AMD Acquires Startup to Burn AI Weights Directly Into Silicon Chips
lemire · x · 2026-08-10
AMD acquired a Canadian startup that burns AI weights directly into silicon, effectively creating read-only-memory (ROM).
- Technical Advantages: ROM is cheaper per bit than the memory GPUs rely on. By fixing the weights, the chip becomes both fast and inexpensive for running a specific model. Their first test chip hit thousands of tokens per second on a small open-weight model.
- Limitations: Similar to old-school video game cartridges, the compute and the data (weights) are fused in the same silicon. This means the processor is hardwired to run only that specific model.
- Outlook: The author hopes AMD will actually build and ship these specialized chips, rather than just holding patents to block competitors.
Related event: AMD Acquires Taalas to Bake AI Weights Directly into Chips(2 posts)→
More from Infra
- Amazon Proposes Per-Request Validated Evidence Caching for Production RAG — _reachsumit · 2026-10-16
- LinkedIn's TransX report: sequence transduction for recsys, +6.02% CTR in A/B tests — _reachsumit · 2026-10-16
- llama.cpp PR boosts CPU inference speed with merged AVX2 tiled accumulators — jacek2023 · 2026-10-16
- Zero's next-gen IVM engine runs in a single DO, live-querying a 100GiB R2 database — threepointone · 2026-10-16
- Meme: driving to the chip factory in 2028 captures AI's compute anxiety — beffjezos · 2026-10-16
- One volunteer maintains the time zone file behind 4 billion phones — danbri · 2026-10-16