Good thoughts Ryan. As a startup that has been working in Generative AI before it was called that, a key learning was the shift away from cloud. As a small startup, pretty bootstrapped, we decided to on-prem our GPU's, thanks to our relationship with Nvidia - the FM/Generative AI world may trigger such a transition as cloud costs become unbearable. Obviously other factors including power consumption costs will need to be factored along with maintenance of GPU's, however we have not looked back since getting off the cloud for our GPU needs.
good article. I think there are a lot of AI-first companies out there, not just FM companies, that have been caught being compared to SaaS companies by VCs. I collected data from my peers two years ago and it showed that, on average, it took an AI-first company $27M to get to $1M ARR. A capital efficient SaaS company needs less than $1M to get to $1M ARR. So the AI-first company shows up to the next round looking like the most capital inefficient business in history when in fact you've done a pretty job. One thing I would add is that the cost difference between an AI-first company and a typical SaaS company isn't just compute. I've found that AI-first companies need to have larger engineering teams (sometimes 2x-3x the size). And these extra engineers are the very expensive kind. so take the R&D expense of a typical SaaS company, double it, and THEN add on all the compute costs of building a FM.
important to note the different types of AI businesses and how they require varying levels of capital intensity. It's exciting to see how exponential growth can be unlocked with the right approach, but also important to note the high risk involved in investing in these capital-intensive startups.
Good thoughts Ryan. As a startup that has been working in Generative AI before it was called that, a key learning was the shift away from cloud. As a small startup, pretty bootstrapped, we decided to on-prem our GPU's, thanks to our relationship with Nvidia - the FM/Generative AI world may trigger such a transition as cloud costs become unbearable. Obviously other factors including power consumption costs will need to be factored along with maintenance of GPU's, however we have not looked back since getting off the cloud for our GPU needs.
good article. I think there are a lot of AI-first companies out there, not just FM companies, that have been caught being compared to SaaS companies by VCs. I collected data from my peers two years ago and it showed that, on average, it took an AI-first company $27M to get to $1M ARR. A capital efficient SaaS company needs less than $1M to get to $1M ARR. So the AI-first company shows up to the next round looking like the most capital inefficient business in history when in fact you've done a pretty job. One thing I would add is that the cost difference between an AI-first company and a typical SaaS company isn't just compute. I've found that AI-first companies need to have larger engineering teams (sometimes 2x-3x the size). And these extra engineers are the very expensive kind. so take the R&D expense of a typical SaaS company, double it, and THEN add on all the compute costs of building a FM.
important to note the different types of AI businesses and how they require varying levels of capital intensity. It's exciting to see how exponential growth can be unlocked with the right approach, but also important to note the high risk involved in investing in these capital-intensive startups.
Great article again. Love the last bit on other business types trying to raise at FM model valuations.
With regards to valuation methods for FM models, I see some analogy with deep tech valuation (well explained here - https://medium.com/cantos-ventures/a-financial-argument-for-deep-tech-2b01bf6e663a)