John’s Background

• Joined 4 years ago during the pandemic, previously energy and chemicals VC (part of SaudiAramco)
• Some areas like equipment manufacturing would expect business org to participate in DD and have business cases
• For Gen AI, more about investing to learn to understand the dynamics
• But financial above return is important.
• Come in at B - C most commonly, $3m-$5m checks, can go down to $1m if it’s really early

Cusp

• We’re not going to play in the foundation model and have a limited role in the middleware layer. Then on application level, see productivity tools where we would be opportunistic
• Applying Gen AI to strategic focus areas (Applied as a Power User) is where something like Cusp falls into. Trying to assess if transformer or diffusion models provide additional leverage to overcome the challenges we have
  • We have a lot of data and expertise. We have talked to a bunch of companies even before gen AI, a huge batch of materials informatics companies
• Two questions we need answered:
  • Does it help us? We have specific materials challenges we’re discussing with Cusp
  • From a business model perspective, how does it monetize?
    • Some companies pitch a new molecule, and its 6-9 months, takes a bunch of projects, not very scalable
    • It’s clear to us that if they came back with a good candidate that we would go through the typical process but that’s multi-year process. It’s the traditional old process. We’re investing in some materials synthesis companies
• Architecture makes sense, graph model + neural network seems like the right path. Sounds like they’re leveraging a powerful MOF engine (Prisma)
  • How translatable is this to non-MOF materials, because MOF is a small market. High performance but very niche
  • The evaluation engine needs to generalize, not sure how big of an undertaking that is
  • Think MOFs are probably the easiest to model but when you go out of MOF, things that are wacky, what would that engine look like, and who will build it, and how long of an effort
  • There’s some literature about MOFs in semis. The challenge we care about is in unwaffered materials. Some papers talk about this but haven’t seen academic research or industrial findings that are going in the direction of MOFs. In the broader manufacturing operation, there are use cases for hazardous gas storage
• We spoke to Cusp about a handful of problem statements in semis. Locade dielectric, needed for transistors. How would you go and about to do it. Understand they don’t have a MOF engine, but want to understand the steps they need to take
• The attention and resources the company puts on CO2 is not on stuff that we care about. Used to work for Lindi and GE doing CO2 capture stuff. There’s industrial capture use cases, why are we jumping to DAC? Industrial capture is like $100/ton. Sure people are willing to pay but what is that market size

Orbital Materials took the vertically integrated path and become a full stack licensor. It’s bold but scary. Think it’s a very different business with a time and capital intensive model