Market Opportunity

The market opportunity for AI in chemical / material sciences is massive. Annual spend on materials & chemicals R&D exceeded €11bn in Europe and $41bn globally in 2022. At the same time, molecular design is slow and expensive, with multiple bottlenecks across the R&D process. Furthermore, many of the most pressing challenges we face, such as those in climate change, clean energy and healthcare, require significant material sciences unlocks to overcome (e.g. unlocking scalable carbon capture technology or more efficient battery technology). Like in biotech, the potential for a significant contribution from in silico modeling and discovery is increasing exponentially with recent increases in compute, data, and models.

The potential application of computational methods in the materials R&D process consists of a few key steps:

• Candidate molecule generation: Candidate molecules are generated at the top of the funnel. Today, generative models are limited in accuracy, so R&D teams often generate hundreds of thousands of potential molecules to increase the likelihood of success, but then need to have ways to pare them down.
• Computational screening: Simulation models are used to screen candidates for various characteristics like solubility, stability, absorption, etc. to narrow down the number of candidates - as opposed to biotech, these scoring mechanisms are pretty well understood and tend to perform well.
• Experimental synthesis: Synthesizability (can you make it in the real world) is still difficult to predict well, and getting real life confirmation is an expensive process with limited bandwidth. As a result, after screening candidates, teams take a small subset (usually <50) and try to synthesize them in the real world. Creating more data and more powerful models should start to create better computational accuracy for this expensive step.
• Various stages of real-world testing: Remaining candidate molecules go through various stages of testing. First, small batch testing is done to assess the material’s stability and basic functionality. After that, laboratory tests are conducted to measure the physical properties of the material. The next step is bench-scale testing, which are slightly larger scale tests to assess scalability. A series of pilot scale tests (~25% of expected full scale volume) are then completed to further understand economic feasibility, as well as various real-world performance, compliance, and safety tests. Closing the loop to the computational screening should make that step increasingly accurate.
• Scaling production: Once all testing has been completed, the final material(s) are ready for full-scale production and commercialization

Though meaningful, advances in generative AI models and molecular simulation models have thus far been limited in practical impact because 1) they sit siloed across research labs and 2) lack the data that industry players own, and would require data that doesn’t exist yet, meaning there is no feedback loop across the cycle. Cusp AI’s thesis is that by building an integrated platform, they can drastically improve the efficiency of the candidate generation process and reduce the cost while increasing the speed of the R&D process, and start creating predictions on synthesizability.

This has the potential to capture significant value because new molecules are patentable. So if Cusp identifies a valuable new molecule, they can protect this IP and monetize by licensing this technology to industry partners. One success alone can be very valuable.

Target market

Cusp is initially targeting the carbon capture market, hoping to generate novel materials that will enable economically feasible carbon capture (direct air capture, “DAC”, as opposed to capture in a factory’s process, that is much easier to do). They have chosen this in part due to what they see as an unfair advantage given their team’s backgrounds (see below). Additionally, they see investment momentum in the space despite the fact that existing material technologies are generally expensive and unreliable.

The underlying materials, a class of polymers known as metal organic frameworks (MOFs), are also used in hydrogen storage, catalysis, water desalination, and other use cases. The current platform is directly generalizable to these use cases. However, when expanding to new classes of materials, some additional tailoring will be required.

DAC Market Learnings

• Even though DAC is still nascent (needing “multiple miracles”) the tailwinds in the industry can help them to de-risk and get early revenues despite the market / technical uncertainty
• Microsoft has publicly committed to “higher quality carbon credits” and is negotiating a commitment with Cusp to acquire $380m of credits at $800 / ton (vs $30 or so in “greenwashing” credits) once technically viable. The deal will be shaped in the next 1-2 months and closed in April 2025, and will combine collaboration, compute, commitment of the credits, a potential up-front payment, and an investment
• It is thought that if they get Microsoft on board in this fashion, multiple hyper scalers might follow, especially Google and Meta. Several players want to be potential future leaders in DAC, most importantly the oil majors. Chad did deals with Total, and Berend worked for a long time at and with Shell, so they believe they could structure a deal that would include up-front payment and defraying of the development cost, if a Microsoft commitment is on the table
• Other important players could be the chemical innovators, like BASF, engineering firms like Siemens, or scale ups like Carbon Capture Inc that the team is talking to as well
• Next to DAC there are a number of other fields where MOFs play an important role, from very adjacent carbon capture at source, to the equally speculative hydrogen storage or much more mainstream like chip manufacturing and gas separation.