Keep pulling the thread on Anj Midha.
Anj Midha's incubation, Periodic Labs, uses LLMs to predict new materials, including superconductors.
In the domain of material science, applying more compute to AI models is currently yielding super-exponential gains per iteration.
A primary reason for poor AI model performance in science is the lack of specialized physics and chemistry data on the public internet, which is mostly composed of text and code.
The US CLOUD Act requires American companies to provide the U.S. government with access to data on their cloud infrastructure, regardless of the data's physical location.
Due to the US CLOUD Act, European companies with mission-critical workloads are seeking local AI infrastructure providers instead of using American hyperscalers like AWS, GCP, or Azure.
For the first time in 15 years, the dominance of cloud hyperscalers is being challenged by startups, driven by data sovereignty requirements.
The core investment thesis for Mistral is to build a fully independent, sovereign AI infrastructure stack in Europe.
Anthropic initially attempted to raise a $500 million seed round but later re-anchored to raise $100 million.
Anj Midha's company, AMP, is structured as a public benefit corporation and provides most of its compute infrastructure to partners at cost.
AMP has secured approximately 1.3 gigawatts of compute infrastructure, which represents about $40 billion in cloud spending over four years.
The financing for AMP's 1.3 gigawatts of compute consists of approximately 20% equity and 80% debt.
Anj Midha estimates Google's current compute infrastructure capacity is between 12 to 15 gigawatts.