Variant invests in technology that expands autonomy—giving users more agency through increased access, knowledge, and ownership. Access to knowledge has always been a bottleneck to progress, particularly in specialized domains, because it stays with the rare individuals who hold it. The handful of researchers who can answer hard technical questions remain siloed and inaccessible to most people and businesses.
A solution we’ve been excited about is domain-specific AI models: models optimized for reasoning within specific domains. As we described in our piece When Domain-Specific Models Win, these models can outcompete frontier labs building generalized LLMs when reasoning within that domain is sufficiently idiosyncratic. In these scenarios, the advantage frontier labs have from economies of scale is neutralized because a smaller, domain-specific model requires less compute/cost and the specialized architecture can enable higher quality. Domain-specific models don’t rely on economies of scale like the frontier labs; they build their moat through cornering verifiability. That is, they create proprietary means to objectively measure the quality of the model’s output and compound its effectiveness over time.
Our investment in Physical Superintelligence (PSI) is premised on our belief that solving hard physics problems represents an instance where domain-specific models will win.
PSI is the first vertically integrated factory for physics discoveries. Its system orchestrates submodels spanning foundational subdomains in physics to understand, model, and solve the most difficult physics problems. PSI deploys this system into large enterprises and research labs bottlenecked by exactly these problems. For example, some of its initial customers are data centers, which can save massive amounts of capital by optimizing their cooling systems and energy usage. Through deployments like these, PSI can verify its outputs against valuable, proprietary “ground truth” data to improve its core system for future use. It can then use this same core system to conduct independent research, what the team calls “physics mining” — solving previously unsolved physics problems to generate new IP. By combining experimental validation with first-mover advantage in new discoveries, PSI can turn its focus on physics into a compounding advantage.
This team was made to build this. CEO Matt Pines spent a decade advising Congress, the White House, and the intelligence community on emerging technology and served as Director of Intelligence at SentinelOne. He also has deep roots in the Bitcoin mining community from his time as executive director of the Bitcoin Policy Institute, which gives the team an edge in servicing miners-turned-data centers as a critical customer in their go-to-market. Dr. Alex Wissner-Gross, cofounder and Chief Scientist, brings deep physics expertise. He became the last triple major from MIT, where he graduated first in his class as a Marshall Scholar. He has since completed his physics PhD at Harvard, holds 26 patents, and has backed and advised more than 40 technology companies. He was also appointed by the Office of the Secretary of Defense to build the country’s largest AI-based all-domain command and control platform. Alex Klokus, cofounder and President, brings a decade of bootstrapping and team building, having previously founded Futurism and is now the founder and managing partner of the SALT Fund. Together, they have recruited a stellar team of physicists, engineers, and AI researchers with roots at Google, OpenAI, Harvard, NVIDIA, MIT, Stanford, Oxford, and beyond.
We couldn’t be more excited to back them.
Disclaimer
This content is provided for informational purposes only and does not constitute investment advice. Please refer to https://variant.fund/disclosures for additional details.