GPT-3 & Capital Requirements

#Tech Twitter is abuzz about #GPT3. It does not take long to understand why. Just watch a few of these videos and you’ll instantly have a visceral reaction.

All of the above are just one use case: using GPT-3 to develop web applications. It’s use cases are much broader. As OpenAI shares on its blog, “Given any text prompt, the API will return a text completion, attempting to match the pattern you gave it. You can “program” it by showing it just a few examples of what you’d like it to do; its success generally varies depending on how complex the task is.” The following tweetstorm is an example:

The implications of what we are seeing are massive. But I’m focusing on the impact on capital requirements when it comes to starting and running a business.

The matrix below summarizes the drivers of capital requirements across 4 eras:

The drivers are:

  • Data Generation: How is software made and what are the associated costs?
  • Replicating a Unit of Data: How does software get replicated for preparation of distribution?
  • Distribution: How does software get to the customers who will use it?

The four eras I’ve defined are:

  • Floppy Drive/CD-ROM: When Microsoft Windows ruled the world and you would receive a new AOL disk in the mail every few months. In this era, companies still had large capital requirements to hire humans to write code, print the units of code onto disks for distribution, and then pay for physical distribution of those disks. Big Impact: We aren’t making heavy machinery anymore.
  • Internet 1.0: When we began getting applications sent to us over the internet. Still, there were rather high capital requirements from needing to hire humans to write code then to buy and manage your own servers. Big Impact: Distribution costs begin to decrease. No more CDs!
  • Internet 2.0: More open infrastructure and APIs help enable faster production of software and lower cost distribution. Companies could simply rent space on AWS. Capital requirements came down dramatically, especially those needed to run the business after product-market fit was found and demand began to be realized. Very high contribution margins are realized and that can usually flow through into large gross margins. The upfront capital requirements to write the code remain, but are slightly offset by APIs enabling developers to get a head start on what they are building. After this upfront investment, the major cost center becomes sales, marketing, and — depending on the service — support. Big Impact: Upfront R&D is reduced and server costs come way down.
  • Post AI: If building an application becomes as simple as a google search, then the upfront capital requirements for a software company begin to approach nil. Potentially, humans won’t be needed to sell or market because AI will just do that. Big Impact: Upfront R&D becomes near zero and distribution approaches zero.

In each era, at least one center of capital requirement dramatically reduced which in turn changed how companies are funded (i.e.; how much money it takes, profile of investors, etc) and how they scaled. Companies went from needing lots of capital to just get started to far less, especially for those started by technical founders. Going forward, capital requirements will continue to decrease and will do so for a larger portion of entrepreneurs, including those who lack the technical skills to write their own code. So in short, we are getting closer to a period where if you have a good idea, then very little will stop you from testing it out. While there are people who may choose to use this newfound power for bad, I do think it will be a net positive for entrepreneurialism.

I’ll try to follow-up with another post that shares some thoughts on how this might impact the asset classes that fund these companies (e.g., angels/VC/PE).

My posts are insightful 6 days a week. Then Giants games happen on Sundays.