Why Are Companies Forcing AI on Us?

Artificial intelligence has been sold as the next great productivity revolution. It will transform customer service, eliminate repetitive work, discover new medicines, personalize everything, and make businesses more efficient… or so the pitch goes.

Yet many companies are struggling to show that their AI investments have produced meaningful results. At the same time, the systems behind that promise consume electricity, water, hardware, and increasingly scarce computing resources.

So why does AI keep appearing in our workplaces, products, schools, search engines, and customer-service channels – even when its value is questionable?

The short answer is that companies are responding to a mixture of genuine opportunity, competitive anxiety, investor pressure, and aggressive marketing. The result is an AI rollout that often happens before anyone has answered the most important question: what problem is this actually solving?

The business case is often weaker than the hype

Companies are not necessarily forcing AI on customers because every deployment is working. In many cases, they are doing it because they fear being seen as technologically behind.

Executives hear that competitors are using AI to cut costs. Investors ask about an “AI strategy.” Consultants and software vendors promise rapid gains. Employees are encouraged to experiment with chatbots and automated tools, even when the underlying workflow has not been redesigned.

That creates a powerful “incentive” to deploy first and measure later.

There are also more practical motivations:

• Cost-cutting: AI is presented as a way to reduce call-center staffing, administrative work, research time, and content-production costs.
• Competitive pressure: A company may adopt AI simply because it believes rivals are doing so.
• Vendor influence: Software providers are adding AI features to existing products, making adoption seem inevitable.
• Investor expectations: Companies may fear that refusing to discuss AI will make them appear stagnant.
• Data and customer capture: AI tools can encourage people to spend more time inside a company’s platform and provide more information about their behavior.
• Organizational control: Automated monitoring, evaluation, scheduling, and decision-making can give managers more oversight over workers and customers.
• Sunk costs: Once a company has spent heavily on data centers, software, and consultants, it has an incentive to keep expanding its AI program.

Some uses are genuinely valuable. AI can help identify equipment failures, forecast energy demand, optimize transport routes, detect fraud, assist with accessibility, and analyze large scientific datasets. Digital tools are already being used to track emissions and improve supply-chain reporting.

However, a useful application is very different from adding a chatbot to every website or inserting generative AI into a task that was already quick, cheap, and reliable.

The problem is that “using AI” has become a goal in itself. A company can announce an AI initiative without proving that it improves accuracy, reduces costs, saves time, raises wages, improves working conditions, or benefits customers.

The hidden environmental bill

AI systems have an environmental impact at several stages, not just when someone types a question into a chatbot.

Electricity and emissions

Training large models requires powerful computers running for long periods. After training, those models must operate in data centers every time they generate text, images, audio, video, search results, recommendations, or automated decisions.

The climate impact depends on factors including:

• the size and efficiency of the model;
• the type of hardware used;
• how many times the system is accessed;
• the complexity of each request;
• the local electricity mix;
• the efficiency of the data center; and
• whether the calculation includes training, storage, networking, and the user’s device.

This makes precise comparisons difficult. Companies generally do not publish sufficient information to reliably calculate the footprint of individual prompts, and there is no universally applied standard for reporting the energy and environmental costs of AI.

One request may have a small footprint on its own. The larger concern is scale. A system used millions or billions of times can create substantial demand even when each individual interaction appears insignificant. Efficiency improvements can also make AI cheaper to use, which may encourage far greater consumption – a phenomenon sometimes called the rebound effect.

Water consumption

Data centers generate heat, and many use water-based cooling systems. Water can also be consumed indirectly through electricity generation and the production of computer chips.

Estimates vary widely because water use depends on the location, climate, cooling technology, energy source, model, and workload. A widely cited estimate for an earlier large language model suggested that roughly 500 milliliters of water could be associated with 10 to 50 medium-length responses, including aspects of training and data-center operation. That is an estimate, not a universal amount for every modern AI query.

The local context matters. Water use in a water-rich region is not equivalent to the same use during a drought or in a community already facing water shortages.

Minerals, manufacturing, and electronic waste

AI depends on specialized processors, servers, networking equipment, cooling systems, buildings, and power infrastructure. Producing that equipment requires mining, chemicals, energy, and complex global supply chains.

Hardware also becomes obsolete. As companies race to acquire newer and faster chips, older equipment may be discarded, repurposed, or moved elsewhere. This contributes to electronic waste and extends the environmental footprint beyond the data center itself.

The impact of large outputs

Text generation is only one part of the AI economy. Image, video, speech, and “agentic” systems can require much more computation, particularly when they generate high-resolution media or repeatedly call other tools.

A short, targeted answer is not environmentally equivalent to producing dozens of images, a long video, or an autonomous workflow that performs many unnecessary steps. The environmental question is therefore not simply “How much AI do we use?” but also “What kind, for what purpose, and at what scale?”

AI is neither automatically harmful nor automatically green

It is tempting to describe AI as either a climate solution or an environmental disaster. Neither description is sufficient.

AI can reduce environmental impacts by replacing more resource-intensive activities or improving the efficiency of existing systems. Examples include:

• optimizing heating and cooling;
• reducing unnecessary transport;
• detecting leaks;
• improving electricity-grid management;
• forecasting renewable-energy production;
• reducing industrial waste;
• monitoring forests, crops, and emissions; and
• helping organizations identify where their emissions occur.

However, those benefits are not guaranteed. An AI system that saves electricity in one part of a business may increase consumption elsewhere. A recommendation engine may encourage more purchasing. Automated content production may create more demand for servers and bandwidth. A “green” claim may count a theoretical benefit while ignoring construction, hardware, rebound effects, or the system’s failure rate.

The relevant test is simple: does the application produce a measurable benefit greater than the resources it consumes?

That benefit should be demonstrated with evidence, not assumed because the product has an AI label.

How can we use less AI?

Individual choices will not determine the entire environmental footprint of the AI industry. Data-center construction, energy policy, corporate procurement, hardware standards, and transparency requirements matter far more than whether one person asks an occasional question.

Still, people can reduce unnecessary demand, and signal that not every task needs an AI layer.

For individuals

• Use ordinary search, documentation, calculators, and reference books for simple factual tasks.
• Write routine messages yourself when doing so is faster and clearer.
• Avoid generating multiple versions of text, images, or videos just because they are easy to produce.
• Use smaller or faster models when they are sufficient.
• Ask focused questions instead of repeatedly regenerating vague answers.
• Turn off optional AI features that scan, summarize, or analyze everything automatically.
• Keep and reuse useful results rather than regenerating them.
• Prefer tools that disclose their energy, water, and emissions information.
• Do not use generative AI for decisions that require human judgment, accountability, or personal understanding.
• Remember that video generation, image generation, and long automated workflows are generally more resource-intensive than a short text request.

At work

Employees can ask whether an AI tool is solving a real problem or merely adding another interface.

Before deploying one, organizations should define a baseline and measure:

• time saved;
• error rates;
• computing and energy use;
• financial cost;
• customer outcomes;
• effects on workers;
• privacy and security risks; and
• whether the same result could be achieved with a simpler tool.

Companies should also establish an “AI off-ramp”: if a system does not deliver measurable value after a defined trial period, it should be reduced, redesigned, or shut down.

The most responsible approach is not “AI everywhere.” It is the least computationally intensive tool that reliably solves the problem.

What companies should be required to do

Consumers cannot make informed choices when companies hide the basic facts. Organizations deploying AI at scale should disclose:

• the purpose of the system;
• its error and failure rates;
• how often humans review its decisions;
• energy and water use;
• emissions associated with training and operation;
• hardware and data-center requirements;
• effects on employment and working conditions; and
• measurable benefits compared with a non-AI alternative.

Public reporting would also make it easier to distinguish useful systems from expensive demonstrations. Researchers have repeatedly called for better, more standardized methods to measure AI’s direct and indirect environmental effects.

Companies should not be allowed to treat AI adoption as evidence of innovation. Innovation is not the purchase of a model, the launch of a chatbot, or the production of impressive demonstrations. It is a demonstrable improvement in the real world.

The right to say no

People are being told that AI is inevitable, but much of its expansion is driven by business decisions. Companies choose where to add it, what data to collect, which tasks to automate, and how much computing capacity to purchase.

That means those choices can be challenged.

We can ask why a chatbot has replaced a human support channel, why an AI summary is necessary, why a simple form now requires automated analysis, or why a service is generating content nobody requested. We can ask whether the system is more accurate, more affordable, more accessible, or genuinely better than what came before.

Sometimes AI will be the right tool. Often, a smaller model, a conventional program, a database, a well-designed process, or a person will do the job better.

Using less AI does not mean rejecting every useful application. It means resisting the assumption that more computation is automatically progress. In a world of finite energy, water, materials, and attention, the standard should be straightforward: use AI when it creates clear value, measure its costs honestly, and leave it out when it does not.

-TeCHS