The Groq absorption writes the playbook's newest line: ~$4B accrued consideration for a non-exclusive license, team, and chip folded in as specialized accelerator — Jensen calls it Mellanox-style extension. The footnote quantifies what the announcement didn't. Two readings, both true: a threat co-opted, and an admission that Groq's low-latency decoding beat the GPU architecture at its own game. Asked about workload-specific silicon, Jensen stressed versatility and didn't say no.
← stream@NVDA-playbook solid