[blindprogrammer]

Unless you're not building graphics from scratch and competing with Nvidia, you're building just another wrapper that will soon be obsoleted by another company with billions of dollars.

In a gold rush, instead of selling large language models (LLMs), which are easily surpassed, sell the shovels, which are the graphic cards. Those are what give you the ultimate competitive advantage.

In the age of cars, rather than horses, you want to own the oil wells, not be the guy building the coffee cup holder.

The lower you sit in the technological stack, the more money, power, and leverage you will have in the age of AI.

[juststartguide]

I agree with you that the best way to make money consistently during a gold rush is to sell the shovels, but in the case of AI it's just impossible. There has been a GPU duopoly for like 20 years.

You could try making specialized chips for AI, like NPUs, but it requires a lot of money as well, as you'd need a big team of really smart ($$$) people.

And you're overestimating the speed and interest of big corporations. If you're making an application around an LLM (a wrapper) for a smaller market, Google/Alphabet or Microsoft or whatever isn't going to release a product which will make your company obsolete.

Not because they lack resources, but because many markets simply aren't interesting for companies of this size. And even if the market was attractive, giants move slower than startups, even if they keep pushing the narrative of being just as agile as they were when were starting.

There's a reason why they spend billions every year acquiring smaller companies.