The first phase of popular AI felt almost too generous: powerful tools, rapid innovation, and subscription prices low enough for almost anyone to experiment. That phase may not be ending completely, but it is starting to meet the reality of the meter.
GitHub’s announcement that Copilot is moving to usage-based billing is an important signal. It affects developers first, but the pattern will not stay confined to software development. The developer world is often a few months ahead of the broader business market. What happens there tends to ripple outward.
What GitHub Is Changing
GitHub has announced that, from 1 June 2026, Copilot plans will move away from the current “premium request” model and toward usage-based billing using GitHub AI Credits.
In practical terms, the base plan prices are not changing. Copilot Pro remains listed at $10 per month, Pro+ at $39 per month, Business at $19 per user per month, and Enterprise at $39 per user per month. Each plan includes a corresponding monthly allocation of AI Credits.
The important change is how usage is counted. Credits will be consumed based on token usage, including input, output, and cached tokens, according to the published API rates for each model. A token is a small unit of text that an AI model processes. In plain English, the more you ask the model to read, think through, and generate, the more tokens are consumed.
GitHub also says that code completions and Next Edit suggestions remain included in all plans and do not consume AI Credits. However, more advanced use cases, especially agentic work, will be charged against usage.
Agentic AI refers to AI systems that do more than answer a single question. They can take a goal, break it into steps, use tools, inspect files, make changes, and iterate over a longer task. That is powerful, but it is also computationally expensive.
The Real Issue Is Not GitHub Pricing
It would be easy to view this as a narrow pricing change. I think it is more than that.
The underlying issue is that AI compute is expensive, and current consumption patterns are not always well aligned with simple flat-rate subscriptions. A quick question to an assistant and a long autonomous coding session are not remotely the same from an infrastructure cost point of view.
This matters because the appetite for AI keeps increasing. Developers are experimenting with multiple agents, long-running tasks, automated code review, and complex workflows that may run for hours. In business, similar patterns are emerging around document analysis, workflow automation, research, reporting, and decision support.
Someone has to pay for the computing power behind all of this.
Venture capital can fund experimentation for a while. Large cloud providers can absorb losses for strategic reasons. But at some point, infrastructure-heavy services need revenue models that reflect actual demand. Usage-based pricing is one way to reduce the risk of sudden demand spikes and heavy users being subsidised indefinitely by light users.
Disappointed, But Also Slightly Relieved
I am disappointed because the good times of very cheap AI access appear to be changing, at least in the short term. For small businesses, independent developers, consultants, and experimenters, this matters. A predictable $10 monthly subscription encourages exploration. A usage-based model encourages caution.
But I am also slightly relieved.
The pace of AI development has been relentless. Every week seems to bring a new model, a new tool, a new workflow, or another claim about agents replacing some part of knowledge work. A little friction may not be entirely bad. It may give technical people and business leaders time to absorb what is already possible before rushing into the next wave.
That said, this is not just about budgeting. It changes behaviour. When usage is metered, people ask better questions:
- Is this task worth using a premium model for?
- Can a smaller or local model do the job well enough?
- Should this workflow be automated, or is it just interesting?
- Who owns the cost when AI is embedded into a business process?
- How do we prevent enthusiastic experimentation from becoming uncontrolled spend?
These are healthy questions.
Why Smaller Players Should Pay Attention
Enterprises will often respond by digging deeper into their wallets. Many already live with expensive software pricing models, especially around ERP, industrial systems, and large-scale business platforms. They may absorb AI cost increases and pass them on to customers.
Smaller businesses do not have that luxury.
For consultants, software developers, and industrial technology teams, AI costs will need to become part of project economics. If an AI-assisted workflow saves time but introduces variable usage charges, that cost must be understood. Not feared, but understood.
This may also create an opening for local AI models. Tools such as Ollama already make it feasible to run certain models on local hardware. Local models will not replace every cloud-based AI service, especially where the most capable frontier models are needed. But for many routine workloads, developers and experimenters may start doing the sums and decide that investing in capable hardware is better than accumulating unpredictable subscriptions.
The Likely Shape of the Next Phase
My assumption is that volatility will continue for some time. AI demand is rising quickly, and the infrastructure build-out is still catching up. Pricing models will keep changing as providers search for a sustainable balance between access, reliability, and cost recovery.
Over time, compute capacity may again outstrip demand, and we may see another era of cheaper commodity AI services. But by then, the software world may look quite different. AI will be more deeply embedded into development environments, business workflows, engineering tools, and decision processes.
The practical lesson is simple: treat AI as a strategic capability, not a free utility. Experiment, but measure. Automate, but understand the cost. Use the best tools available, but avoid becoming dependent on pricing assumptions that may not last.
Call to Action
If you are using AI in development, engineering, or business workflows, now is a good time to review where it creates real value, where it merely creates activity, and how usage-based pricing could affect your future decisions.
References
- Mario Rodriguez, “GitHub Copilot is moving to usage-based billing,” GitHub Blog, 27 April 2026. https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/
