AI & Materials Discovery - Part 1: Four Paths to the Frontier
Much has been penned already on the explosive growth of AI. The capital, land, energy, and human resources devoted to AI are staggering. Global AI investment in 2015 was $12.75B. This year it has surpassed $350B.
Nestled within the litany of AI developments is an area we at Pangaea Ventures are following closely. AI for the discovery of new materials (something we highlighted in our Hard Tech Report earlier this year), is promising to accelerate the development of better batteries, stronger alloys, novel catalysts, more effective biomolecules, and sustainable alternatives to everything from plastics to concrete.
Yes, AI is primed to exponentially accelerate and expand the transformative potential of hard tech, and investors are taking note to the tune of billions invested in this opportunity in 2025 alone.
But of course, this isn’t a new story. Bleeding-edge computational methods have promised to revolutionize materials discovery for decades, with plenty of cautionary tales along the way.
Given this, there are two questions we’re interested in exploring:
Does the most recent generation of foundational AI models truly enable a new paradigm in terms of materials discovery?
Can these companies do anything fundamentally different from a business model perspective to capture meaningful value for themselves and for investors?
Before we answer these questions in Part 2, let’s first take a look at the most prominent approaches to AI-driven materials discovery.
Let’s dive in.
AI Platforms for Materials Discovery
Before we tackle the questions above, let’s take a look at the prominent approaches to AI-driven materials discovery. We can largely segment the technologies into four different approaches, each with distinct capital requirements, timelines, and value capture strategies: software-only platforms; purpose-built hardware and software; product-specific companies; and full-stack hardware-enabled AI platforms.
Software-Only Platforms
The most crowded segment in the space, software-only platforms are just as they sound: companies develop and provide AI-enabled software and machine learning (ML) tools (often combined with access to material databases) and predictive modelling capabilities delivered as a SaaS product.
As with any SaaS business, margins can be high (70-90%), aided by the ability to scale without incurring lab infrastructure or material development costs. Startups can serve multiple customers, industries and applications simultaneously, have faster development cycles, and can grow quickly through subscription revenue. Ideally, these companies should not require substantial amounts of outside capital to reach profitability.
There are a number of earlier stage companies seeking their first institutional rounds, and there have been a handful of successful outcomes, including Vancouver BC’s Good Chemistry Co., acquired by SandboxAQ in 2024 for a reported $75M, and Schrödinger, Inc., a NY-based company founded in the 1990s that went public in 2020.
The economics of today’s most advanced AI poses a real threat to the software-only model. Recent analysis of reasoning-based AI reveals that computational costs can far exceed traditional SaaS subscription revenues, forcing companies to either absorb losses or dramatically raise prices. Materials discovery platforms adopting these more sophisticated models will face the same dilemma: forgo the most powerful AI capabilities, or pass higher costs on to customers already wary of expensive, uncertain R&D tools. This cost-capability mismatch could potentially undermine the capital-efficient promise of the software-only model.
Purpose-Built Hardware and Software
This segment takes the previous, and goes a step further. The companies in this segment have layered on modest-scale, purpose-built experimental capabilities (think bench-scale robotics or small validation labs), with an eye towards internally generating proprietary data which can be used to train and validate material property predictions.
Although there have been some recent mega rounds with grandiose promises of creating revolutionary new materials, the reality is most have fallen into the prickly trap of becoming glorified R&D companies. High capital requirements for equipment and personnel have further eroded margins.
In spite of a deluge of customers willing to engage in R&D projects, service revenue alone has not been able to sustain the companies developing costly solutions like self-driving labs (i.e., automated setups that use robotics and automation to run experiments and iterate on results discoveries, with limited human intervention). Simply accelerating discovery of a new material delivers limited value in and of itself. The real value comes downstream, in development, scale-up, engineering, and deployment.
Product Specific Companies
This category encompasses startups convinced the best use of AI in materials discovery is to develop, manufacture, scale up and productize specific materials themselves, rather than license their platform or provide R&D services. These companies use AI internally to accelerate material development and aim to capture value by vertically integrating into a particular market, often landing on a fairly narrow vertical. In other words, these companies are attempting to solve for the market challenges of the Purpose-Built Hardware and Software segment.
The advantage is crystal clear: these companies are typically aiming to enter established markets with proven pathways to scale-up and revenue. They can build physical IP moats within their chosen market and generate revenue without depending on customer adoption of a platform, or attempting to negotiate IP rights before actually making anything.
The drawback of course is that AI does not necessarily mitigate the high costs and protracted timelines of bringing new materials to market. Indeed, once a startup commits to a particular market, and scaling and manufacturing costs start to dominate, it faces a critical dilemma of whether or not to continue investing in the underlying discovery platform. This tension is intensified in difficult fundraising environments where capital efficiency and focus become paramount. The discovery platform that justified initial investment may become an expensive distraction from the operational challenges of actually producing materials at scale.
Hardware Enabled AI Platforms
Finally, we have the segment of extremely well-capitalized companies building full-stack systems that integrate AI models with extensive robotic automation, high-throughput synthesis, and characterization equipment. These companies each claim that their foundational generative AI models can help achieve an unprecedented level of scientific intelligence, creating massive amounts of proprietary data to continuously improve their AI models and unearth new performance materials that have never been identified or synthesized before. These are what we’d call breakthrough innovations (if they pan out).
While there’s a range of well-funded startups advancing today, they might benefit from a glance to the past. Take San Jose-based Intermolecular, which was founded in 2004 and pioneered “high productivity combinatorial” (HPC) technology for semiconductor materials discovery. The company raised over $60M across three venture rounds before going public in 2011. It raised almost $100M and reached a market capitalization of over $400M. Despite technical success and over 10 years of serving the semiconductor industry, Intermolecular struggled to achieve strong economics as a standalone business, and in 2019 was acquired by Merck for just $62M. Companies today should take note that even in a high value market like semiconductors, accelerating R&D does not automatically translate into a highly profitable and sustainable business model.
The likelihood of success for hardware-enabled AI platforms varies significantly by target market.
Drug discovery and formulation has, maybe, the highest probability of translating discoveries into early revenue because novel molecule or formulation IP is valuable to pharmaceutical companies in advance of scale up. We dove into this (and many other subsets of hard tech) in our Hard Tech report.
Decarbonization materials such as electrocatalysts, active battery components and energy storage materials face a much lower probability of success because new materials face long commercialization horizons in markets where commodities dominate, and incremental improvements are of little value.
Performance materials, semiconductors and alloys face similar qualification and scale-up challenges, but may yield some initial success in niche markets that are not as price sensitive. However, the trajectory of companies like Intermolecular should serve as a warning.
Perhaps one of the only categories where a new material would be truly game changing at the discovery phase is high temperature superconductors, which could unlock new paradigms across electricity transmission, computing, and medical technology. This moonshot philosophy is what underpins the $1B+ valuation in Periodic Labs, where the risk and reward are both in the extreme.
Up Next
In Part 2, we’ll get to the heart of the matter, and assess whether or not any of these technology segments will truly unlock a new paradigm in materials discovery, and if indeed any can be considered attractive from a venture perspective.
The content provided in this blog post is for informational and educational purposes only and does not constitute investment advice, financial advice, or a recommendation to buy or sell any securities or financial instruments. No information herein should be relied upon as a substitute for professional investment advice tailored to your individual or corporate circumstances. Always consult with a qualified financial advisor or other professional before making any investment decisions. Investing involves risk, including the possible loss of principal. Past performance is not indicative of future results.