The GitHub Copilot News is Just the Latest Sign That the Era of Free Compute is Over
The era of model freedom is just beginning
Remember the final season of the TV show Lost? Things got super exciting, but the characters’ complicated backstories started becoming even more complex — and they sort of stopped making sense. Audiences were up in arms about every plot twist, and discussion surged online about whether the island even existed in the first place.
It’s starting to feel a bit like that in AI, especially in the realm of agentic engineering. Major announcements from GitHub Copilot, Cursor, Anthropic, OpenAI and others have made it hard to track all the changes. Yet despite the turbulence of the past few weeks, there’s actually a solid thread running through it all. We’ve reached the end of the era of free compute.
How did we get here? And could these changes actually be a good thing for the future of AI coding?
GitHub Just Made the Economics Visible
Starting June 1, GitHub Copilot moves to usage-based billing, replacing premium request units with GitHub AI Credits calculated by token consumption — input, output, and cached. Plan prices aren’t changing (though it looks like new Pro and Plus subscriptions are still blocked). What those dollars buy is shrinking. Each plan now includes AI Credits equal to its subscription cost — $10 for Pro, $39 for Pro+. Once those credits are gone, there’s no safety net.
That last part is the real signal. A fallback to a free, simple model — as before — will no longer be available.
GitHub’s own CPO put it clearly: “Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount. GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable.”
Copilot has evolved from an in-editor assistant into an agentic platform capable of running long, multi-step coding sessions using the latest models and iterating across entire repositories. Agentic usage is becoming the default for platforms like Copilot, Cursor and Kilo, and this brings significantly higher compute and inference demands.
GitHub isn’t alone in the move to usage-based pricing. Anthropic made a similar move recently, quietly steering users away from flat subscriptions and toward consumption-based API access. The direction across the industry is identical: the era of subsidized, all-you-can-eat AI is ending — and the reason is always the same underlying constraint.
Compute.
The SpaceX–Cursor Deal Is Really About Compute
A lot of folks read the SpaceX–Cursor deal as a productivity story. A big engineering org standardizes on a fast AI coding tool. That’s true. But it misses a bigger signal.
SpaceX has struck a deal with Cursor to develop a next-generation “coding and knowledge work AI” — including an option to acquire the company for $60 billion later this year. SpaceX described the partnership as combining Cursor’s “product and distribution to expert software engineers” with SpaceX’s Colossus supercomputer, which the company claims has the equivalent compute power of a million Nvidia H100 chips.
A rocketry and satellite company is deploying a million-H100-equivalent supercomputer and pairing it with an AI coding tool. This is not a software story, even keeping in mind that SpaceX also recently absorbed xAI, maker of the popular Grok models.
The backstory makes it clearer. xAI began renting computing power from its data centers to Cursor, with the coding startup using tens of thousands of xAI chips to train its latest AI model. Cursor isn’t just an IDE. It sits at the demand layer, deciding where millions of developer requests get routed and how much inference gets consumed. When SpaceX standardizes around it, they’re not picking a coding assistant. They’re securing guaranteed throughput in a constrained market.
The deal’s structure reflects this: SpaceX will either pay Cursor $10 billion for its work or acquire the company outright for $60 billion. That’s not a typical enterprise software procurement. In my view, that’s really a compute alignment deal.
Even if you look deeper and look at the models themselves, the truth is in the compute.
Cursor’s own model work was built on strong open foundations like Moonshot AI’s Kimi K2.5, a sign of how fast iteration happens when good models meet the right workflow layer. Yet xAI is not retreating from the frontier; it’s already claiming that a 3 trillion parameter Grok release is coming as soon as May. And if that model goes live – or if some new, equally large combination of all of these model iterations emerges – what will be the only real bottleneck?
Compute, compute, compute.
This is exactly the world Kilo Code was built for
While GitHub Copilot continues to iterate to figure out how to package their offering to developers, and nobody knows quite what will happen with SpaceX, Kilo has always had transparent pricing and wide BYOK coverage.
To make things even more interesting, yesterday OpenAI announced that they were refiguring their exclusive deal with Microsoft and opening up the provider landscape again. Their deal with AWS Bedrock expands opportunities not just for Kilo Code and how we serve models to our 2M+ users, but also for our enterprise customers.
The expanded partnership between OpenAI and AWS puts OpenAI’s frontier models, including GPT-5.5, directly onto Amazon Bedrock alongside Anthropic, Meta, DeepSeek, and the rest (they seem to be adding more open models like Kimi K2.5 recently as well – here’s the current list of available models on Bedrock).
Amazon Bedrock is built on the principle that customers should choose the best model for every use case. For years, the Azure/OpenAI relationship meant those two were effectively one choice. Much of this was made possible by Microsoft’s willingness to open its relationship with OpenAI in exchange for being freed from revenue-sharing commitments — leaving OpenAI free to partner with any cloud provider it likes, with AWS potentially a blueprint for future deals.
That’s a meaningful shift. OpenAI models on Bedrock inherit the enterprise controls customers already depend on — IAM, guardrails, encryption, CloudTrail logging — and from what we’ve read so far, usage can likely be applied toward existing AWS cloud commitments.
When Kilo committed to open-source development, model freedom, and model choice, the prevailing wisdom was that developers would consolidate around one or two dominant providers. Pick your vendor, buy the subscription, and get on with it. That was the pitch from every direction — and for a moment, it was almost convincing.
But it wasn’t right.
What’s actually happened is that every major AI platform has converged on the same problem: compute is expensive, subsidies are unsustainable, and the only honest path forward is to make it easy and clear for users to simply pay for what they use. That’s rational. But it means developers who locked themselves into a single provider are now at that provider’s mercy — on pricing, on model availability, on what gets deprecated when the economics shift again.
Kilo was built on the opposite assumption: that developers should never have to care which vendor controls the model, or where the compute sits, or what this month’s pricing change means for their workflow. It’s great to see some of the bigger labs coming around to this idea too.
Model choice is a right, not a premium feature. And open source isn’t a consolation prize; it’s the only foundation that stays stable when closed systems reprice overnight.
PS. Looking for model freedom? Try out Kilo Pass, with instant access to 500+ models and never any surcharge.