# Automat

Automat is an AI-focused investment collective that pools capital from technology executives and researchers to fund early-stage artificial intelligence companies. It was founded in 2023 by former OpenAI and Google DeepMind members.

Automat is an investment collective focused exclusively on artificial intelligence startups. Founded in 2023, the group aggregates capital from a network of technology executives, AI researchers, and operators to provide seed and early-stage funding. The collective operates with a thesis that transformative AI companies will emerge from deep technical research and applied machine learning, rather than incremental product improvements.

The organization was established by a group of former engineers and researchers from leading AI labs, including [openai](https://www.wikiprompt.org/wiki/openai) and [google-deepmind](https://www.wikiprompt.org/wiki/google-deepmind). Its founding members include [jakob-uszkoreit](https://www.wikiprompt.org/wiki/jakob-uszkoreit), a co-inventor of the [transformer](https://www.wikiprompt.org/wiki/transformer) architecture, and [lukasz-kaiser](https://www.wikiprompt.org/wiki/lukasz-kaiser), another transformer co-author. They were joined by [niki-parmar](https://www.wikiprompt.org/wiki/niki-parmar), who previously led applied AI at a major cloud provider, and [mark-chen](https://www.wikiprompt.org/wiki/mark-chen), a former product lead at a large language model startup. The collective's headquarters are in San Francisco, California, with a small operational team that manages deal flow and due diligence.

## Investment Strategy

Automat focuses on pre-seed and Series A rounds, typically writing checks between $500,000 and $5 million. The collective prioritizes companies working on fundamental AI research, including [neural-network](https://www.wikiprompt.org/wiki/neural-network) efficiency, [large-language-model](https://www.wikiprompt.org/wiki/large-language-model) alignment, and novel [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) architectures. Unlike traditional venture capital firms, Automat does not manage a single pooled fund; instead, it coordinates syndicate investments where individual members commit capital on a deal-by-deal basis. This structure allows the group to move quickly and avoid the pressure of deploying capital within a fixed fund lifecycle.

The collective has backed approximately 15 startups since its founding. Notable portfolio companies include a [groq](https://www.wikiprompt.org/wiki/groq) competitor developing custom AI inference chips, a [samba-nova](https://www.wikiprompt.org/wiki/samba-nova) rival focused on [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) hardware, and an [ai21-labs](https://www.wikiprompt.org/wiki/ai21-labs)-adjacent natural language processing firm. Automat also invested in an [inflection-ai](https://www.wikiprompt.org/wiki/inflection-ai) spin-off working on conversational AI for enterprise customer service. These investments were made between 2023 and 2025, with the largest single check of $4 million going to a [figure-ai](https://www.wikiprompt.org/wiki/figure-ai)-like robotics startup in early 2024.

## Membership and Network

The collective's membership is capped at 50 active investors, drawn from a broader network of over 200 alumni and advisors. Members include former researchers from [mit-csail](https://www.wikiprompt.org/wiki/mit-csail), [stanford-ai-lab](https://www.wikiprompt.org/wiki/stanford-ai-lab), and [berkeley-ai-research](https://www.wikiprompt.org/wiki/berkeley-ai-research), as well as engineering leaders from [amazon-web-services](https://www.wikiprompt.org/wiki/amazon-web-services), [azure](https://www.wikiprompt.org/wiki/azure), and [google-cloud](https://www.wikiprompt.org/wiki/google-cloud). The group also includes several academics, such as [michael-jordan](https://www.wikiprompt.org/wiki/michael-jordan) and [anima-anandkumar](https://www.wikiprompt.org/wiki/anima-anandkumar), who serve as informal technical advisors rather than financial contributors.

Membership requires a minimum annual commitment of $250,000 and active participation in at least two investment reviews per year. The collective holds quarterly technical deep-dive sessions where members present on topics like [multi-head-attention](https://www.wikiprompt.org/wiki/multi-head-attention) improvements, [learning-rate-schedule](https://www.wikiprompt.org/wiki/learning-rate-schedule) innovations, and [data-augmentation](https://www.wikiprompt.org/wiki/data-augmentation) techniques. These sessions serve both as educational forums and as a way to source investment opportunities from emerging research.

## Deal Process

Automat's deal flow comes primarily from member referrals and direct outreach to academic labs. The collective maintains a rolling application process, with startups submitting a technical whitepaper alongside a standard pitch deck. A three-member screening committee, rotated every six months, evaluates each submission against the group's technical criteria, which include novelty of the approach, potential for [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) or efficiency gains, and the strength of the founding team's research background.

Once a startup passes initial screening, it is invited to present at a monthly investment meeting. These meetings are held virtually, with recordings made available to members who cannot attend live. A deal is approved when at least 60% of participating members commit capital. The average decision time from first contact to funding is 45 days, which the collective cites as faster than traditional venture capital due to the absence of a formal partnership structure.

## Impact and Reception

Automat has gained recognition in the AI investment community for its technical rigor and speed. In 2024, the collective was featured in a [stanford-ai-lab](https://www.wikiprompt.org/wiki/stanford-ai-lab) case study on alternative funding models for AI research. The group has also been credited with helping early-stage companies navigate the competitive hiring market by connecting them with its alumni network of [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) specialists.

However, some industry observers have noted that Automat's narrow focus on technical novelty may cause it to overlook commercially viable applications. The collective has responded by adding two members with product management backgrounds in 2025, including a former executive from [samsung-electronics](https://www.wikiprompt.org/wiki/samsung-electronics)' AI division. As of early 2026, Automat reports a portfolio-wide valuation increase of 3.2x on invested capital, though it does not disclose individual company performance.

The collective's long-term goal is to establish a self-sustaining research ecosystem, where portfolio companies contribute findings back to the membership. Automat has begun hosting an annual symposium, the first of which took place in October 2025 in San Francisco, attracting over 300 attendees from academia and industry.

## Governance and Future Plans

Automat is structured as a limited liability company, with a three-person managing board elected by members annually. The board oversees operational expenses, which are funded by a 2% annual fee on committed capital. The collective has no external limited partners, and all profits from exits are distributed pro rata to participating members.

In 2026, Automat plans to launch a fellowship program for early-career AI researchers, offering $100,000 grants to support independent work on [residual-network](https://www.wikiprompt.org/wiki/residual-network) improvements and [loss-functions](https://www.wikiprompt.org/wiki/loss-functions) design. The collective is also exploring a partnership with [tsmc](https://www.wikiprompt.org/wiki/tsmc) to provide portfolio companies with access to advanced chip fabrication for custom AI accelerators. These initiatives reflect Automat's ambition to move beyond passive investing and become an active contributor to the AI research landscape.

---
Source: https://www.wikiprompt.org/wiki/automat
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-09T01:56:08.243056+00:00