I think this is a really good take, and not one I’ve seen mentioned a lot. Pre-Internet (the world Microsoft was started for), the man expense for a software company was R&D. Once the code was written, it was all profit. You’d have some level of maintenance and new features, but really - the cost of sale was super low.

In the Internet age (the likes of Google and Netflix), it’s not much different, but now the cost of doing business is increased to include data centers, power, and bandwidth - we’re talking physical infrastructure. The cost of sale is now more expensive, but they can have significantly more users/customers.

For AI companies, these costs have only increased. Not only do they need the physical infrastructure, but that infrastructure is more expensive (RAM and GPUs) and power hungry. So it’s like the cost centers have gone up in expense by log-units. Yes, Anthropic and OpenAI can still access a huge potential customer base, but the cost of servicing each request is significantly more expensive. It’s hard to have a high profit margin when your costs are this high.

So what is a tech company founded in the 1970s to do? They were used to the profit margins from enterprise software licensing, and now they are trying to make a business case for answering AI requests as cheaply as possible. They are trying to move from low CapEx + low OpEx to and market that is high in both. I can’t see how they square this circle.

It’s probably time for Microsoft to acknowledge that they are a veteran company and stop trying to chase the market. It might be better to partner with a new AI company that is be better equipped to manage the risks than to try to force a solo AI product.