The end of cheap AI plans

If you've been enjoying AI tools at bargain prices, enjoy it while it lasts. The era of cheap AI is ending, and the signs are everywhere. For the past two years, AI companies have been running the Silicon Valley playbook we've seen before: subsidize heavily, acquire users, worry about profits later. It worked for Uber, it worked for Amazon, and it's been working for AI. But just like those predecessors, the bill is coming due.

The GLM wake-up call

Zhipu AI's GLM pricing saga is perhaps the clearest example of how fast things can change. In late 2025, their GLM Coding Plan went viral internationally with a promotional price of roughly $3 per month. Developers flooded in. Demand was so intense that new sign-ups had to be capped at 20% of normal capacity because inference requests overwhelmed the infrastructure. Then, on February 11, 2026, Zhipu announced the end of promotional pricing, citing growing demand and rising compute costs. First-purchase discounts were removed, and overseas pricing jumped to around $10 per month for the Lite tier. But that was just the beginning. By April 2026, alongside the release of GLM-5.1, Zhipu raised API prices again by 8% to 17%. The Pro plan, once a hidden gem at $180 per year, spiked to over $680, a roughly 3x increase. Their pitch had been "frontier capabilities at a fraction of the cost." That fraction got a lot bigger.

It's not just GLM

Zhipu isn't an outlier. They're a leading indicator. OpenAI, Anthropic, Google, and xAI are all overhauling their pricing structures. The old model of near-unlimited use for a flat $20 per month is crumbling. In its place, we're seeing high-end plans above $200, restricted access to premium features, usage caps on previously "unlimited" tiers, and the introduction of ads on free plans. As Axios reported in March 2026: "AI may never be as cheap to use as it is today." May Habib, CEO of Writer, put it bluntly: "These LLM companies are going to go public and they're going to raise prices because they have to." The numbers back this up. OpenAI projects $115 billion in cumulative cash burn through 2029. Anthropic is targeting break-even by 2028, but only through aggressive enterprise pricing. Hyperscalers committed $380 billion combined to AI infrastructure through 2026. Someone has to pay for all that compute, and increasingly, that someone is you.

The Uber playbook

Industry experts have started calling this AI's "Uber moment," and the comparison is apt. Remember when Uber rides were magically cheap? That wasn't because the service was inherently affordable. It was because venture capital was subsidizing every ride. The same dynamic has been playing out in AI. Every query you send to ChatGPT, Claude, or Gemini costs more to serve than what you're paying. The companies are eating the difference to build market share. But subsidies don't last forever. As AI companies eye IPOs and investors demand a path to profitability, prices have to come up. The question isn't whether it will happen, it's how fast.

The pricing pivot

What's replacing the old flat-rate model is more complex and, in many cases, more expensive. Usage-based pricing is becoming the norm. Sam Altman has described a future where "intelligence is a utility like electricity or water, and people buy it from us on a meter." That sounds elegant, but metered pricing means your costs scale with your usage, and that bill can surprise you. Tiered access is fragmenting what used to be unified products. Want the best model? That's the $200 plan. Want faster inference? That's extra. Want to keep unlimited usage? That's going away. Enterprise focus is shifting where companies invest their attention. Consumer plans are becoming loss leaders (or getting ads), while the real product development targets businesses willing to pay premium rates. Google Workspace added Gemini AI into Business plans with price increases of 16% to 33%. Adobe rebranded Creative Cloud with a $10 per month hike tied to AI features. Atlassian raised Jira pricing 5% to 10%. These aren't optional AI add-ons. They're baked into the base price whether you want the AI or not.

What about the "cheap" alternatives?

DeepSeek remains remarkably affordable, with V3.2 priced at $0.14 per million input tokens. Open-source models can be run locally for the cost of hardware. But even the budget options face pressure. The fundamental economics haven't changed: running inference on large models requires expensive GPUs, massive amounts of electricity, and specialized infrastructure. DRAM costs are rising. Energy constraints are real. Every "cheap" provider either has a benefactor subsidizing costs (often a government), is running at a loss, or is about to raise prices. Zhipu was the cheap alternative. Look where their prices are now.

What this means for you

If you're building products on top of AI APIs, your cost structure is about to change. If you're a consumer enjoying $20-per-month access to frontier models, expect either higher prices or more restrictions. A few practical takeaways:

The era of artificially cheap AI was always temporary. The technology is genuinely transformative, and transformative technology tends to be priced accordingly. The companies that built these models need to recoup hundreds of billions in investment, and that money has to come from somewhere. Cheap AI was the hook. Now comes the real price.

References