One model per chip: AMD bought Taalas, whose test part hit 17,000 tokens a second
Taalas casts a model's weights into transistors. Its 6nm test chip served Llama 3.1 8B at roughly 17,000 tokens per second, and AMD just bought the company.
Taalas casts a model's weights into transistors. Its 6nm test chip served Llama 3.1 8B at roughly 17,000 tokens per second, and AMD just bought the company.

Samsung signed an MOU with Broadcom worth more than $200 billion through 2030 across memory, 2nm foundry and advanced packaging. Nvidia locked up SK Hynix the same day.

IBM showed a 0.7nm 'nanostack' transistor with ~100 billion transistors on a fingernail. It's real research, but the name is marketing and shipping silicon is years out.

Bloomberg says Apple is in early-stage talks with Intel and Samsung Foundry to build base M-series chips in the US starting 2027. Intel closed at a record $543B market cap.

Elon Musk said Tesla plans to manufacture at Terafab in Austin on Intel's 14A process. It's the first major external customer Intel CEO Lip-Bu Tan needs to keep the foundry alive.