Generative AI (genAI) platforms lure you in with free plans, then offer new features for a monthly fee – but as many users are discovering, flat-rate pricing is becoming a thing of the past as rapidly changing market dynamics push AI into a new economic model based entirely around tokens.
The computational load of everything a genAI system does – whether parsing your prompts, doing back-end processing or generating responses – is measured in computational tokens, and the more complex a task the more tokens it requires, with users’ consumption rising as their skills grow.
Tech giants were happy to wear the cost of these tokens while growing user numbers, reflected in their multi-billion dollar investments in AI data centres.
But with ever more complex genAI capabilities increasing token usage accordingly, the economics were becoming a problem.
Costs will rise even faster with the mainstreaming of AI agents which run multiple operations and complex searches to collate data, process information or complete transactions on your behalf.
The problem is they are swamping the Internet and need up to 24 times as many tokens as normal queries.
Similarly, as genAI is integrated into ever more tools – producing massive jumps in token usage, as is likely after Apple this month gives billions of iPhones, Macs and iPads access to a significantly more powerful genAI based Siri – overall consumption will surge even more.
So, why is this a problem?
Any given genAI system can only process so many tokens at once, and slow performance is the result of token consumption exceeding token availability.
It’s a surprisingly common situation that is pushing tech giants to build massive AI data centres as quickly as possible.
New data centre capacity has driven down the incremental value of genAI tokens, but that capacity is expensive.
While tech giants (and their cash-rich investors) have been willing to wear that cost to boost uptake, providing unlimited tokens for a flat rate has become untenable.
That has seen genAI platforms increasingly restricting usage, whether by limiting the number of tokens a user can consume per day, selling more expensive monthly plans with higher token limits, or limiting how quickly a given user can consume tokens.
Will this make my AI usage cheaper or more expensive?
That depends on what you’re doing with genAI.
Casual or disciplined users may have no problem staying within the confines of a free plan or monthly plans offering a set amount of tokens, but for users that have integrated genAI into their everyday work, the story is very different.
Recent months have seen dramatic price changes: in April, for example, OpenAI and Anthropic began charging business users for compute resources, while on 1 June, Microsoft changed to a usage based pricing model for its GitHub AI Copilot platform.
Users have been rebelling, with one GitHub user complaining on Reddit that they were going to cancel after bills rose from $40 (US$25) per month to $1,050 (US$750), while another called the new model “ridiculous” after bills soared from $70 (US$50) per month to $4,200 (US$3,000).
Just how the new plans impact any given user’s costs will vary by usage patterns, but it’s clear that for genAI power users the general trend is towards more expensive services.
This will likely force them to evaluate genAI in terms of how much capability it delivers per spend on tokens.
What does it mean for businesses adopting AI?
Analysts have flagged genAI platforms’ transition to a token-based economy as a key management issue, since software development projects using genAI tools will need to be allocated appropriate budgets to cover the cost of the tokens they use.
Development managers must also analyse token usage to identify which developers are working inefficiently, which risks a chilling effect on naturally inefficient vibe coders who design and iterate apps more for functionality than cost efficiency.
Such emerging ‘tokenomics’ “reframe something fundamental about how leaders make decisions,” Microsoft said after a recent survey surfaced tokenomics as one of five key changes that genAI is driving within businesses where ’tokenmaxxing’ – unchecked AI usage – has become too common.
The rise of tokenomics has direct implications for headcount: as genAI increasingly does things humans would otherwise have done, managers will have to weigh the cost of buying the tokens necessary for that task, against employing a human with all their associated costs.
“The allocation question that follows,” Microsoft notes, “is immediate and concrete: who gets tokens, how many, and for what work? And the cost piece is moving fast as AI models and systems become better, faster, and more efficient: what it costs today won’t be what it costs next year, or even next quarter.”
Who will win and lose from this trend?
Tokens are the currency of genAI, and the immediate winners from the shift will be AI platforms, which have finally found a way to stem what had become massive financial losses as the tech giants spent billions on new genAI power that was being sold to monthly users for a relative pittance.
Yet even OpenAI CEO Sam Altman admits the costs are becoming “a huge issue”, and with the number of tokens consumed now directly tied to the economic value they create, the new approach may cost users more in the short term.
However, it will also help users benchmark the efficiency of their developers and the software tools, like AI agents, that they deploy.
Tokenomics is delivering other insights as well: in March, for example, observers noted that China was using 140 trillion AI tokens per day, a world leading pace that reflects that country’s breakneck AI adoption and serves as a proxy for national AI capability in the global AI arms race.
Given AI’s unstoppable momentum, it’s clear the shift to token-based pricing will boost tech giants’ revenues in the short term.
But as users become accustomed to economising, they may reap additional benefits as AI evolves from being a novelty into a productivity investment.
