($) 2025: Year of (Software) Picks and Shovels
API platforms, databases, and the neoclouds.
This post is for our premium members only. If you are reading this in full, thank you for being an Interconnected Premium member! If you aren’t, I hope you become one by scrolling down and tear down that paywall! (Better still, gift our premium subscription to a few curious friends, who would enjoy my writing in this holiday season!)
Recently, I joined a couple of podcasts to talk about open source AI and the US-China tech competition. If you are interested, please give either this Lawfare podcast episode or
’s NatSec Tech podcast (or both) a listen!When I tell people that Interconnected Capital (my hedge fund) is focused exclusively on the “picks and shovels of AI”, most assume I’m a hardware investor. This assumption is doubly confirmed by the fact that I also write a lot in this newsletter about GPUs, semiconductor manufacturing, etc. So when I share with them that the majority of my portfolio is in software names, and doing very well, they are quite surprised!
Of course, their assumption and their surprise both make sense. In the last three years, anyone who is investing in AI in the stock market, whether based on expertise or out of FOMO, has rightfully focused on the “hardware picks and shovels” – chips, foundries and equipment that make those chips, server racks and cooling systems, and even a layer below into the world of energy. The level of investment into these atomic layers of the AI infrastructure (stuff you can feel and touch) won’t stop; every large tech company is still “investing full-tilt” as the ever quotable CEO of Broadcom, Hock Tan, likes to say.
However, the bits and bytes layer of AI infrastructure, or the “software picks and shovels”, will be stepping into the limelight next year as the building blocks of AI applications. These applications have to create enough value in the real world to justify the massive hardware spending that will continue until the end of this decade.
The operative hype word is “agent”; everything in AI has taken on an agentic flavor. Some agents will amount to nothing more than cute demos or vaporware, while other agents will solve hard problems better than most human beings can or drastically reduce the cost of many rote tasks that many people are currently getting paid to do. It is currently too hard to figure out which agents for which tasks or problems will stand the test of time. It is a bit less hard, however, to figure out what types of software infrastructure products must be relied on to build these agents and throw them into the wild to find their “agent-market-fit”.
So in this (most likely) last post of the year, I’d like to share what I think are the three categories of software picks and shovels that will become more valuable: API platforms, databases (the right kind), the GPU neoclouds (with software chops), and how to spot the winners. Each category has many pure plays in the public market and even more in the private market. Each category should occupy more investor mindshare in the agentic year of 2025. Let’s dig in…