Investment thesis

AI that takes on real work.

We back two kinds of companies: vertical AI that takes on real work in labour-heavy industries, and the companies removing what holds AI adoption back.

The opportunity

Software has captured a small share of what companies spend on work. AI changes that: it can do the work, not just support it.

US wage bill vs US SaaS spend, drawn to scale

US wage bill: ~$11T

US SaaS spend: ~$116B

SaaS spend is about 1% of what the US pays for work. [Source]

Labour spend, ten US industries

~$8.75T

Addressable by AI today

~$520B

Where the gap is widest

Between what AI can already do and what is actually deployed:

  • Professional services
  • Health care
  • Finance and insurance
  • Public administration

Theme 1

Vertical AI in markets facing labour displacement

We back software that does real work inside specialised industries.

  1. Large labour spend

    Buyers already pay for the work through salaries, outsourcing or services firms, and talent shortages push adoption.

  2. Hard-to-serve markets

    Workflows depend on domain knowledge and frequent exceptions, and fragmented customers make the segment costly for horizontal platforms.

  3. A durable advantage

    Proprietary data that compounds with use, deep integration into systems of record, and ownership of the customer relationship.

Theme 2

Removing the constraints on AI adoption

AI can already do far more work than it is trusted with today. What holds adoption back is everything around the model. We back the companies that clear these bottlenecks.

Compute and energy

Affordable access to compute, and the power and hardware needed to run it at scale.

Data and integration

Turning scattered enterprise data into something AI can use, and connecting AI to the legacy systems where work actually happens.

Trust

Evaluation, security and compliance, so that enterprises can hand real work to AI with confidence.

The physical world

Robotics and automation that carry AI into physical work such as warehousing and manufacturing, where labour shortages are most acute.

What we avoid

Saying no clearly is part of the thesis.

  • Technology in search of a problem

    We start from work that someone already pays for, not from what a model can do.

  • Thin products built on model access alone

    Without the workflow depth that keeps a horizontal platform from serving the same need.

How a company fits

We ask four questions of every company.

  1. Is the problem large, and already paid for today?

  2. Is it hard for a general platform to solve?

  3. Will the advantage compound with use?

  4. Is this the founder who has lived the problem, with the right intent and capability?