Foundation models for brain data are a hot topic in the neurotech world.

Hemispheric and its $52M Series A raise catalyzed a lot of hype and assumptions around scaling laws applied to EEG data without any substantial discourse around the obvious representation problem that exists with EEG data. (This was my impression based on the people I spoke with who cited Hemispheric and made erroneous generalizations about what their work meant for neurotech writ large; I never spoke with Hemispheric directly).

Companies like Piramidal, Alljoined, Constellation, and Conduit are just some of the many companies leveraging brain data en masse to build foundation models with interesting applications. Data collection > models > applications. There’s a race to the top in terms of who can collect the most data the quickest.

But I remain puzzled for one reason: Neuroscience as a Service (NaaS) is not an established market yet. While I’m excited about all of these companies, I have seen first hand just how much effort is required to get a deal going with a corporate partner and a neurotech. Some founders talk to me about their GTM assumptions, which are very clearly unsubstantiated. Assuming that the market will come to you if you collect enough data is the wrong move here. You have to be trailblazing on the corp dev & BD side of things simultaneously - just like any other startup, distribution is 90% of the battle.

I’m further perturbed by the lack of progress in addressing the representation problem in EEG. I’m optimistic that some companies are using modalities other than EEG (I know some are using td-fNIRS), but I still don’t understand how to make the venture case for these models (i.e. generating a significant return in less than ten years) when the hardware moat and physics/engineering insights aren’t owned outright by the company. I recently parted ways with Kernel, but I’m still extremely bullish on the tech and have had many conversations about how this company could come out on top as the NVIDIA of neurotech.

Kernel was able to accomplish so much over the past 3 years and much of these accomplishments have been under-publicized (which I’m thinking about resolving with my own writing). The company has spent years pioneering what the NaaS offering is. I’m not quite sure anyone else has. The institutional knowledge that comes with building a completely new modality for brain data collection and trying to commercialize it across several verticals at once is immense.

I’m super bullish on new modalities in neurotech/BCI where the company owns the hardware IP. Hoping to gain a better understanding of who’s who in this segment over the next few months.

~Sam

Original LinkedIn post

Original X post