Waltes

Big tech companies like Microsoft, Google, and Anthropic are selling paid versions of their AI technologies after investing hundreds of billions of dollars in developing Large Language Models (LLMs). If you’ve ever used a free version of ChatGPT or any of its competitors, you know you’re getting a sweet deal. However, these companies are eager to recoup their investments, which is why they offer premium versions of their AI tools, boasting extra features for tasks like coding or billing. Meanwhile, third-party firms are leveraging these AI agents to create and sell tailored services. But here’s the kicker: pricing these services is proving to be a real headache.

You see, the economics around tokens—the fundamental units that power LLMs and agentic AI—are changing rapidly. When you interact with an LLM like ChatGPT, your queries are transformed into tokens, which are then converted back into text, software code, or commands to automate various processes. But the tricky part? This entire procedure is riddled with unpredictability. Minor changes in input can lead to wildly different outputs, and different models yield different responses altogether.

As companies utilize multiple AI agents to make decisions and take actions, the complexity of token consumption escalates. Even though the cost per token has plummeted recently, the overall volume of tokens consumed by businesses and consumers has skyrocketed. According to Goldman Sachs, external token consumption is expected to increase a whopping 24 times between 2026 and 2030, reaching an astonishing 120 quadrillion tokens a month as companies ramp up their use of AI agents.

But here’s where it gets complicated: businesses and individuals often have a shaky understanding of how many tokens they’re actually burning through. This realization usually hits them hard when they either run out of tokens or receive their monthly bill. Even Microsoft has reportedly started to scale back on external token consumption, revealing that people struggle to manage these costs effectively. “It’s a non-deterministic output; hence it’s a non-deterministic value,” says one insider. Some folks might even slip under the radar by using flat-fee personal accounts—a tactic that big vendors probably don’t appreciate.

Now, when the major AI platforms begin feeling the heat from shareholders demanding profits, predictions suggest that they’ll tighten the screws on pricing. The situation can spiral out of control, especially when companies integrate AI into products that could be used by thousands. AI costs could balloon unexpectedly. For instance, managers might realize they need tokens not just for core software development but for testing, security, or implementing safeguards. “It’s particularly challenging when you’re dealing with agentic processes,” one expert points out.

However, companies are still left with the daunting task of passing those costs onto their customers. “Nobody’s really cracked the code on this,” he adds dryly. Options for managing costs could include simply raising prices across the board, charging based on results, or bundling services. But this leads to a nightmare of variable pricing, constantly changing every few months—something customers absolutely detest. After all, who can budget for that?

As we navigate this ever-evolving landscape of AI tokenomics, one can’t help but wonder: how will companies adapt their pricing models to keep pace with the rapid changes in token consumption? The future remains uncertain, but the stakes have never been higher.

Kaynak: Orijinal Haber