Business idea: on-prem data center installs

 In light of all the concerns about data center installs (and mounting debt, community pushback, electricity interconnect issues): some companies (like Revolut) have opted to self-host open-source AI models. Once (a) the bills for Anthropic and OpenAI start to get over tens of millions of dollars and (b) the open-source models are good enough for most needs, doesn't it soon start to figure out how to cut frontier model spend?

What's old is new again -- perhaps more large companies will look to de-cloud their AI spend. The business idea is to be an "easy button" for companies to plan and install this. Have strong connections/purchasing power to Nvidia (or other GPU companies - TPUs?), have strong advice on number of chips, locations of data centers, models to use, model upgrade schedules; maybe have an API that allows you to toggle between LLM models (on-prem vs. maybe sometimes you do need/want the frontier model). Maybe a longer-term services/recurring component too. Make it as easy for a Fortune 500 company to install as Anthropic, but own none of the assets yourself (and push the capex budget onto the company itself). A capex-light, project-finance-y play.

Two related things: 

1. About 9 months ago, I was doing some research into a wider movement for companies to move their data off cloud to self-hosted data centers. Reasons mostly cost and privacy. A colleague thought this idea was sacrilege ("everything is moving to the cloud, not away!") ... but who knows? Maybe this will cycle back.

2. I was talking to another start-up who envisioned that larger companies could have local power banks and data center resources. (E.g. your corporate HQ would have a mini data center in their basement, and a couple of large batteries outside.) This "distributes" the data center resources, which would have more community support than the monolithic, jet-engine-powered data centers of today.

Anti-AI writing

In high school, I remember a tension in how much hip new slang to use. Too much, or too late, or too little, and you’re uncool. In college, there’s a similar tension — for complex topics (say a debate on politics), you want to show you’ve done the homework, but don’t want to parrot someone else’s words. (Side note: this is probably why Twitter became so big — you could get the unpolished kernel of an idea out there without sweating about making it sound smart.)

I still write most of my stuff by hand today, so when I write, I find myself thinking: “how do I make it clear that I wrote this myself?” Brevity and clarity. Slightly unpolished. Interesting, non-linear “shower time” thoughts. Choosing to bring in more backstory/history into an industry background than strictly needed. Annotations about people that an AI wouldn’t pick up (eg “great founder vibes for XYZ reason”). 

Some of these are what I hope are seen as “my style” of writing/thinking, some are areas that I recognize AI cannot touch, while others are intentionally anti-AI. Example: partial sentences in emails, that my high school grammar teacher would tsk at. Sprinkling funky (but accurate) words here and there. (Fun fact not on resume: I cared a lot about writing in high school — read Orwell’s and Stephen King’s Why I Write and a few others — and was briefly an MBA application editor right after graduating college, despite my heavy sciences resume.)

If my career depends on writing and thinking independently, I do think it starts with simple things like diction and syntax. Again, may be specific to me and my pursuits, but it does feel like the only way to “outrun” AI.

E-comm agents -- are they real?

I feel like I don't "get" the whole idea of e-commerce agents. I've always struggled with: why do I need a personalized shopping agent to pick and buy stuff for me? (Can I really trust it to buy the right thing?)

The big question: is it something about me and my shopping behavior, or is this shared? Some discrete examples (and some thoughts/hesitations):

• Groceries -- for a while we've now been able to buy groceries with a (human) agent (e.g. via DoorDash). I haven't used it often. A few reasons: (a) it costs too much, (b) I don't trust the buyer (who is incentivized for speed) to pick the right produce, (c) I enjoy the "randomness" that comes in shopping in a store sometimes -- random things on sale, a fruit or item that catches your eye. We did buy DoorDash with Costco while we lived in New Haven, in part because Costco was 20 minutes away. 
• Travel -- the hard part about travel is not necessarily booking the tickets, but the coordination? I need to coordinate with my wife (and often other friends, family, etc.) to see if a given date works, if people can take the time off from work, etc. etc. Yes, booking flights and hotels is a pain, but it's not the key friction. Can the agent coordinate with friends and family and make them decide more quickly?
• Clothing -- in the past 10 years, I've shifted towards "slower" buying -- buying more expensive, quality pieces of clothing but knowing they will last a while. I'll look on Facebook marketplace, and pull the trigger only after a few weeks or months of mulling it in my head. ("If I still want it after 2 weeks, then I actually want it.") Can agents really help me? (And if it tries to suggest a new brand, ... can I really trust it?)
• Home commodities -- things like toilet paper can be re-ordered automatically via Amazon (I haven't done it, but my wife has ... so our household has (dumb) "agents" buying household commodities already). There's not a ton of differentiation in price or quality, and buying the brand name doesn't hurt the wallet (all smaller-ticket items).

It is very likely that I'm an old man who doesn't impulse buy enough things, and so I just don't get it. And some of these e-comm agent companies will undoubtedly be successful (there is a lot of money in ads and e-comm). But ... I still hesitate to believe that in 5 years, I will be using e-comm for all my purchasing. After all, I still like going into the store to feel materials, try things on, etc. -- the same as we used to do in department stores in the days of Sears and JC Penny!

AI is an extension of strategy

This feels a bit business-school-y to say ... but broadly speaking, new technology can be used in one of a few ways:

1. Strategy enhancement -- the technology doubles-down on existing moats and rights to win

2. Create new strengths -- expand the business in new directions

3. Operational efficiency -- make admin tasks quicker

4. Required technology adoption -- must-have technology just to keep up the times

I've been spending more and more time with different ideas and strategies on how to adopt AI well at our investment shop. We've had the chance to talk with a few other investment firms (as well as dive into other folks' LinkedIn posts, hear start-up pitches, hear vendor pitches, etc.) 

From what I've seen, I believe that the most successful AI adopters will fill up bucket #1 -- AI will extend an existing (winning) strategy even further. This would be a useless post without a few examples:

1. One AI-forward venture shop's core strategy is early-early-stage AI start-ups, targeting companies before they're necessarily even formed. They built tools (with AI) that allows them to (a) read, digest, and map the latest computer science papers and (b) track and rank people (e.g. X person left Y company, another person's stock is likely fully vested on Z months after being acquired, etc.). It feels like they are a step ahead of the masses (and a small step ahead of tools like Harmonic), allowing them to get a narrow lead in sourcing.

2. I've also built a few tools (using AI) for myself that doubles down on our core strategy at Connecticut Innovations. Our biggest constraint is that companies have a presence in the state, so I found a source (CT Business Registry) that gives us a good jumping off point. Layer in a few other AI assistance, and this has become a renewed source of sourcing alpha.

3. One much more contentious area is investment memo writing (of which I've written about before). I've seen venture shops with great exited companies have lengthy, almost unreadably-long memos generated with a heavy hand of AI. I'm sure there are also great, principled memos written about losing start-ups. 

Through the above lens, though, the question changes from "should we use AI to write memos" to "is writing an investment memo a core part of our investment strategy." I think for some places (like the Yale Endowment), to write is to think (a sentiment that I espouse). But, perhaps, the difference is one of strategy, not capability -- they intend to be some of the most patient capital, really hoping to partner over the next 20+ years. If it takes an extra 6-12 months to do diligence (and to write a memo and grapple with minutiae and diction and brevity), then so be it. 

Venture, especially now, may be different. In some AI deals, there is no time to think or do deep deep diligence; you must see and believe the uber-high-growth story. Perhaps here AI memos are good. Or perhaps for a single GP whose core strength is their network and business building, memos are a necessary evil, an artifact produced out of necessity (but not part of the alpha). Through this lens, memos are operational, not a strategy. 

I, of course, disagree, especially where I am today. I'd like to be a good investor (not just a good venture investor), and so the painstaking writing is worth it. 

Digital therapeutics are challenging for VC

 Recently did some digging into a digital therapeutic company (non-telehealth). I've heard the sector is very challenging, but after digging, good to put some concrete thoughts behind it:

• Digital therapeutics require all the same things that bio therapeutics require -- regulatory approval (albeit fewer trials than bio) and solid clinical evidence, but then also payer adoption and provider prescribing. 
• Payer adoption is hard (and where players like Akhili failed) -- you need to show clinical proof of medical necessity for payers to want to cover the digital therapeutic. This feels like valley of death number 1.
• Even if payers cover it, you need to get providers to prescribe it. Which means a sales/marketing team and a fleet of account managers, similar to pharmaceuticals but with a more untested care modality (digital vs. drug). Can be expensive and slow. 
• Provider un-enthusiasm. A more speculative observation that heads into behavioral aspects of healthcare, but I believe that (a) doctors understand how drugs work and are generally not hesitant to prescribe them if the benefit is clear, but (b) doctors don't really believe that digital therapeutics work. I think in general, many providers are slower to change prescribing practices, but especially slow when confronted with a totally new modality of care (digital vs pharma). It might be the sort of thing that's a generational change (like getting comfortable typing notes into a computer instead of handwriting), not a multiyear change.
• Digital therapeutic adoption/compliance. Most pharmaceuticals you take a few times a week, then forget about it -- it takes less than a minute to take a pill (and still compliance rates are sub-par). However, many digital therapeutics have a longer duration -- e.g. "watch X thing for 1 hour a day, 5 days/week". To adopt it well, patients need to change behaviors and habits, which can be difficult. (This thought needs a little more workshopping, but prescribing self-directed therapies -- like many digital therapeutics, like prescribing exercise -- feels harder. Would love for some data to refute/support this claim.)

AI will never have all the data

Some comments you hear after a pitch include things like "the founder was kind of annoying" and "man, that was hard to sit through, couldn't imagine being on that board." Strongly worded, yes, but honest (and often directionally accurate). No investor ever would want to put those words into any internal database.

Similarly: few firms require that you write a memo for investments you didn't make. (I've been trying to do this for a few that I've dug deep on but ultimately passed on -- but it takes a lot of time!) This valuable data is not captured anywhere -- how can you try to build an "investment analysis" AI if you don't have these "negative" data points? Will there not some weird bias inherent in the system?

Short of a neural tap (of which I hope I am not alive for), this information will never get recorded, and machines can only learn on things that ... exist somewhere. Transient but critically important thoughts will struggle to find their way to recorded data points, for better and for worse. 

Who's done the reading?

Back in high school, one of the core challenges a teacher needed to figure out was: who has done the reading? I remember one of my English teachers doing a pop quiz, that asked a simple question about a short story that tested: did you actually do the assigned reading? It's easy to fake it (or just be quiet) in class, and that was a good (or at least, memorable) solution. 

I'm reminded of this in two cases: (1) Reddit posts (and other similar comment-driven forums like Twitter) - I'd guess a slim majority actually read the article, and most read the comments, and (2) LLM content - I suspect most people are content with the LLM output and seldom read the underlying content in full. (And for #2: I think this is okay in many circumstances -- most LLM use cases don't demand information perfection.)

But with the age of AI comes a similar question to my high school days: who is doing the underlying readings, who is doing the hard work to try to understand and really dive in, who is using their brain as the primary processing engine (and AI primarily as a data gatherer)? Perhaps no surprise, I think I straddle this line well, but I believe others don't. For example, I see others using AI to read through a deck to help generate diligence questions, and others throwing large prompts into AI to generate content that they then edit down. Is it inherently bad? No, but I think it can erode the ability (and urge) to struggle with thinking. 

If we assume the above premise is correct (that some people misuse AI as a shortcut to thinking), the big question becomes: how can you tell who is who? And does it matter? "Surface readers" will provide equal value in many scenarios, and in some disciplines (like pre-seed venture), it may not even matter -- being on top of the current trends may matter more than, say, digging into the unit economics. But, if we assume it is a problem AND that it matters, then what questions or conversations do you have to figure out who did the readings, who's asking good questions? No good answers yet, but something that kicks around in the back of my mind as I think through what a good, modern, AI-infused investor looks like.