<\/p>\n
To understand this whole matter, first let us look at what has happened in the giant tech company Microsoft. Microsoft had allowed thousands of its software engineers to use ‘Cloud Code’ AI tool for tasks like coding, troubleshooting and code review. This tool is quite powerful. In the first six months, the engineers liked this tool very much and used it extensively. But the real trouble started when the company faced the bill for using this AI tool.<\/p>\n
When thousands of engineers started using this tool day and night, its total expense became a big headache even for a big company like Microsoft. The result was that Microsoft decided to shift its engineers from this expensive tool to its own cheaper AI tools and the GitHub Copilot stack. The news was published in many places that Microsoft has banned the use of AI by its employees.<\/p>\n
However, calling it a ban on AI is not entirely correct. More precisely, the company did vendor consolidation. Tools under our control were given priority instead of expensive tools from outside, so that expenses could be controlled. The real issue was not the utility of AI, but its cost per unit.<\/p>\n
Something similar happened with the world’s largest cab service company Uber. Uber introduced this cloud code tool for its engineers in December 2025, and by March, about 84% of the company’s 5,000 engineers were using it. The situation inside the company was such that about 70% of the software code was coming directly from the AI \u200b\u200bsystem. But the second and scary aspect of this was that the expenses of some heavy user engineers who used AI a lot were coming from $ 500 to $ 2,000 (about Rs 40,000 to Rs 1.6 lakh) every month.<\/p>\n
Talking about this, Uber’s Chief Technology Officer Praveen Neppalli Naga said that the AI \u200b\u200bbudget that the company had fixed for the whole year was completely exhausted by April. The company had even created an internal leaderboard to create a competition among the employees to enthusiastically use AI, where engineers were being ranked on the basis of how much they were using AI. This means that a game-like environment was created for the adoption of AI, but there was no strong plan to control the expenditure.<\/p>\n
After all, why does using AI cost such a huge bill? Actually, tools like Cloud Code work on token-based pricing. This simply means that whenever you ask AI a question, write code or check a large file, AI divides the entire text into small pieces called ‘tokens’.<\/p>\n
If you understand in simple words, AI takes money to process your every word, space and comma. Just like in the olden days, a meter of every second ticked while talking to a PCO, similarly in AI tools, with every query, every code review and every debugging session, the token meter keeps ticking and its cost gets added. When one or two people run it, the expense is not known, but when thousands of engineers of a company run the prompt repeatedly around the clock, then the consumption of tokens reaches lakhs and crores and the bill suddenly becomes many times bigger than expected.<\/p>\n
Regarding this increasing expenditure, the estimates of organizations like famous investment bank Goldman Sachs and research company Gartner also point in the same direction. In the coming years, the consumption of tokens within companies may increase manifold. Although it is a matter of relief that due to increasing competition among tech companies, the price per token is continuously decreasing, but despite this the total bill of the companies may increase. This will happen because in the coming times, AI Agents will do more complex work. When the task is big and difficult, AI will have to spend a lot of data and tokens to process it. This means that even if the price per token decreases, the total use of AI will increase so much that the AI \u200b\u200bbills of companies will keep increasing.<\/p>\n
Meanwhile, Brian Catanzaro, Vice President of NVIDIA, the world’s largest AI chip and GPU manufacturing company, has said something very important. He told that the cost of running computers and processing data for his own team has exceeded the salary paid to human employees and their total expenses. This statement is very big because Nvidia itself sells the infrastructure and chips required to run AI to the whole world. If the compute cost within the company that builds the AI \u200b\u200binfrastructure is becoming more expensive than the employees, then this concern may become even more serious for other companies.<\/p>\n
However, amid this concern, there is also a strong counter-argument, which needs to be heard and understood. Many experts believe that just saying that \u201cAI is becoming more expensive than humans\u201d is an incomplete statement. The real math behind this is that if a heavy AI user engineer at Uber bills around $16,000 a year, and that engineer’s total annual salary and benefits are around $577,000, then this AI expense is not even 3% of his salary. In such a situation, if that AI tool increases the working capacity of that engineer by just 3%, then the entire expenditure incurred on AI is easily recovered. So the real problem is not that AI is expensive, but the question is whether the benefits companies get from AI outweigh its costs or not?<\/p>\n
Looking at these situations of Microsoft and Uber, it would be completely wrong to understand that companies are retreating from AI or rejecting it. No company is completely stopping the use of AI, rather they are now working on how to keep this expenditure under control with discipline. Anyway, AI models are now becoming increasingly cheap. New Chinese models like DeepSeek have increased the pricing pressure by coming into the market, due to which other companies are also having to reduce their prices. Self<\/p>\n
A model like Google’s Gemini Flash is giving very strong performance at a very low cost.
The lesson for companies around the world is that to successfully implement AI, it is not enough to just adopt it, but to control its expenses, strong cost governance will have to be adopted. Companies need to teach better prompt discipline to their employees, so that they do not waste tokens by asking unnecessary questions again and again. Also, companies will have to decide which tasks require expensive AI tools and which tasks can be handled with cheaper or open-source models.<\/p>\n<\/div>\n