I think the money in AI is in small, highly specialized AI firms that work on one very specific thing..
I spoke with the CEO of a small, defense focused AI company today that supports a very narrow need of the USAF.. they are making TONS of money.. and there is currently zero competition for their market space.. their product is somewhat unique in what it does and who it does it for..
Granted, once one of the big defense firms like SAIC, CACI, BAH, etc find out about it, they will offer the ownership team of this company 9 figures or more and they will cease to exist the next day.. they'll get absorbed into one of the mega-primes.. the strategic value of what they are doing is way to great not to be super attractive to the mega's.. it would get them straight into the offices of the most senior GO's in almost every department within the USAF.. and likely would have some appeal to the other services as well..
While this firm isnt publicly traded.. I'd love for them to let me pull even a small equity stake in their business.. there is practically nothing they could do to not all retire with many, many millions in their pockets from just this one software package they have put together..
I tend to agree.
A specific consumer need, a specific application, big enough to be a lucrative market, not so big that you just get crowded out by the big boys.
I'm generally of the opinion that these will be the players that survive in the AI space over the next decade, and are likely to be the ones with a lot of the market share, not the big boys trying to do 'all the AI, for all of the applications'. Build one product, generate a revenue stream, line extensions into adjacent categories once you have free cash flow to do so.
Your defense example is a good one, and I've also worked with AI companies (or machine learning companies, as this was 5 years ago before AI was cool) doing stuff like logistics network optimization, manufacturing process optimization, predictive maintenance modelling, forecasting, etc. Some start ups, some young companies recently acquired.
I also have friends doing medical consulting stuff who are talking to companies developing tools for MRI scan assessment, medical assistants for GPs, etc.
The other factor that to me is important, is the strategy for acquiring compute power.
I'm very, very concerned to see an AI company investing heavily in their own data centers. Sure, it's cheaper in the very long run, but you lose so much agility that I don't think it's smart in this early stage.
Far better for most to rent that capacity from Google, Amazon, Meta, let them eat the massive capex and the massive depreciation on quickly obsolescing (and currently underutilized) assets. Only pay for what you need today, and ensure you have contracts in place that require the partner to provide more compute on demand to cover actual growth.
Or alternatively, get the customer to pay for the hardware, and provide maintenance contracts only. Again, revenue generating contract, then infrastructure investment, not vice versa.
If AI truly does boom, you can always build your own data center when you have long term profitable contracts in place to ensure good utilization.
