LLMs have been incredibly useful in boosting productivity, but lately I’ve started to feel that this paradigm may change soon and quite significantly.
Even though Spec Driven Development has greatly improved productivity, we seem to be reaching a turning point.
In the past, several platforms offered so-called “coding plans”, where for $10, $50, $100, or $200 you could get a fixed number of requests every 5 hours (or per day/week, depending on the provider). This was an excellent deal for developers.
However, due to current datacenter limitations, this model may not last. There is investment and hardware available, but there’s a shortage of critical resources like electricity and water to sustain expansion. Spain is already a good example, where even housing development is being impacted due to energy infrastructure limitations.
We are now facing a real “compute” problem. Coding plans may disappear and shift towards pure pay-per-inference or token-based pricing — which will be significantly more expensive.
For example, for a freelance developer, using services like Anthropic might become unviable. Around 80 requests to implement a simple feature (roughly 10 minutes of work) can cost about $6. By the end of the day, this can easily exceed $30/day using models like Claude 4.6 Opus.
At the same time, most platforms (“harnesses”) are removing free tiers (as recently seen with Qwen Code), and when they do offer them, they tend to rely on lower-tier models like GLM, Minimax, Xiaomi, or Qwen.
For those who can work with these tier-2 models, there are still some alternatives. Minimax offers a $20 plan with around 4500 requests per 5 hours. Ollama claims better performance, but lacks transparency regarding limits. Qwen 3.6 Plus, while decent, costs $50 and often requires waiting for available capacity, likely due to compute constraints.
Meanwhile, European companies are still far from using AI efficiently, whether in task automation or security. It’s concerning to imagine a future where they become dependent on external compute providers.
Are we heading back to partial reliance on human labor?
Will we soon return to our old friend… Stack Overflow?
