I'd count this as a fun fact, but most of these AI startup "fundraises" are GPU swaps where most of the money is in compute and that serves as collatoral, so relatively little of it goes to the actual hard sciences.
I'm fully onboard for hoping AI funds a hard science revolution and am happy Charles is writing about this, but we need to see the balance sheets before getting too existed!
I'm much less bullish than you are here, but hopefully I can be proven wrong.
I wouldn't pattern match what is happening in the broader LLM space with the cost profile for autonomous labs - I'm sure GPU costs are a part of it, but autonomous labs are very CapEx heavy on lab equipment and robotics.
If you look at Lila AI's careers board (https://www.lila.ai/open-roles) they are hiring for AI scientists (mostly non-LLM's e.g. GNN and materials data), but the vast majority of hiring is for scientists, technicians, and operators of expensive scientific equipment. Periodic Labs admittedly has a bit more of an AI + robotics approach.
I can speak more directly to the smaller startups that have raised - in all of those cases, most of the development and capital cost is on the lab automation side, which is still quite difficult and boutique (more to come on how the state of science instrument software is the primary bottleneck to autonomous labs, at least in my view).
To be clear, it is not obvious to me that any of these companies will work either or that we will have a science revolution because of these investments. My goal was simply to provide a sense of scale for how the AI bubble is touching this particular aspect of the economy.
Great post. Just saw that September one where you mention automation-native scientific instruments. Do you have an idea of the size and nature of R&D capex for the instrument makers? Like what are Thermo Fisher, JEOL, etc. doing? It feels like offering a MCP server built on top of an open API would be such an easy way to gain market share.
another charles banger
I'd count this as a fun fact, but most of these AI startup "fundraises" are GPU swaps where most of the money is in compute and that serves as collatoral, so relatively little of it goes to the actual hard sciences.
I'm fully onboard for hoping AI funds a hard science revolution and am happy Charles is writing about this, but we need to see the balance sheets before getting too existed!
I'm much less bullish than you are here, but hopefully I can be proven wrong.
I wouldn't pattern match what is happening in the broader LLM space with the cost profile for autonomous labs - I'm sure GPU costs are a part of it, but autonomous labs are very CapEx heavy on lab equipment and robotics.
If you look at Lila AI's careers board (https://www.lila.ai/open-roles) they are hiring for AI scientists (mostly non-LLM's e.g. GNN and materials data), but the vast majority of hiring is for scientists, technicians, and operators of expensive scientific equipment. Periodic Labs admittedly has a bit more of an AI + robotics approach.
I can speak more directly to the smaller startups that have raised - in all of those cases, most of the development and capital cost is on the lab automation side, which is still quite difficult and boutique (more to come on how the state of science instrument software is the primary bottleneck to autonomous labs, at least in my view).
To be clear, it is not obvious to me that any of these companies will work either or that we will have a science revolution because of these investments. My goal was simply to provide a sense of scale for how the AI bubble is touching this particular aspect of the economy.
Great post. Just saw that September one where you mention automation-native scientific instruments. Do you have an idea of the size and nature of R&D capex for the instrument makers? Like what are Thermo Fisher, JEOL, etc. doing? It feels like offering a MCP server built on top of an open API would be such an easy way to gain market share.