AI Is Getting Cheaper, Smarter, Greener

Scaling Without Additional Humans

There is a chart doing the rounds on social media. It shows that to take home £120,000 in Dubai you must earn £120,000. The equivalent in the UK is £205,000, which is more than in France, Germany or the US. Payroll taxes are the reason.

Dubai is short of people. To attract them, there is no income tax. Revenue is raised on spending, with higher amounts reserved for more luxurious items. The basics of food, shelter and education remain tax free.

We see strong demand at MSBC for AI services in the UAE, of which Dubai is part. While much of the economy is driven by construction of offices and residential communities, there is a plan in place to scale businesses without human capital. Government is at the forefront of development.

AI requires energy and Dubai has little oil. It relies on its wealthier and more powerful neighbour Abu Dhabi. Between them there is energy to desalinate water, cool buildings and power data centres. These economies are planning for the next stage of economic growth.

Falling Operating Costs

Our experience is that demand for AI in the UK is at an earlier stage. There are questions over growth in the economy and worries among workers about losing jobs. While the population is ageing, this is a slow process and does not create the same labour shortage that faster growing Dubai is experiencing. High electricity costs are an additional hurdle across Europe.

The challenge is the efficient conversion of energy into intelligence. Businesses may think of this as the running costs for a given amount of useful AI. It is my view that usefulness will increase while operating costs fall. The signs are promising.

Google released data showing the energy consumption of a Gemini model prompt. The average one consumes 0.24 watt-hours of energy, results in 0.03 grams of CO₂ emissions and requires five drops of water. That’s the energy equivalent of watching TV for nine seconds, while 150 prompts have the same emissions as charging your phone.

AI processors are around 60% of the total energy costs. The balance comes from idle capacity, data centre cooling systems and overhead. Google calls this a full-stack calculation.

Model improvements and software optimisations mean the energy consumption is 33 times lower than in May 2024. At OpenAI, ChatGPT-5 decides when to answer queries quickly and when to deploy more energy consumptive reasoning. The preference is to work smarter rather than to charge consumers based on usage. Google’s figures are a rebuttal to those complaining about the environmental costs of AI, but not yet enough.

The data is for the median prompt. That means half of all prompts consume more energy and often much more, given this does not include deep reasoning, image or video generation. There is no data on total energy consumption across all prompts and the data covers inference, or running costs, while excluding training the model.

For these reasons the environmental debate will go on. It will not stop AI developing, however. The energy and emissions data will keep improving.

Nvidia is committed to making its chips at least twice as productive and half as energy consumptive every year. This is not the result of regulation and pressure from protest groups. It is a competitive advantage that keeps the firm at the forefront of the industry by making its products evermore valuable for customers.

Drivers and Passengers

Time is always the answer to those who want to ban progress because they believe they are alive at the optimal stage of human achievement. AI is no different in this regard to any previous technology. It will get cheaper, better and more environmentally friendly because there is an economic incentive to do so.

Countries will decide whether they are AI drivers or passengers. The data centres planned in the Middle East have the potential to meet the compute requirements of all but the far reaches of Europe. The continent can support its nascent AI industry, or import technology from abroad. The decisions taken in the next few years will impact the rest of the century.

Businesses will decide where they are on the AI adoption curve. In the UAE we meet companies that will spring from spreadsheets straight into AI-controlled processes. European companies might believe themselves more efficient and as a result lack urgency. They expect their market size to protect them.

Competition, customer demand and regulation will determine the pace of AI adoption. The risk for businesses that do not take a global view, or hide behind rules that slow the rollout of AI, is that international competition will move too far ahead to be caught.

Dubai has its challenges. Attracting bright, young Europeans to work there is not one of them. When they get there they find companies that are ready and willing to use AI. Being paid tax-free to be at the frontier of technology is a powerful incentive. UK companies must deal with the disadvantage of high payroll taxes. One way is to use AI adoption to become more exciting places to work.

Questions to Ask and Answer

1. Are there smarter, lower-cost ways to integrate AI tools?

2. Am I experimenting, applying, or embedding AI in my business?

3. What am I doing to make my company attractive to young talent?

Find out more. Hit reply and ask about becoming an AI-ready organisation.