The open data era fed the AI era. The AI era is running the same loop. The circle is not accidental.
Let’s be real over what has been happening here.
The entire tech industry is burning through capital at a rate we have never seen before. On March 25th, OpenAI announced on Twitter they are closing shop to their Sora project, just 6 months after the announcement of Sora 2, burning about USD$1 million a day. Anthropic has raised billions. Almost every other week, we hear big techs like Google and Microsoft spending at large scales on AI and data centres that dwarf anything the previous tech era required, absorbing losses on AI products in exchange for market position. Yet, most of these endeavours remain unprofitable.
Every prompt cost these companies money. In most cases, they’re charging you nothing. The free tiers, the generous API limits, the accessible pricing. None of it reflects the actual cost of what’s being delivered.
This is not a secret. It is the stated strategy.
Growth over profits. Build first, monetise later.
Sound familiar?
The ZIRP Playbook, Running Again
In last week’s article, I wrote about how the open API era was funded by zero interest rate policy. The cheap capital that made it rational to burn money on growth rather than profit. That era ended in 2022 when rates began rising and investors demanded returns.
What I didn’t fully articulate then: AI arrived at exactly the moment that the previous version of this playbook was collapsing. And it picked up the same handbook.
The logic is structurally identical. The motivation isn’t. But the behaviour is. Subsidise access now. Build network effects and switching costs. Establish the infrastructure as indispensable. Then, once the dependency is deep enough, adjust the pricing to reflect the actual value being extracted.
The difference is scale. The previous era’s losses were in the hundreds of millions. The current era’s losses are in the tens of billions. And the infrastructure being built isn’t just software — it’s physical.
The Costs Nobody Is Putting on the Invoice
Here is where the parallel breaks down.
When a social platform burned investor money on free APIs in 2012, the consequence of failure was financial. Investors lost money. But this came with the clear understanding of acquiring users and developers. This provides a win-win situation for everybody. Platforms experience user growth and brand recognition. Users get excellent user experience and community to belong to. And developers have fun just building and experimenting.
AI’s cost base just feels… different…
On the surface, it doesn’t feel like it would benefit anyone at all.
Data centres running large language models consume electricity at a scale that is measurably reshaping energy demand in entire regions. Water consumption for cooling is significant and, in some cases, competing directly with local supply. The carbon footprint of training a single large model is comparable to the lifetime emissions of multiple cars. These are not estimates from critics. They are documented figures from the companies themselves, where disclosed at all.
Then, there is the potential risk of massive unemployment like we have never seen before. The previous tech era displaced some categories of work while creating others. The optimistic argument was always that automation creates new jobs even as it eliminates old ones. AI is moving faster and broader than previous automation waves. It touches on almost everything: writing, design, legal research, customer service, coding, finance operations. The new jobs argument is getting harder to make with a straight face when the displacement is happening across knowledge work simultaneously. And the social anxiety around job security and economic stability is compounding in real time.
None of these costs appear on an AI company’s profit and loss statement. They are externalised — paid by energy infrastructure, local water supplies, workers in affected industries, and eventually governments managing the social consequences. The business model looks more attractive than it is because the full invoice isn’t being presented.
What Enshittification Looks Like for AI
If the pattern from my previous article holds, what follows is predictable.
Phase one is already ending. The fully free tiers are degrading or disappearing. OpenAI has already begun testing advertising on their ChatGPT’s Free and Go tiers. The trajectory is clear.
Phase two is monetisation. This is where the pricing starts to reflect something closer to actual cost recovery, and the features that drove adoption get gated behind paid tiers. Developers who built products on cheap API access find their margins compressed or their products unviable. Some will pay. Some will not.
Phase three is extraction. This is where it gets structurally uncomfortable. The data that users and developers generate while using AI products — the prompts, the workflows, the outputs, the preferences — is itself training data for the next generation of models. The user becomes part of the product. The tool they paid to use is simultaneously using them.
We are somewhere between phase one and phase two right now. Phase three has a business logic that is hard to argue against from the inside of a company that needs to survive.
The Full Circle
Let me draw this out plainly.
The open data era was funded by cheap capital. Platforms built communities. Communities generated content. Content became the asset. When the capital got expensive, the asset got locked. AI trained on that locked content.
But the driver this time isn’t cheap capital — it’s fear. Companies are burning billions not because money is free but because losing feels existential. That is a different motivation entirely. And in some ways, a more reckless one.
Along the way, something else got lost. The community. This is more about who can build the most powerful AI first, and in that race, the average user has stopped being the point. The voices telling these companies to slow down are getting louder. They’re not being heard.
Yet the voice going the other way is just as loud. Non-technical people are picking up these tools and talking about building the next innovation SaaS. With ideas that sound different but still feel rather familiar — the newest modern CRM, the next task tracker platform. For the first time, from my own experience, I am seeing more builders than users. And almost none of them stopping to ask what these companies are actually building toward.
What This Means If You’re Building Right Now
Don’t architect your product around AI pricing that exists today. The generous API limits, the accessible tiers, the free tooling. This is the loss-leader phase. Price increases are coming. Access restrictions are coming. The companies running these products at a loss have investors, and those investors will eventually require a return.
Because once somebody gains that monopoly, someone else will have to pay the price. More often than not, it’s the consumer.
Build with AI. It is genuinely useful. But build with one eye on what your product looks like if the cost of your core dependency doubles or triples in eighteen months. Because that scenario is not pessimism. It is the pattern, running again.
And this time, the bill doesn’t just hit your balance sheet.