At first, a new ride-hailing app on your smartphone takes you anywhere in the city virtually for free, while generously showering its drivers with bonuses. Then, your favorite messenger becomes an indispensable part of your life, boasting a completely ad-free, clean, and seamless interface. Years later, however, that same ride-hailing service becomes tenfold more expensive, drivers’ cuts are slashed, and the messenger begins flashes paid subscriptions and banners at every turn.
Today, the global tech market is witnessing the largest and most expensive rerun of this exact scenario. Only this time, it is not about ordinary apps, but rather artificial intelligence technologies that are said to reshape the future of humanity. We might currently be living through the final days of the “free and cheap AI” era, which is now drawing to a close.
The three-stage degradation
Renowned author and techno-critic Cory Doctorow coined the term “enshittification” to explain the evolutionary decay of digital platforms. The term hit the bullseye so precisely that the American Dialect Society named it the 2023 Word of the Year, the Australian Macquarie Dictionary followed suit in 2024, and by 2025, Doctorow’s book of the same name made it to the Financial Times Business Book of the Year longlist.
This three-stage model accurately maps the transformation of any major tech platform:
- Stage 1: Platforms offer their services at a loss—either dirt cheap or completely free—to aggressively aggregate a massive user base and build behavioral habits.
- Stage 2: Once the audience is locked in, the platform hooks third-party business entities—advertisers, content creators, and merchants. The user experience is no longer the product; it becomes the raw material.
- Stage 3: Having bound both sides to the ecosystem, the platform shifts its entire focus toward squeezing maximum revenue from both users and businesses to deliver value to shareholders.
Doctorow’s crucial caveat: this is not an iron law of capitalism. It is the result of impunity. The four forces that historically kept platforms disciplined—competition, government regulation, interoperability, and employee resistance—have weakened one by one over the past decades. Facebook, Amazon, and Google decayed in this exact vacuum. The ultimate question now is: where do today’s AI giants—OpenAI, Microsoft, Google, and Anthropic—stand within this model?
A “Honeymoon” built on debt
For the past few years, we have enjoyed access to the world’s most advanced neural networks almost entirely for free or via low-cost subscriptions. This era has survived not on the actual revenues of AI companies, but on billions of dollars in subsidies from venture capital and tech conglomerates.
The numbers present a stark reality. Market leader OpenAI generated $13.1 billion in revenue in 2025—a 250% increase year-over-year. Yet, in that same year, the company’s operational losses ballooned to $20.9 billion, with a net loss reaching $38.5 billion. Currently, 1.1 billion people use ChatGPT at least once a month, but only about 5% of them pay for a subscription. Meanwhile, the company has locked in nearly $1.4 trillion in infrastructure commitments for data centers through the early 2030s. Behind every generated line of text, image, or code lies massive server overhead, water-cooled data centers, and astronomical electricity bills.
Tech critic Ed Zitron flagged this phenomenon as early as 2024, dubbing it the “subprime AI crisis”—a nod to the 2007 mortgage crash. His thesis posits that thousands of companies have integrated generative AI into their products at artificially suppressed, sub-cost prices. According to Zitron’s estimates, if OpenAI and Anthropic were to price their services near actual cost, tariffs would need to scale tenfold to a hundredfold. (In fairness, some of Zitron’s figures have been flagged as overstated in independent reviews, but his core thesis—expenditures completely disproportionate to revenues—remains uncontested.)
In other words: the current cheap tokens, generous free tiers, and $20 “all-inclusive” subscriptions are not market prices. They are the price being paid to get you hooked. And someone, sooner or later, will have to foot the bill.
The cycle has already begun to turn
This is no longer a theoretical anxiety. Throughout 2025 and 2026, symptoms of Stage 2 emerged in rapid succession—backed not by speculation, but by hard dates and metrics.
- In May 2024, OpenAI CEO Sam Altman called the combination of AI and advertising “inherently dystopian” and a “last resort.” By April 2025, product recommendations surfaced in ChatGPT Search. In November, an ad infrastructure was discovered in the Android beta app—which the company officially denied. A month later, ads were formally announced. On February 9, 2026, advertising officially launched on ChatGPT in the US—appearing on the Free and the new $8 “Go” tiers as “sponsored” blocks beneath responses. By May, the pilot expanded to the UK, Japan, Brazil, Mexico, and South Korea.
- With the rollout of GPT-5, OpenAI forced free users onto cheaper, downgraded models. They stripped the ability to manually select models from the $20–$35 tiers, reserving that choice exclusively for the $200 Pro tier. A tier that previously offered six models now offers two. The company packaged this as a feature, claiming a “real-time router selects the best model for you.” The AI-powered code editor Cursor also introduced sharp price hikes for its users—despite raising nearly a billion dollars in investment. The reason is simple: the underlying models became more expensive, and that cost was pushed downstream to the user.
- The most vivid example is Microsoft. The company bundled its Copilot AI assistant into Microsoft 365 subscriptions by default, driving the Personal tier price up from $70 to $100 per year—a 42% spike at the lowest tier. In return, users receive a mere 60 “AI credits” per month, which do not roll over. Crucially, this aggressive push backfired: three years in, fewer than 4.5% of Microsoft’s enterprise users actually pay for Copilot. The friction grew so intense that in March 2026, Microsoft was forced to pull Copilot from core apps like Photos, Notepad, and Widgets—marking the first major official retreat from its “AI everywhere” strategy.
Search: the damage is already quantified
The thesis that “cramming AI into a product does not inherently improve it” finds its clearest proof in Google Search. A Pew Research Center study analyzing the real browser activity of 900 Americans revealed that on pages featuring an AI Overview—the AI-generated summary at the top of search results—only 8% of users click through to an actual website. On pages without an AI summary, that click-through rate stands at 15%.
The first randomized controlled trial in 2026—involving 1,065 users—delivered even more definitive metrics: AI Overviews slash click-through rates to websites by 39.8% and drive “zero-click” searches up by 34.5%. Most telling of all, there was no measurable improvement in how users rated their search experience. In short: media traffic was cannibalized, while the end-user gained nothing. The consequences are concrete: Business Insider cut its workforce by a fifth after its organic search traffic collapsed by 55%, and HuffPost lost half of its search traffic. For the media industry, this is not an abstract trend—it is an existential crisis.
Companies are selling this transition under the guise of enhancing “user experience.” However, sentiment data tells a different story. An NBC national poll clocked AI’s net favorability at minus 20 points. Among the 18–34 demographic, it plummeted to minus 44.
Yet, a critical nuance exists: people are not anti-AI. They are anti-forced, anti-useless AI. Amazon integrated AI-generated review summaries quietly at the top of pages without loudly screaming “AI”—today, it stands as one of the site’s most heavily utilized features. Apple took a similar route with Apple Intelligence, deploying it contextually and entirely at the user’s discretion. The same technology, two distinct approaches, yielding two completely different outcomes.
Market metrics validate this shift. According to a June 2026 report by Sensor Tower, ChatGPT’s global market share dropped from 65% in December 2024 to 46% by May 2026—falling below the halfway mark for the first time in its history. The lost ground was captured by two rivals: Google Gemini grew its share to 28%, while Anthropic’s Claude expanded its monthly user base from 60 million to 245 million in just five months—a staggering 452% annualized growth rate making it the fastest-growing platform. Users are voting against poor experiences, not with their words, but with their feet.
The end game: four scenarios
The escalating cost of AI infrastructure and the pressure to monetize will inevitably push the market down one—or several—of four paths:
- The Chatbot as the New Google: Advertising merges with agentic commerce. OpenAI is already building direct checkout pipelines through ChatGPT with giants like Walmart and Etsy. For advertisers, this is the holy grail: tracking not just a click, but the entire transaction.
- The Bubble Bursts, the Tech Remains: The AI economy is a fragile chain: profitable Nvidia → debt-fueled data centers → loss-making model builders → loss-making startups. If a single link snaps, the structure buckles. However, a crash does not mean total erasure. Deflated GPUs and sophisticated open-source models will remain. The bubble dies; the technology endures.
- Deep lock-in via memory: AI memory—the personal context, preferences, and workflows uploaded to an assistant over years—could become the most potent lock-in mechanism in tech history. The current “benevolent era” persists only because this lock-in infrastructure is still being built.
- Ecosystem-Wide Degradation: AI consumes internet content, regurgitates it, and re-consumes its own output—a cycle where quality degrades at every turn. Clinically termed “model collapse,” models trained on synthetic data lose semantic diversity, returning a distorted, hyper-narrowed version of reality.
In light of this systemic shift in the tech ecosystem, businesses and users must recalibrate their long-term strategies. First, mitigate strategic dependency. Avoid building business pipelines or core daily operations solely around the current pricing of a single proprietary AI provider; these rates are highly subsidized and can pivot sharply without warning. Second, seriously evaluate open-source alternatives. Deploying models like Llama or Mistral on private infrastructure hedges against vendor tariff shocks in the long run. Third, audit AI efficiency strictly. Calculate whether the convenience of proprietary tools genuinely justifies their impending full-market price tag.
Doctorow’s most vital insight offers a silver lining: enshittification is not a law of physics; it is a symptom of unchecked monopoly. The forces capable of halting it are still very much alive in the AI market—competition is fierce, open-source models are maturing, and user backlash is driving product pivots. Yet, the historical formula remains undefeated: first, the product works for you. Then, if left unchecked, you start working for the product. The venture-backed party is winding down. Ahead lies a more pragmatic, sober era of artificial intelligence—one governed by the strict laws of economic gravity.
