The Hidden Risk With AI

Remember when Uber was cheap? When Uber first launched they subsidised rides in order to build up their customer base, mature their platform and most importantly outcompete local cabs. Now Uber basically own the 'taxi' market and even as a Londoner I never get in a black cab.

But Ubers are now as expensive as taxis and in some places even more so. The playbook is simple: take over a legacy market with a mixture of technology and lower prices, operate at a loss but run your competitors out of business then raise prices as an effective monopoly. This is the disruptor model and companies like Spotify, Airbnb and, most depressingly, Ticketmaster did this to great effect.

Now we're seeing this with AI companies. A lot of junior positions have been eliminated already because a senior employee with an AI agent does more work than a senior and junior employee does without an agent. But the companies that build the AIs are operating at a loss. Anthropic and OpenAI are not profitable Nvidia is because it makes hardware and Meta, Google and Microsoft all heavily subsidise their AI operations.

Right now that is keeping tokens cheap and businesses (and in some cases individuals) are becoming dependent on using AI tokens to function. Eventually the AI industry will have become profitable so tokens will become more expensive; and this might happen sooner than expected. The war in Iran is almost certainly going to cause a global recession, a huge el nino is forming which could cause global disruption and AI itself is already causing unemployment (the irony). Thus the amount being invested globally will drop and the big AI companies may well lose their ability to operate at a loss this year with the cost of tokens then rising accordingly.

If this happens any business that is dependent on AI is somewhat screwed. Companies will be forced to reduce their token usage and scramble to hire people to run the operations that they previously tried to automate. Some won't be able to and will fold.

This all happening this year is perhaps unlikely but sooner or later token prices will rise. So how do you insulate your business from this eventuality? Firstly, if you can, buy your own server, build your own models and run them locally. That won't make sense for most businesses so the best thing to do is make sure AI is additive not a dependency. Make sure your staff can still do their jobs without the help of an agent and that you can still run core business functions with a reduced token count. Juniors are a great way to mitigate the risk as you have cheap staff in house who are improving so they can step into a more functions as token usage decreases.

I'm very much a proponent of AI and believe that it can help businesses to the point where my company helps small businesses get their infrastructure AI ready. But we should all be careful that we don't build our business that is completely dependant on something we do not control.

I'm curious to hear how else businesses can prepare for tokens becoming more expensive.