The price of AI capability is falling at a rate that is genuinely hard to internalise. Performance that commanded a premium not long ago is now available far more cheaply, and the trend shows no sign of stopping. Most of the discussion treats this as good news — cheaper AI, more access, lower costs. It is good news for users. But I think it quietly threatens the business model of a large part of the software industry, and not enough people are thinking through what comes next.
Here is the tension. A great deal of software is priced on the assumption that the capability it provides is scarce and valuable enough to charge a recurring fee for. That model works when the underlying capability is hard to replicate. But when the core capability of many software products is, at root, AI — and the cost of that AI is collapsing toward something close to free — the foundation of the pricing model starts to wobble. If the thing you are charging for is becoming a cheap commodity, your ability to charge a premium for it erodes.
Consider what this means for a product whose main value is some AI-powered feature. As the cost of that capability falls, two things happen. First, competitors can offer the same feature more cheaply, because the input they are building on has become cheaper for everyone. Second, customers increasingly realise they could approximate the capability themselves, because the underlying intelligence is now accessible and affordable. Both forces push prices down. The moat that the feature once provided fills in as the capability behind it commoditises.
I think this forces a hard rethink of where software value actually lives. If the AI capability itself is becoming cheap, then charging for the capability is a losing strategy. The value has to come from somewhere the commoditisation does not reach: proprietary data the model cannot get elsewhere, deep integration into a customer's workflow that is painful to replace, trust and reliability in high-stakes settings, or the accumulated context that makes a product genuinely better for a specific customer over time. These are the things that do not get cheaper just because intelligence does.
This is, in a way, a repeat of a familiar pattern with a new input. Whenever a key input becomes abundant and cheap, the businesses built on its scarcity struggle, and the value migrates to whatever remains scarce. Cheap intelligence is wonderful for the world and dangerous for any company whose entire proposition was access to intelligence that used to be expensive. The very abundance that benefits everyone undermines the businesses that monetised the shortage.
I do not think this means software is doomed — far from it. It means the basis of competition is shifting. The winners will be the ones who treat cheap AI as a foundation to build on rather than a product to resell, who use the falling cost of intelligence to create value that does not itself collapse in price. The losers will be the ones who assumed they could keep charging premium prices for a capability that is rapidly becoming a commodity available to all.
There is a timing problem that makes this especially treacherous. Business models change slowly, but the cost of AI is falling fast. That mismatch means some companies will keep pricing as though the old scarcity holds, right up until customers and competitors force the issue. The repricing, when it comes, may be sudden rather than gradual, and the companies that did not see it coming will be caught flat.
So the question I would put to anyone in the software business is uncomfortable but necessary: if the AI capability at the heart of your product becomes nearly free, what are you left selling — and is that thing scarce enough to build a business on, or are you quietly relying on a scarcity that is already disappearing?
The above reflects my personal views only and is intended for informational and discussion purposes. It does not represent the position of any employer or organisation.