Milestones
- Asimov tapeoutTarget Dec 2026Current milestone.
- Production readyTarget 2027Not yet reached.
Upcoming
- Dec 2026Asimov tapeoutin about 7 weeks (next)
- 2027Production ready
- Q2 2027Production availability
by PositronUS
Enable cost-effective inference on models with up to 16+ trillion parameters and 10+ million token context windows.
Taping out end of 2026; production H2 2027. Supports 16+ trillion parameter models.
Checked 11 Oct0 updates this week
Prioritizes memory bandwidth over compute, since inference is data-bound.
Multiple Asimov chips integrate for large systems.
| Spec | Asimov |
|---|---|
| Memory capacity per chip | 288GB–2,304GBR (reported) |
| Maximum Titan system memory | 18.4TBR (reported) |
| Manufacturing process | TSMC N3PR (reported) |
| Target model parameters | 16+ trillionR (reported) |
| Target context window | 10+ million tokensR (reported) |
R reported by the company
Positron
Atlas, Asimov inference chip, Titan system
Positron designs hardware and software for AI inference, reducing cost and power. Founded in 2023, it develops Atlas (in production) and Asimov chip (2027). September 2026 raised $875M at $5B.