There's a quiet assumption running through many businesses about AI right now: it's brilliant and cheap today, so why would that change? Cheaper, faster, better, on repeat, forever. AI has become a comfort blanket. But that assumption is probably wrong, or at least far more complex than the headlines suggest once you start tugging at the loose threads.
Here's the bit that doesn't get said out loud very often: the AI prices we're all enjoying right now are, in large part, subsidised. Companies like OpenAI, Anthropic and Google are investing heavily to win market share and lock people into their ecosystems, which makes the AI tools feel astonishingly cheap for what they do. Underneath that price tag, though, the real cost of running large AI models - the compute, the infrastructure, the sheer scale of it all - is significant. Very significant. What we're seeing isn't the natural price of AI. It's just the introductory offer.
Every major AI provider wants to become the default platform businesses build around. Whether you're using ChatGPT, Claude, Gemini or other large language models (LLMs), the same commercial incentives apply. Once your workflows, prompts and integrations live inside one ecosystem, switching becomes much harder. That's what makes today's pricing so compelling: it's not just about attracting customers, but about making them stick around.
Right now, artificial intelligence is fantastic at speeding human engineers up. It writes code fast, spots patterns, and helps automate the "boring" bits. What it still isn't great at is being left entirely alone to produce production-ready software from the get-go, with no human required. Someone still has to check the edge cases, think about architecture, worry about security, and generally make sure the thing can withstand some prodding once real users touch it.
That's simply where the technology is today. AI isn't replacing engineers, it's making good ones considerably more productive. It’s not a replacement hire. It's the world's fastest (and occasionally overconfident) junior developer.
Meanwhile, the cost of running the newer, smarter AI models is going up, not down. More capable reasoning takes more compute, and more compute costs more money. At the same time, providers are still discounting heavily to grab their share of users. That combination, rising costs plus subsidised pricing, isn't sustainable indefinitely. Not even for the tech giants. We've already watched this play out elsewhere in tech: services launch cheap and brilliant, win everyone over, then quietly (or not so quietly) double or triple their prices once the market's theirs to lose. AI pricing is a business decision, not a law of physics. It can and will move.
That doesn't necessarily mean AI will become expensive overnight. But it does mean today's pricing shouldn't be treated as the baseline forever.
This is where it gets interesting. If AI pricing doubles or triples, will it still be the cheapest way to build software? The honest answer: it depends what you're making.
For quick scripts, throwaway prototypes, small internal tools, AI is probably still the cheapest option even at a higher price point. But for anything that actually needs to work reliably in production, the maths changes fast. The cost of a mistake, a security hole, a scaling failure or a badly designed database can dwarf whatever you saved on the build. And asking AI to fix a mistake it introduced? Well, you can probably imagine where that loop ends. Lean entirely on AI for high-stakes work, and it starts to look less like a bargain and more like a gamble.
The real answer isn't AI versus engineers. It's AI and engineers. AI brings the speed; engineers bring the judgement: deciding what to build, how it should behave, and how it needs to hold up under pressure. AI-assisted engineering, not full automation, is where the genuine cost efficiency lives.
This also shifts what software development is really about. It's moving away from "who can type code the fastest" and towards "who can make the right calls." Engineers who know how to direct AI, rather than simply prompt it and hope for the best, are becoming more valuable, not less. They're the ones getting the most output for the least cost, which is exactly the sort of thing that matters when budgets get tighter.
If prices do rise sharply, and suddenly, the sensible response isn't to panic and abandon AI. It's to get smarter about when to use it. Businesses will become more selective, reserving AI for where it genuinely adds value and relying on human oversight where the stakes, and the cost of getting it wrong, are highest. That's not AI losing its place. That's AI finding its proper one, alongside a human engineer architecting its presence. AI doesn't, and won't, fully replace human work. It just changes how that work gets distributed, and who's still needed to make the calls.
In the meantime, it's worth taking stock: how is AI being used across your team? Is it being leaned on in the right ways? And does the person using it have the skills to know when to step in? If prices do rise sharply, companies will naturally need to optimise usage, relying more on human oversight to reduce unnecessary AI calls rather than eliminating either altogether.
AI isn't going anywhere. Today's pricing, however, might. The businesses that get the best long-term value won't be the ones that rely on AI the most; they'll be the ones that know when to rely on it, and when not to.
Let's see what we can make together.
From startups to scaleups and enterprise, we're always happy to talk to impact-focussed and ambitious organisations. Use our quote creator to get a better idea of scope, timescales and next steps.
