PulseAugur
EN
LIVE 11:16:27

Ex-Databricks AI chief launches oscillator chip model, claims 1000x power cut · 2 sources tracked

Unconventional AI, founded by former Databricks AI chief Naveen Rao, has launched Un-0, an image generation model. This model operates on a simulated oscillator-based chip architecture, which Rao claims could reduce AI inference power consumption by up to 1,000 times. The startup has secured $475 million in seed funding at a $4.5 billion valuation from investors including Lightspeed, a16z, Sequoia, and Bezos, and plans to develop complete inference stacks. AI

IMPACT This novel architecture could significantly reduce the energy footprint of AI inference, potentially lowering costs and enabling wider deployment.

RANK_REASON Frontier-lab model release with system card and significant funding. [lever_c_demoted from frontier_release: ic=2 ai=1.0]

Read on Mastodon — fosstodon.org →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

Ex-Databricks AI chief launches oscillator chip model, claims 1000x power cut · 2 sources tracked

COVERAGE [2]

  1. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    25 June 2026: Unconventional AI — Naveen Rao, ex-Databricks AI chief — released Un-0, an image model on a simulation of an oscillator-based chip. No transistors

    25 June 2026: Unconventional AI — Naveen Rao, ex-Databricks AI chief — released Un-0, an image model on a simulation of an oscillator-based chip. No transistors — coupled ring oscillators do the maths. Rao claims it could cut AI inference power up to 1,000x. Demo images match Sta…

  2. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    Unconventional AI, led by Databricks former AI chief Naveen Rao, has released its first model built on oscillator-based architecture that could reduce AI power

    Unconventional AI, led by Databricks former AI chief Naveen Rao, has released its first model built on oscillator-based architecture that could reduce AI power use by up to 1,000 times. The startup plans to build entire inference stacks from the ground up. https:// techcrunch.com…