DeepSeek recently slashed the price of its V4-Pro model by 75 percent. On the surface, that looks like unambiguously good news. Cheaper AI means lower costs for businesses, right? The reality is more complicated, and it contains a lesson every Irish business owner should understand before investing heavily in AI tools.
The problem is what industry analysts are now calling the “100x problem.” While the cost of each individual AI query is dropping fast, AI agents are consuming tokens — the units of data AIs process — at a rate that far outpaces those price cuts.
How one question becomes 35,000 tokens
Here is the key difference. A traditional chatbot turns one user question into one AI call. You ask a question, the model answers, done. An AI agent does something very different. It plans, retrieves information, uses tools, verifies its results, summarises, and makes follow-up decisions. The user sees one answer, but behind the scenes, the system has made multiple AI calls totalling tens of thousands of tokens.
Consider a simple agent query: “What did our top customer ask about last week?” In reality, that one sentence triggers roughly seven priced operations — the user prompt, system instructions, retrieval of context, a tool selection call, tool execution, a summarisation call, and a follow-up decision. Total input tokens: roughly 35,000. On a frontier model, that single question costs between ten and forty cents. Multiply by a million queries a month, and you are looking at a six-figure line item.
That is the 100x problem. A 75 percent price cut helps, but it does not fix a product architecture that turns one prompt into dozens of billable operations.
Why this matters for Irish small businesses
If you are a small business owner using AI through a subscription service — say, an AI-powered CRM, an automated customer support tool, or an AI writing assistant — you might assume the vendor absorbs these costs. In the short term, many do. But the economics do not add up.
Several software vendors are privately reporting negative gross margins on heavy AI users, according to recent industry reporting. A customer paying €40 per month for an AI-powered tool can easily cost more in inference than their subscription covers. That is not sustainable. At some point, either prices will rise, usage limits will tighten, or features that seemed free will develop costs.
The same applies if you are building your own AI-powered features. A tool that costs a few cents per query today could become a significant monthly expense as you use it more — and as its capabilities expand, you will naturally use it more.
What smart businesses are doing differently
The companies managing AI costs well are doing four things that any business can apply, regardless of size.
First, they treat AI cost as a first-class metric, tracked per feature and per use case, the same way they track cloud hosting costs. If you do not know what your AI tools actually cost per transaction, you cannot manage them.
Second, they set cost ceilings. Decide what a reasonable cost-per-thousand-queries looks like for each AI feature, and set alerts when costs exceed that level. Engineering teams will rarely enforce cost discipline on their own.
Third, they use model routing — sending simple queries to cheaper, smaller models and reserving expensive frontier models for the tasks that genuinely need them. This alone can cut inference bills by around 60 percent without any noticeable drop in quality.
Fourth, they audit their prompts. A system prompt that grew from a simple instruction to a 4,000-token document over six months is a six-figure expense in slow motion. Most teams have never read their own production prompts from end to end.
The bottom line
DeepSeek’s price cut is good news. AI is getting cheaper, and that trend will continue. But the structural shift underneath matters more: AI agents consume far more resources than the chatbots they are replacing, and that consumption is growing faster than prices are falling.
For Irish businesses, the practical advice is straightforward. Understand what your AI tools actually cost — not just the subscription price, but the usage behind it. Budget for costs to rise as you use AI more deeply. And build the habit of treating AI as an operational expense that needs managing, not a magic box that solves problems for free.
The companies that thrive in the next two years will not be the ones running the cheapest model. They will be the ones whose AI is both smart and cost-aware.