Apparently, AI is going to kill us. Before that, it is going to take our jobs, hack us, replace developers, write our emails and manage our social media, badly. There is no shortage of bad news.
But I keep coming back to a different problem. What if we are all looking at the wrong thing? Because I think there is a much more immediate risk sitting right in front of us.
The companies building the best AI in the world are spending billions trying to stay ahead. They cannot really stop spending because the moment they stop improving, somebody else catches them. At the same time, all that work is making AI cheaper. Open models get better. Smaller models get better. Hardware gets better. China gets closer. The gap between the best model and the next best model gets smaller.
So the companies at the front have a strange problem. They have to keep spending huge amounts of money to build something that could eventually become cheap enough that nobody wants to pay them huge amounts for it.
That is what this article is really about. Not whether AI wakes up one morning and decides it does not like us. It’s about whether the companies building it can survive the race they have created – and whether, if that race starts going wrong, the damage stays inside the AI industry.
I do not think it does.
Nobody Really Knows Where This Ends
Before I go any further, there is something I really believe. I don’t think there is one person alive today who knows exactly where AI is going. I definitely don’t. And I get nervous around people who think they know it all. The technology is moving too quickly for that.
What I do have is years of experience focusing heavily on this space. Studying it, building with it, working around it, speaking to customers, vendors, security companies and people much cleverer than me. And I actually use it. A lot. That matters.
One thing I learned a long time ago is that if you want to understand something properly, you have to get close to it. Whatever your view of Elon Musk, I always understood the story about him sleeping at the Tesla factory when production was going badly. I have done plenty of that in my own career. Late nights, being right in the middle of a problem, sitting with the people actually doing the work rather than looking at it from a distance. You see things differently from inside.
That is how I try to look at AI. Use it. Build with it. Break things. Watch how quickly it changes. Talk to the people trying to secure it. Watch what customers are really doing with it. And follow the money. The money is where things get really interesting.
Everyone Agrees AI Should Slow Down. Great. Now What?
Something weird has happened recently. Leaders from some of the world’s biggest AI companies have started publicly suggesting that frontier AI development may need to slow down.
Anthropic’s Dario Amodei recently called for the pace of frontier AI development to slow, with OpenAI’s Sam Altman and Elon Musk publicly backing the need for greater caution.
That is worth paying attention to. These companies spend most of their time trying to beat each other. When the people building the technology start saying, “We might want to be a bit careful here”, I think we should listen.
And if the big Western AI companies manage to work together and slow the pace, that solves part of the problem. But only part of it. Because China has not disappeared. Open models have not disappeared. Researchers around the world have not disappeared. The technology does not suddenly stop improving because four American companies agree to take things more slowly.
And that creates a horrible problem. Imagine the biggest Western AI companies agree to slow down. They spend more time testing. More time on security. More time making sure the next model behaves properly. All very sensible. But what happens if, while they are doing that, a Chinese company releases something almost as good? Then another. Then an open model appears that companies can run themselves. How long does the agreement last?
That is the bit I find really interesting. We keep talking about slowing AI as if this is a decision for the technology companies. It is bigger than them now. It is about money. It is about national security. It is about defence. It is about jobs. It is about which country has the best technology. And it is about some of the biggest companies on the stock market.
Nobody wants to lose. That is what makes slowing down so difficult. The Western companies might agree with each other. Their global competitors do not have to.
And They Can't Stop Spending
This is the next part people sometimes miss. These companies need money. Lots of it.
Building the best AI models costs a huge amount of money. You need chips. Lots of chips. Then data centres. Then electricity. Then cooling because those chips get hot while helping me work out whether an email sounds too rude. Then you need some of the best technical people in the world. Then you build the model. Then millions of people use it, which also costs money. Then, just as everybody is excited about what you have built, somebody releases something better. So you go again.
To put the scale into perspective, the Financial Times recently reported that OpenAI expects almost $280 billion in negative free cash flow between 2026 and 2030 as it continues investing heavily in computing power and infrastructure.
The whole thing is slightly mad. You spend billions building the best AI in the world. Then your next job is to spend billions making the AI you just built look old. And you cannot really take a year off. Because the investment only keeps coming if people believe you are still near the front. Customers stay because they believe you will keep improving. Developers stay because they believe your model will keep getting better. Your value depends heavily on what people think you are going to build next.
So the rate of improvement is not just about technology. It keeps the whole machine moving.
Slow the innovation too much and suddenly the question changes from “How much better will the next model be?” to “Why are we putting all this money into this company?”
That is a much less comfortable question.
The Best Model in the World Has a Very Short Shelf Life
Imagine spending billions becoming number one. For a few months everybody talks about you. Developers move over. Businesses start testing you. You are top of every comparison. Amazing!
Then somebody gets close. Then somebody else offers something cheaper. Then an open model comes along that is not quite as good but does everything most businesses actually need. Then somebody works out how to run something similar using cheaper hardware. Suddenly the thing you spent billions creating starts looking quite normal.
That is brutal. AI companies are spending huge amounts of money creating an advantage that sometimes lasts months. And every time they make AI better, they also make the older AI cheaper. Yesterday’s amazing model becomes today’s normal model. Today’s expensive model becomes tomorrow’s cheap one. At some point we could end up with very capable AI available almost everywhere for very little money. Brilliant for us. Potentially horrible for the company that spent billions expecting everyone to keep paying a premium.
And this is where China becomes really interesting.
What if China Wins Because It Has Less?
For a long time, the story seemed obvious. America has the best chips. The biggest American technology companies have huge amounts of money. So, America stays comfortably ahead. I am not convinced it’s that simple.
China has not had the same access to the very best hardware. That is clearly a disadvantage. But having less can force you to get very good at using what you have. If I can solve every problem by buying another few thousand expensive chips, there is less pressure on me to find a smarter way. If you cannot do that, you have no choice. You have to squeeze more out of less.
And there is evidence that the gap is narrowing. Mozilla’s latest State of Open Source AI research found that leading open-weight models, many developed in China, now trail closed frontier systems by around 4.4 months on one capability measure. Its analysis also found substantial cost differences for some comparable workloads.
And this is the question I cannot get out of my head: What if the West is learning how to spend more on AI while China is learning how to need less?
Think about that. China does not need to build something twice as good as OpenAI or Anthropic. It does not even need to be the best. It just needs to get close enough. Then make it cheap. Or make it open. Or both. That completely changes the conversation. Because most businesses do not actually need the smartest AI model ever created. They need something that does the job.
"Good Enough" Might Be the Most Dangerous Words in AI
This is where the whole thing turns upside down. Imagine you run a normal company and somebody offers you two options. The first is the best AI model in the world. Amazing. The second is nearly as good, costs far less, and you can run it with more control over your own data. Which one do you choose?
Sometimes the best one. But definitely not every time. Businesses have never bought technology just because it was number one. They buy what works. They buy what they can afford. They buy what solves the problem. That is why “good enough” matters so much. Because building the absolute best AI in the world costs billions. Building something good enough for millions of companies could be a completely different game. And if somebody gives that model away, or charges almost nothing for it, things get very uncomfortable very quickly.
Open technology does not need to beat the expensive option everywhere. It just needs to be good enough in enough places. Then people start asking why they are paying so much.
Then You Look at the Stock Market
This is where I become a bit more careful. I am not predicting a stock market crash. If I could predict the market that accurately, I would probably be somewhere much warmer checking my portfolio. Although knowing my luck, our Head of Marketing, Emily, would still find me and ask, “Abba, where’s the next article?”
But I do think there is something here we should take seriously. AI is no longer a small technology story. Some of the biggest companies in the world are tied into it. Then around them you have chip companies. Cloud companies. Data centres. Energy. Networking. Software. Entire supply chains. And markets that have become very used to the idea that AI spending keeps going up.
We got a glimpse of how sensitive those expectations can be this month. Following the recent calls for a slower pace of AI development, shares in major semiconductor companies fell, with Nvidia down 3%, AMD 4% and Intel 6% on 14 September. That does not prove an AI crash is coming. It does show how closely expectations around AI development are now connected to market sentiment.
That is fine while everybody keeps believing the next round of investment will create even more value. But imagine the chain reaction if that confidence changes. An AI company slows development. Investors start questioning how quickly the next breakthrough arrives. Spending on chips slows. Data centre plans slow. Infrastructure forecasts change. Technology valuations come under pressure. That does not stay inside one AI company. It spreads.
I am not saying AI is going to crash the entire stock market. Nobody knows that. What I am saying is that AI has become so connected to the value and future plans of some of the world’s biggest companies that a serious change in the economics would be felt far beyond the companies making the models.
That is why continuous innovation matters so much. They are not just racing for bragging rights. They are carrying enormous expectations with them. And they need the next model, and the model after that, to keep the story going.
Fear, Investment and Incentives Are Becoming Difficult to Separate
Some of the concerns around AI are absolutely real. Cyber attacks. Bad people using AI. Agents being given access to important systems. Models becoming more capable faster than we can properly understand them. We should take all of that seriously.
But there is also an uncomfortable dynamic here. The more transformative AI is believed to be, the more strategically important the companies building it become. That attracts investment. Investment accelerates development. Faster development creates new safety concerns. Those concerns, in turn, reinforce just how consequential the technology is believed to be.
I am not saying the risks are made up. They are not. But risk, investment, competition and expectations are now feeding into one another. And that makes this race even harder to slow down.
We Are Still Talking About Software
I also think we need some perspective. Today’s AI is amazing. I use it constantly and I still have moments where I think, how did it just do that? But we have not suddenly created a new species. We are still talking about software. Very clever software. Very powerful software. Software that can absolutely cause serious problems when given the wrong access. But software. Humans build it. Humans train it. Humans connect it to systems. Humans decide what an agent can access. Humans decide whether it can send an email, move data or take an action.
It does not need to be alive to be dangerous. Bad software with access to my bank account is still a problem. I do not care whether it has feelings while transferring my money. But sometimes the conversation about some future machine taking over the planet distracts us from the very strange thing already happening in front of us.
The biggest technology companies in the world are locked in a race. They need huge investment to keep running. Some of their leaders are now openly talking about slowing the pace because of where the technology may be heading. But they cannot slow down too far because China is still there. Open models are still there. The market is still there. Investors are still there. And nobody wants to be the first one left behind.
That Is What Actually Worries Me
Not ChatGPT waking up tomorrow morning angry with humans. Although given some of the things we ask it, I would understand. It is the race.
The companies building AI need continued investment. Continued investment needs continued belief. Continued belief needs progress. Progress needs new models. New models cost more money. And every new model helps make yesterday’s intelligence cheaper. At the same time China is trying to get closer using less.
That is the loop. And nobody knows where it breaks.
Maybe it does not. Maybe AI creates so much value that all this investment looks cheap in ten years. Maybe the companies spending billions today become the biggest companies the world has ever seen. That could happen.
But there is another outcome. Maybe they spend billions making intelligence so good, so cheap and so easy to access that eventually the real winner is not the company with the best model. It is the company that can deliver almost the same thing for almost nothing. And if that company comes from China, the problem is much bigger than OpenAI losing some customers. It becomes a technology problem. An economic problem. A geopolitical problem.
No Terminator required. Just competition. Much less exciting film. Far worse board meeting.
We keep asking whether AI will become powerful enough to kill us. I think there is a more immediate question.
What happens if AI becomes cheap enough to kill the companies that built it first?
This is the fifth article in a series from MTI’s Chief AI Officer exploring the forces shaping enterprise AI – from adoption, governance and security to agents, economics, data readiness and business value.
Next in the series: Is AI Creating Security Blind Spots?
What are AI agents actually doing inside your organisation, and why should boards be paying attention?
Continue the conversation
The AI landscape is changing quickly, but the questions facing organisations remain practical: where does AI create genuine value, what risks does it introduce, and how do you build the right foundations to use it responsibly? MTI helps organisations navigate those questions across AI readiness, governance, security and adoption.
About The Author
