Development
A chipmaker buys a startup that etches model weights into silicon
Claim
AMD bought Toronto's Taalas, which etches weights into silicon rather than using high-bandwidth memory, after a 6nm chip served an 8B model at almost 17,000 tokens a second, each chip locked to one model until re-spin, while Nvidia weighed the blunter fix of less memory in its next architecture.
Reported metric: ~17,000 tok/s
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to August 6, 2026 (theinnermostloop.substack.com)
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.