Is AI Becoming Too Dangerous - Or Simply Too Expensive?
AI companies are spending unbelievable amounts of money building more powerful systems. At the same time, we are hearing more warnings about how dangerous those systems could become. Both may be true. But I can’t help wo
Ni Anjan N Ghosh··27K nabasa
Is AI Becoming Too Dangerous - Or Simply Too Expensive?
AI companies are spending unbelievable amounts of money building more powerful systems. At the same time, we are hearing more warnings about how dangerous those systems could become. Both may be true. But I can’t help wondering if there is another question we should be asking.
I use AI quite a lot. I find it useful, fascinating and sometimes genuinely surprising. What it can do today compared with even three or four years ago is remarkable.
But it also gets things wrong. Sometimes very confidently.
Ask it something complicated and it can give you an excellent answer. Ask another question five minutes later and you may find yourself correcting it. Anyone who uses AI regularly would have experienced this.
That’s fine. The technology is still developing.
The bigger question is what it is costing to make it better.
The amount of money now going into AI is difficult to understand. We are talking about huge data centers, thousands of expensive processors, electricity, cooling, land and all the infrastructure needed to keep these systems running.
Reuters recently reported that global investment in data centers could eventually cross $30 trillion by 2050. That’s $3 followed by 13 zeros!
I had to read that number twice.
The reason companies are willing to spend so much is quite simple. They believe AI will eventually change the way we work, write software, do research, run businesses and perhaps even discover new medicines.
They may well be right. But someone eventually has to pay for all this.
And that’s where I find the story interesting.
Millions of us are already using AI. Some pay for it, many don’t. Companies are also using it for coding, customer service, research, writing, analysis and plenty of other things.
So, there is clearly demand.
But using something and being willing to pay enough for it are two different things.
We have become used to technology being cheap. Give someone a powerful AI assistant for free and they will happily use it every day. Start charging a serious monthly amount and suddenly the same person starts wondering whether they really needed help writing that email.
Businesses are different, of course. If AI can save a company crores of rupees or make employees more productive, paying for it makes perfect sense.
But we still don’t know whether the money AI eventually earns will justify the enormous amount being spent on building it.
And those numbers are getting very large.
Anthropic, for example, reported $4.6 billion in revenue for 2025. That’s impressive.
It also lost more than $8 billion.
Again, that doesn’t mean the company is failing. Many technology companies lose money for years while growing. Amazon itself is probably the most famous example.
But investors don’t put in billions simply because something is interesting. At some point they expect those billions to come back, preferably with friends.
This is happening just as another AI discussion is getting louder.
Safety
We are regularly being warned that future AI systems could become dangerous. They may help with cyberattacks, deceive people, be misused by criminals or eventually become difficult for humans to control.
Some researchers even talk about AI becoming a threat to humanity itself.
I don’t dismiss these concerns.
If a powerful AI system can help someone attack a computer network or create something dangerous, obviously we need safeguards. The companies building these models are themselves testing for such risks.
But there is something about the timing that makes me curious.
We are being told two things at the same time.
AI still makes mistakes and needs to become much better.
And AI is becoming so powerful that we may need to slow down because it could become dangerous.
Both can be true.
But there is a third possibility sitting quietly between them.
Maybe making AI significantly better is also becoming extremely expensive.
The early improvements came quickly. Models became noticeably better from one generation to another. But improving them further now requires more computing power, more infrastructure and more money.
So, when the next big jump becomes harder, slower and more expensive, what exactly are we looking at?
A technology approaching dangerous levels of intelligence?
Or a technology that is becoming increasingly difficult and expensive to improve?
Perhaps both.
That’s the part I think deserves more discussion.
Because saying, “We need to slow down because this technology is becoming incredibly powerful” sounds very different from saying, “We need to slow down because building the next version is costing a fortune.”
Especially when you are talking to investors.
There is also something slightly amusing about the position AI finds itself in today.
One moment we’re discussing machines that may eventually become smarter than humans.
The next moment I’m asking one to correct a document and wondering why it changed the paragraph I specifically told it not to touch.
That is quite a journey between present and future.
And perhaps that’s where some perspective is needed.
AI doesn’t have to be either a disappointment or the beginning of a machine takeover.
There is a large space between those two possibilities.
It can be an extremely useful technology that continues to improve, changes many jobs and industries, creates enormous businesses and still doesn’t become the all-powerful intelligence some people predict.
The internet is a good example.
During the dot-com boom, investors poured enormous amounts of money into internet companies because everyone knew the internet was going to change the world.
They were right. The internet did change the world.
A lot of those companies still disappeared, and investors lost plenty of money along the way.
The technology succeeding didn’t automatically mean every business built around it would succeed.
AI could follow a similar path.
We may be watching one of the most important technologies of our time develop in front of us. We may also be watching too much money chasing expectations that will take much longer to become reality.
Again, both things can be true.
And this is why I think the safety discussion and the money discussion need to remain separate.
If AI poses genuine risks, let’s talk about them. Put safeguards in place. Regulate things that need regulation. Make companies responsible for what their systems can do.
But let’s also keep asking the boring questions.
How much does all this cost?
Who is paying for it?
Are customers willing to keep paying?
Does it genuinely save businesses enough money?
And when does the income start catching up with the spending?
Those questions don’t make someone anti-AI.
I certainly am not. I use the technology regularly and I’m excited to see where it goes.
I just don’t think excitement should stop us from looking at the bill.
So, the next time I hear that AI development may need to slow down because the technology is becoming too powerful, I’ll take that argument seriously.
But I’ll probably wonder about something else too.
Maybe AI really is becoming too dangerous to move this fast.
Or perhaps making it move this fast is becoming too expensive.