Observed arrival · 2026-11-13
Overlay wants to keep the whole model on the chip
Overlay is a pre-silicon hardware and AI research lab designing an inference chip that keeps model weights in on-chip SRAM.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams building latency-sensitive AI agents
- Worth noticing
- The page says the design is bit-exact in simulation; test silicon is next, and a white paper is not yet available.
Field notes
The page distinguishes its first all-SRAM chip from a longer-term mixed-die concept that would put stable base weights in etched ROM and changing weights in rewritable SRAM. It says the compiler places model parts, the design is bit-exact in simulation, and a test chip is next. Performance comparisons specify one user; the GPU figures include both a measured standard-engine result and a separately labeled tuned-engine estimate.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
One card from the complete issue