As executives at top AI companies grow more attuned to rising privacy and safety concerns and urge a slowdown in AI development, experts argue that a broad slowdown is unrealistic and may not solve these issues so easily. Meanwhile, debates over returns on investment (ROI) raise an uncomfortable question: do calls for a slowdown come from genuine safety concerns, or do economic pressures also play a significant role?
Artificial intelligence (AI) models are advancing so quickly that, at this juncture, even researchers struggle to understand and control their behavior. For instance, agentic AI systems can use various tools to make decisions with some autonomy and sometimes take desperate actions (even crossing ethical boundaries, e.g., hacking websites without authorization) to achieve their goals. AI companies, yet, are racing to develop more powerful systems to attract customers, secure investment, and gain market dominance. These commercial pressures make it difficult to balance rapid development with growing concerns about AI safety.
A broad slowdown would be difficult to achieve as AI companies compete to stay ahead, according to a recent discussion by Stanford’s Institute for Human-Centered Artificial Intelligence. The discussion highlighted this contradiction (the urgency of AI safety versus the pressure to innovate), while experts stressed the need for independent testing, stronger oversight, and greater investment in AI safety.
The question, ultimately, is not whether AI poses risks; rather, it is whether the industry’s current incentives allow those risks to receive the attention they deserve.
Safety and the Limits of a Slowdown
AI agents can now coordinate tasks, access digital resources, and work with less human supervision. This makes their behavior harder to predict. Researchers at Stanford have raised concerns about how agents behave when they interact with one another, and about the weak methods currently used to evaluate advanced systems.
Experts, therefore, call for several layers of protection: AI firms may limit an agent’s goals, make rules it cannot break, create models that check each other, and assign human experts who step in when needed.
But a slowdown does not guarantee any of this. A company can delay a product and still leave its testing practices unchanged. Another company can keep building and still act responsibly, if it invests properly in risk assessment.
So what are AI companies actually willing to change right now? That is the real question. Safety must be judged through independent testing and accountability, rather than public calls for a slowdown in AI development.
The Financial Question Behind AI Development
The economics of AI make this debate harder. Building advanced models takes huge computing power, specialized chips, a lot of electricity, and skilled research teams. Companies must earn enough revenue to justify these costs, and their rivals are chasing the same goals.
The evidence of AI’s business value is mixed. According to Stanford HAI’s 2025 AI Index, 49% of companies that use AI in service operations reported cost savings. In marketing and sales, 71% reported higher revenue. But most of the savings were below 10%, and most of the revenue gains were below 5%.
The costs are not only financial. A study conducted by UN University estimates that data centers could use 945 terawatt-hours of electricity by 2030, have a water footprint of 9.3 trillion liters (equal to the basic annual domestic needs of 1.3 billion people), and generate 2.5 million tonnes of e-waste a year. More than 90% of AI-specialized computing capacity sits in just two countries, interestingly, the United States and China, while over 150 nations lack significant domestic AI infrastructure. The study warns that this imbalance limits economic opportunities for those countries, which may also bear the environmental costs without sharing in the benefits of AI-driven growth.
Those numbers above from Stanford HAI’s AI Index, however, do not tell us whether AI companies are making money. They only show what other businesses gain from using AI. How the developers themselves are doing is a separate question.
Still, the findings raise a bigger question about expectations. If businesses are not yet sure how much value AI can deliver, developers must convince customers and investors that these increasingly expensive systems will pay off in the end.
Commercial ambition and responsible development often pull in different directions. A company can worry about safety and still want to stay ahead of its rivals. So which of the two should we ignore? Perhaps neither.
There is not enough evidence to say that calls for a slowdown are only a way to protect profits. But the industry should not escape scrutiny just because its leaders talk about safety.
Who Controls the Infrastructure?
Every AI model runs on chips, and NVIDIA sits at the center of that trade, without any shadow of a doubt, as it commands approximately 80% to 90% of AI accelerator market by revenue in recent years. It builds the processors most developers rely on, along with the software that keeps them locked in. Its fiscal 2026 financial results show how much money flows through that one company.
Dependence on advanced infrastructure can make competition expensive for developers without comparable resources. The race to build more capable models is therefore also a contest over access to capital, computing power and technical expertise.
The physical costs matter, too. Analysis of AI data centers’ environmental costs highlights the resource demands associated with expanding infrastructure. Electricity and water consumption should form part of decisions about where and how new facilities are built.
These pressures do not prove that the industry needs to slow down across the board. They do, however, challenge the assumption that faster growth automatically represents better progress.
What should Bangladesh do to keep AI accountable?
Most AI companies run their own safety tests and publish their own results. Outsiders cannot verify them. A company may promise to slow down, but nobody can tell if it did.
A responsible company would act differently. It would hold back a launch when its tests find a problem. It would pay for the fix, even if a rival gets ahead in the meantime.
A broad slowdown is therefore may be difficult to coordinate in a competitive global industry.
For Bangladesh, these questions arise mainly from its position as a user of global AI services. Yet the country also has an opportunity to build solutions around local needs.
Most people in Bangladesh use mainly two AI chatbots. According to Statcounter’s September 2026 data, ChatGPT has 74.76% of the market. Google Gemini has 21.64%. Together, that is more than 96%. Surprisingly, both are US-authorized products and have also received massive criticism over their systems that fail to protect user privacy.
That reliance, hence, raises questions about who controls users’ data, who is held responsible when things go wrong, and how Bangladesh can ensure fair access to AI. Bangladesh should strengthen its capacity to evaluate AI systems, and develop its own models for meeting local needs in various sectors (including education, healthcare, and public services) rather than immediately competing with global model developers. At least 20,000 ICT students graduate from public and private institutions, report says, and enter the job market every year. They, of course, need realistic career opportunities as Bangladesh invests in AI education, promotes AI literacy, and develops policies to govern its use.
Bangladesh should also continue working with other nations at the United Nations, where it has placed a draft AI policy before the rights body, so that countries can jointly build safety guidelines that protect people across borders.
There is a precedent for this kind of cooperation. Concern over nuclear technology first led to the proposed United Nations Atomic Energy Commission and later to the International Atomic Energy Agency. Whether Bangladesh can lead the creation of a similar body to oversee AI infrastructure is uncertain, but the effort is worth making if it helps protect privacy of millions of people.
A responsible strategy would encourage innovation and still require safeguards that match the risks. The country should also make sure that technology investment does not cost it its privacy or its environmental standards.
Governments and users must demand evidence, and do not accept promises until a credible, independent commission has verified AI companies’ systems and approved them for commercial use. AI developers should prove that safety gets as much money and attention as the push for more powerful models.
Ultimately, AI progress should be judged not only by the power of its models or the scale of investment they attract, but also by whether their risks are understood, their benefits justify their costs and the companies developing them can be held accountable.
References:
[1] Stanford Institute for Human-Centered Artificial Intelligence (September 22, 2026). Can AI Be Slowed Down? Stanford HAI Experts Weigh the Risks, Rules and Race Ahead. [https://hai.stanford.edu/news/can-ai-be-slowed-down-stanford-hai-experts-weigh-the-risks-rules-and-race-ahead]
[2] Stanford Institute for Human-Centered Artificial Intelligence (April 7, 2025). The 2025 AI Index Report. [https://hai.stanford.edu/ai-index/2025-ai-index-report/economy]
[3] UNU Institute for Water, Environment and Health (June 3, 2026). Rising Emissions, Depleting Water and Vanishing Land—UN Scientists: AI Is Threatening Natural Resources for Billions. [https://unu.edu/inweh/news/environmental-cost-of-AIs-Enrgy-use-carbon-water-and-land-footprints]
[4] NVIDIA (February 25, 2026). NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2026. [https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Fourth-Quarter-and-Fiscal-2026/]
[5] Statcounter Global Stats. AI Chatbot Market Share in Bangladesh – September 2026. [https://gs.statcounter.com/ai-chatbot-market-share/all/bangladesh]
