pay a human full price for.
The AI lab is currently paying $500,000 to $850,000 a year to research engineers whose job is to teach its AI models how to design silicon chips. Meanwhile, the engineers actually building Anthropic’s first ASIC, the real chip work, are getting $320,000 to $485,000. Same ballpark of skills. Very different paycheques.
Why this matters
Back in May Anthropic’s throwaway mention of “logic chips” alongside its Samsung partnership had Korean media buzzing about a possible foundry tie-up. Two months on, Anthropic all but confirmed the plan, telling Business Insider it’s building out an in-house silicon design team and taking a “multi-chip approach.”
That confirmation is what makes this pay gap so telling. Both roles reportedly need the same core expertise: full ASIC/FPGA flow, RTL to tape-out, UVM/formal, physical design, PPA, DFT, and EDA tools. Yet teaching the AI to do the job pays nearly double what doing the job pays.
Anthropic isn’t alone in chasing AI-designed silicon either. Moonshot’s Kimi K3 model reportedly built a working chip design in 48 hours using around 8,700 tokens, complete with a 4.0 mm² area and a Nangate 45nm library, all built on open-source EDA tools. The simulated chip hit over 8,700 tokens per second in decoding throughput.
Turns out chip engineers aren’t safe from AI’s appetite either. Even the people building the machine get outbid by the people training it.
For more updates like this, check out the gaming news section!
More Stories
OpenAI Security Test Incident Reveals Two New AI Testing Cases
SAPPHIRE EDGE+ Apex Brings Faster AI Performance for Next Generation Robotics
MSI Launches PRO MAX EDGE AI Plus Desktop for Faster Local AI Performance