Token Lockdown: Why AI Tools Are Getting More Expensive - And Worse

The landscape of generative AI has undergone a dramatic transformation since its early days, when platforms like ChatGPT and Claude offered near-unrestricted access to their capabilities. Users could experiment freely, pushing the boundaries of what these models could do without worrying about strict usage limits or prohibitive costs. This period of openness was crucial in driving adoption, fostering innovation, and demonstrating the potential of large language models to a global audience. However, as the technology has matured and its applications have expanded, AI providers have begun to implement increasingly aggressive monetisation strategies. These changes mark a fundamental shift in how users interact with generative AI—from an era of exploration to one of commercialisation, where every token processed carries a financial cost.

The recent decision by Anthropic to impose weekly rate limits on Claude Code users, including those on paid plans, serves as a clear indicator of this broader trend. No longer content with simply offering premium tiers, AI companies are now enforcing hard restrictions on usage, effectively turning tokens—the basic units of computation in AI systems—into a carefully controlled commodity. This move is not an isolated incident but rather part of a wider industry pivot toward stricter monetisation controls. What was once a playground for experimentation is rapidly becoming a pay-per-use service, where access is metered and unrestricted generation is a luxury reserved for those willing to pay premium rates.

This shift is driven by the harsh economic realities of sustaining large language models. The computational resources required to train and deploy these systems are staggering, with operational costs running into millions of dollars per month for leading providers. As user demand continues to grow, companies are under increasing pressure to recoup these expenses, leading to a wave of pricing adjustments, usage caps, and tiered access models. The implications are far-reaching: developers, businesses, and casual users alike must now navigate an ecosystem where AI-generated content is no longer limitless but instead treated as a finite—and billable—resource.

The Early Days of Generative AI: Open Access and Experimentation

When OpenAI first launched ChatGPT in late 2022, it represented a seismic shift in how the public interacted with artificial intelligence. The model was made freely available, allowing anyone to engage in extended conversations, generate creative content, and even write functional code without upfront costs. Similarly, Anthropic’s Claude and other competitors followed suit, offering generous free tiers to attract users and showcase their capabilities. This period was characterised by a sense of boundless possibility—developers built integrations, businesses explored automation opportunities, and individual users experimented with everything from essay writing to poetry generation.

The open-access model was not just a marketing strategy; it was a necessity. Generative AI was still in its infancy, and providers needed real-world usage data to refine their models, identify edge cases, and improve performance. By allowing unrestricted access, companies could gather vast amounts of feedback while simultaneously building brand loyalty. However, this approach came at a steep cost. Every query processed by these models required significant computational power, and as adoption skyrocketed, so did infrastructure expenses. Reports estimated that running a single ChatGPT conversation could cost OpenAI several cents in compute resources—a figure that quickly added up given the platform’s millions of daily users.

Despite these costs, providers initially absorbed the financial burden, betting that widespread adoption would eventually lead to profitable monetisation strategies. Free tiers acted as loss leaders, enticing users to explore the technology before eventually converting them into paying customers. Yet, as the novelty of generative AI wore off and operational expenses continued to climb, it became clear that the industry could not sustain an unlimited, cost-free model indefinitely. The shift toward monetisation was inevitable, and it began with the introduction of premium subscription plans and pay-per-use APIs.

The Monetisation Shift: From Free Tiers to Token-Based Pricing

The first major sign of change came with the launch of ChatGPT Plus, OpenAI’s subscription service offering priority access and enhanced features for a monthly fee. This was quickly followed by other providers introducing similar paid tiers, effectively creating a two-tiered system where free users faced slower response times and reduced functionality. However, subscription models alone were not enough to offset the soaring costs of AI infrastructure. Providers needed a more granular way to monetise usage, and the solution came in the form of token-based pricing.

Tokens—the chunks of text that AI models process—became the new currency of access. OpenAI’s API, for example, transitioned to a per-token billing structure, where developers paid based on the volume of text they generated or analysed. This approach allowed companies to directly correlate usage with revenue, ensuring that heavy users contributed proportionally more to operational costs. At first, these changes primarily affected enterprise clients and developers using API access, but the trend soon expanded to consumer-facing products as well.

Anthropic’s recent enforcement of weekly rate limits for Claude Code users, including those on paid plans, underscores how deeply this monetisation strategy has taken root. Unlike earlier restrictions, which primarily targeted free-tier users, these new limits apply even to paying customers, signalling that no segment of the market is exempt from usage controls. The message is clear: generative AI is no longer an open-ended resource but a metered service where access is carefully regulated to maximise profitability.

This shift has profound implications for how businesses and individuals use AI. Content mills and spam operations, which once relied on unlimited cheap generations, now face hard ceilings on output. Developers building AI-powered applications must factor token costs into their budgets, much like they would with cloud computing expenses. Even casual users encounter friction, as free tiers become increasingly restrictive, pushing them toward paid subscriptions. The era of treating AI as a boundless utility is over; in its place is a system where every token carries a price, and providers are becoming ever more sophisticated in how they meter and monetise consumption.

The Economics Behind AI Restrictions: Why Providers Are Tightening Controls

The driving force behind these monetisation efforts is simple: generative AI is astronomically expensive to operate. Training state-of-the-art models like GPT-4 or Claude 3 requires thousands of high-end GPUs running for months, with costs often reaching hundreds of millions of dollars. Even after training, inference—the process of generating responses to user queries—consumes substantial computational resources. Industry estimates suggest that answering a single complex prompt on a large language model can cost providers anywhere from a few cents to over a dollar in electricity and hardware depreciation.

These costs are compounded by the fact that generative AI has moved beyond niche experimentation into mainstream business applications. Companies now use LLMs for everything from customer support automation to legal document analysis, dramatically increasing the volume of daily queries. Without strict monetisation controls, providers would face runaway expenses that could jeopardise their financial sustainability. This reality has forced AI firms to adopt strategies similar to those seen in cloud computing, where usage is tightly monitored and billed accordingly.

However, there is also a strategic dimension to these restrictions. By imposing rate limits and paywalls, providers can segment their user base more effectively, pushing high-volume clients toward expensive enterprise plans while still retaining casual users on free or low-cost tiers. This tiered approach maximises revenue extraction from commercial users—who derive significant value from AI tools—while maintaining a foothold in the broader consumer market. Additionally, limiting access helps prevent abuse, such as automated spam generation or API exploitation, which can degrade service quality for legitimate users.

Yet, the most telling aspect of this shift is that even paying customers are no longer immune to restrictions. In the past, premium subscriptions guaranteed unfettered access, but now, providers like Anthropic are applying hard caps regardless of payment status. This suggests that the industry is moving toward a model where usage is commoditised, and even the highest-tier users must pay extra for additional capacity. The implications are clear: generative AI is transitioning from a disruptive, open-access technology into a tightly controlled service where profitability takes precedence over unlimited availability.

The Broader Implications: Who Wins and Who Loses in the Pay-Per-Token Era?

The move toward stricter monetisation has far-reaching consequences, creating clear winners and losers in the evolving AI ecosystem. On one side, AI providers and cloud infrastructure companies stand to benefit enormously. Firms like OpenAI, Anthropic, and Google DeepMind now have a scalable revenue model that aligns directly with usage, ensuring that increased demand translates into higher profits. Cloud providers such as Microsoft Azure and AWS also gain, as businesses requiring heavy AI capabilities must invest in expensive compute instances to run models at scale.

For enterprise clients with deep pockets, these changes are manageable—if inconvenient. Large corporations can absorb the costs of API usage and premium subscriptions, treating AI as just another operational expense. Many are even willing to pay a premium for guaranteed access, custom model fine-tuning, and priority support. In this sense, the monetisation shift reinforces the divide between well-funded organisations and smaller players, further entrenching AI as a tool for those who can afford it.

However, the picture is far bleaker for independent developers, startups, and individual users. Those who once relied on free or low-cost access to build innovative applications now face significant barriers. Rate limits constrain experimentation, while per-token pricing makes scaling prohibitively expensive for bootstrapped projects. The democratising promise of AI—where anyone could leverage cutting-edge technology—is giving way to a reality where access is dictated by financial resources.

Perhaps the most concerning trend is the normalisation of paywalls around core AI functionalities. Features like code generation, long-form content creation, and advanced reasoning are increasingly locked behind premium tiers, leaving free users with severely limited capabilities. This risks creating a two-tiered digital landscape where those who can pay enjoy the full benefits of AI, while others are left with watered-down alternatives. Over time, this could stifle innovation, as the next groundbreaking AI application may never be built simply because its creator couldn’t afford the necessary API calls.

Is There an Alternative? The Role of Open-Source AI

In response to the tightening grip of commercial AI providers, many have turned to open-source models as a potential counterbalance. Projects like Meta’s Llama 3, Mistral’s Mixtral, and Falcon 180B offer weights that can be freely downloaded and run locally, theoretically bypassing restrictive API pricing. Advocates argue that open-source AI could democratise access, allowing individuals and small businesses to deploy powerful models without being subject to corporate monetisation strategies.

However, the reality is more complicated. While open-source models have made significant strides in performance, they still lag behind proprietary offerings in terms of accuracy, coherence, and versatility. Training and fine-tuning these models require substantial computational resources—something that remains out of reach for most independent developers. Additionally, self-hosting an LLM at scale introduces its own costs, including GPU leasing, electricity, and maintenance. For many, the convenience of paid APIs still outweighs the logistical challenges of running open-source alternatives.

That said, the open-source movement is gaining momentum, particularly among privacy-conscious users and organisations wary of vendor lock-in. Some companies are already experimenting with hybrid approaches, using proprietary models for high-stakes tasks while relying on open-source alternatives for less critical functions. If this trend continues, it could pressure commercial providers to offer more flexible pricing, lest they lose customers to freely available alternatives.

The Inevitable Commercialisation of Generative AI

The imposition of rate limits on Claude Code users is not an anomaly—it is a harbinger of the industry’s future. Generative AI, once hailed as a revolutionary force that would democratise creativity and knowledge, is rapidly becoming another commoditised digital service, governed by the same profit-driven logic as cloud computing and SaaS platforms. Tokens, once a mere technical metric, are now the currency of access, and providers are refining their ability to meter and monetise every interaction.

For users, this means adjusting to a new reality where AI is no longer an infinite resource but a paid utility. Developers must budget for token costs, businesses must weigh ROI on AI investments, and casual users must accept that the most powerful features come at a price. The days of unfettered experimentation are fading, replaced by an era where AI’s potential is unlocked not by curiosity alone, but by financial capacity.

Whether this commercialisation stifles innovation or simply reflects the maturation of the industry remains to be seen. What is certain, however, is that the age of free and unlimited generative AI is over—and the pay-per-token future is here to stay.

United Kingdom | AI, Technology, ML | |