One of the more revealing stories in AI right now is not about a new model release. It is about pricing.
Research firm SemiAnalysis recently published estimates suggesting that some of the heaviest users on premium AI subscriptions may be consuming far more compute than they pay for. In its test, the firm reportedly pushed usage limits across paid tiers from Anthropic and OpenAI, then compared subscription revenue with estimated infrastructure cost. The headline claim was striking: a power user on Anthropic's top Claude Max tier could potentially generate thousands of dollars in monthly compute cost, while a heavy ChatGPT Pro user could cost OpenAI even more.
Whether every estimate is perfect is almost beside the point. The broader signal is clear. Leading AI labs appear willing to subsidise extreme usage in order to win attention, loyalty, and workflow share from serious users. In other words, they are not just selling access to a chatbot. They are buying a position inside the daily operating habits of developers, analysts, founders, consultants, and knowledge workers.
For businesses, that matters far more than the internet spectacle of "this person used $14,000 of compute for a $200 subscription". The real question is what these pricing strategies tell us about the AI market, and how businesses should respond before they become dependent on tools whose economics may not stay this generous forever.
As of June 2026, Anthropic publicly lists Claude Max plans from US$100 per month, with higher-capacity options above that, while OpenAI's ChatGPT Pro offering includes premium paid tiers above the standard consumer plans. That tells us something immediately: these companies know there is a class of user whose value is not measured only by monthly subscription revenue.
The obvious interpretation is that AI labs are competing for market share. That is true, but it is not the whole story. They are competing for workflow dependency. The user who builds their daily work around one lab's interface, model behaviour, memory, coding flow, and surrounding ecosystem is much harder to lose later. Once a team writes prompts a certain way, builds internal processes around one product, and starts trusting one output style, switching becomes more expensive than the monthly subscription fee suggests.
This is why heavy-user subsidies can make strategic sense even when they look irrational at first glance. The lab may lose money on the subscription in the short term, but gain several longer-term advantages:
Put simply, the subscription itself may not be the final product. It may be the customer acquisition layer for a much larger revenue opportunity.
Traditional SaaS businesses usually like predictable margins. Once the software is built, serving one more customer is relatively cheap. AI is different because usage can drive substantial variable cost. Every long reasoning session, code generation request, image task, or large-context interaction consumes real infrastructure resources.
That means AI subscription pricing is inherently harder to stabilise than pricing for a CRM, CMS, or accounting platform. The vendor is constantly balancing three pressures:
This creates a pricing environment that behaves more like a subsidised transport network or a venture-backed delivery platform than a mature software category. Vendors are still learning what "normal" usage looks like, what users will tolerate, and what margin structure can survive once the growth phase cools down.
That is why businesses should avoid reading today's AI subscription prices as permanent reality. They are better understood as a temporary market-clearing mechanism in an aggressive land grab.
When AI labs subsidise power users, they are effectively buying distribution into professional work. That distribution matters because AI adoption is sticky in subtle ways.
A team that starts using one assistant for coding may later use the same vendor for document drafting, research, support automation, internal knowledge search, and workflow orchestration. A founder who gets hooked on one model's style might later approve an enterprise deal. A consultant who relies on one AI tool for proposal drafting may gradually build client delivery processes around it.
In that sense, consumer and prosumer subscriptions are functioning as beachheads. The lab is not only monetising an individual user. It is trying to establish itself as the default thinking layer across an organisation.
That helps explain why vendors can tolerate apparent losses on the heaviest users. The power user is often not just a subscriber. They are a distribution node.
For Australian businesses, the main risk is not that OpenAI or Anthropic will suddenly disappear. The risk is that many firms will build critical internal habits around AI tools without fully thinking through pricing volatility, data governance, vendor concentration, or replacement cost.
That risk is easy to miss because the current experience feels deceptively affordable. A few hundred dollars per month can give an individual access to capabilities that would have seemed absurdly cheap even two years ago. But if those prices are partly subsidised, then businesses should assume at least some of the following will happen over time:
That does not mean businesses should avoid AI subscriptions. It means they should treat today's pricing as strategically useful, but not guaranteed.
Australian firms also have an additional layer to consider: many are adopting AI without large internal technical teams. That makes them more vulnerable to tool lock-in, because the path of least resistance is often to build around whatever works first. If the economics later change, those businesses may find that their processes, prompts, documentation, and team habits are more vendor-specific than expected.
The best response is not alarmism. It is better procurement thinking.
1. Separate experimentation from dependency.
It makes sense to exploit generous AI pricing while it exists. Teams should experiment, learn, and build capability now. But they should also be clear on which workflows are optional and which are becoming business-critical. If your sales operations, internal reporting, development workflow, or support triage now depends on one vendor's subscription product, that deserves more deliberate planning than an employee quietly paying for a premium plan on a company card.
2. Design processes that can survive model switching.
Where possible, document prompts, review steps, quality checks, and workflow logic in ways that can be adapted across vendors. Businesses do not need to avoid model-specific optimisation entirely, but they should avoid invisible lock-in. If changing provider would break your process completely, you are more exposed than you think.
3. Budget for the post-subsidy world.
When AI proves valuable, assume the long-term cost may be higher than the current trial-like price. The winning internal business case should still hold if pricing becomes less generous, not only while it is temporarily cheap.
Subsidised power usage suggests we are still in the customer-acquisition phase of AI, not the mature-margin phase. That is important because it changes how businesses should interpret vendor behaviour.
In a mature market, software pricing usually reflects a reasonably stable view of value and cost. In the current AI market, pricing often reflects strategic urgency. Vendors want data, attention, distribution, and embedded use cases. They want to become the default interface before the economics fully settle.
That does not mean the products are unsustainable in a dramatic sense. It means the product packaging is still fluid. Features may move. Limits may change. Bundles may be restructured. Plans may split into narrower segments for developers, researchers, teams, and enterprise buyers. Businesses should expect that kind of motion rather than acting surprised when it arrives.
There is also a second-order effect here. As premium AI tools become part of mainstream work, pricing itself becomes a competitive weapon. The lab that can most effectively subsidise serious users may be able to shape the market before rivals catch up. In that environment, subscription pricing is not just a billing model. It is a strategic moat-building mechanism.
One of the biggest mistakes businesses make with new technology is mistaking current price for intrinsic value. If a premium AI subscription costs less than a traditional software licence or junior staff time, it is tempting to think the capability itself is inexpensive.
It is not. The capability is powerful, resource-intensive, and strategically contested. The low apparent price may simply reflect that vendors are currently choosing to absorb part of the cost for competitive reasons.
That distinction matters because it changes how businesses should plan adoption. If AI becomes embedded in core operations, then governance, vendor management, security review, training, and process design matter just as much as the monthly fee. Treating AI as a cheap add-on usually leads to messy implementation later.
None of this is an argument to sit on the sidelines. If anything, subsidised access creates a useful window for businesses to learn quickly at relatively low cost. Companies can test where AI genuinely improves throughput, where it creates risk, and where it changes the structure of knowledge work.
The key is to use this phase intelligently. Learn aggressively, but architect carefully. Capture value, but do not assume today's plan structure is permanent. Build internal fluency, but avoid accidental dependence on one vendor's quirks where that dependence is unnecessary.
The businesses that do this well will get the upside of early adoption without being blindsided when the economics evolve.
If SemiAnalysis is directionally right, then the most important insight is not that a few power users are getting an extraordinary deal. It is that leading AI labs believe those users are worth subsidising because control of the workflow matters more than short-term margin.
That is a useful signal for any business trying to make sense of the AI market. We are not just watching a software category compete on features. We are watching a foundational platform race compete on habits, dependency, and future enterprise capture.
Businesses should respond accordingly: with curiosity, urgency, and a little more pricing scepticism than the current hype cycle encourages.
Ready to build an AI strategy that still makes sense when pricing changes? Get in touch with us to design practical AI workflows, governance, and vendor choices that hold up beyond the subsidy phase.
