Trending Articles
Trending Topics
Trending
This article is part of our Opinions section, where we invite industry professionals to share their views on the most pressing technology questions of our time.
For almost two years now, artificial intelligence has grabbed headlines almost daily. Often, these paint pictures of a future where AI has put millions out of work or rendered entire industries useless. Thousands have gotten fired across the globe as executives have decided that employees’ work was best done by chatbots.
However, as time has passed, this has turned out to be premature, to say the least. Recently, Swedish fintech Klarna was forced to rehire staff it had fired and replaced with AI. It turned out that AI just wasn’t up to the job. Klarna is just one example of companies scrambling to undo their mistakes; if that’s the case, is it worth investing this much into AI?
AI is expensive. Though most AI companies don’t like to advertise how much their products cost to develop, we can make some educated guesses thanks to the amounts being poured into companies like OpenAI and Anthropic by the likes of Microsoft and the rest of Big Tech, and it numbers in the billions of dollars.
Because of this, when China’s DeepSeek was launched in January 2025, it garnered widespread interest for costing only $6 million to build and train, according to its parent company. It showed the weakness of the Western approach to AI, where Chinese efficiency whittled the bloated Western budgets down to nothing.
Except that very little of that was true. While DeepSeek is undoubtedly a powerful AI that has some very interesting applications, it didn’t cost $6 million to develop, but $1.6 billion. Windows Central has more details on how this breaks down, but the upshot is that the quoted $6 million covers the cost of the hardware needed to run DeepSeek, not the R&D budget or any of the other costs associated with building a new technology from scratch.
In fact, DeepSeek may have cost more to develop than its counterparts: one publication speculates ChatGPT may have cost as much as $400 million to build, a quarter of DeepSeek’s price tag.
Creative accounting aside, DeepSeek proves that there are no shortcuts when developing AI. Building these models takes time, effort, and, above all, money. The example of Klarna, and many companies like it, show that many of AI’s promises aren’t quite as promising as they seem. This begs the question of how to move forward.
The fact is that right now, large AI deployments do not make sense for most companies as they are unlikely to deliver the returns most will hope for. Klarna did not need to fire 700 staff, and most organisations don’t need to rebuild their entire tech stack to implement solutions of unproven value.
Instead, I would advocate for a modular, problem-first adoption of AI. Companies should figure out first where their pain points are and then see if AI could help there. If so, great. If not, find a solution that would actually work somewhere else.
Right now, AI is making huge strides in areas like research, document analysis, outlining drafts of articles and presentations, and things of that nature. It’s also proven helpful to software developers, with several praising its usefulness as an assistant programmer that can take much of the scutwork away from its human counterparts.
None of this involves replacing humans; they’re just too valuable and, quite simply, much better at their job than any existing AI. The power of AI is its ability to let teams scale productivity without having to increase headcount right away; you can turn one worker into 1.1 or even 1.2 workers; you can’t replace entire departments.
When investing in AI, it pays to first think about what success would look like without it. Do you want to grow a specific metric in your company, like increasing sales? Or do you want to simply grow content without growing your marketing department?
Bringing your goal into focus should narrow down what you’ll be using AI for. There’s no shortage of tools out there, and not all of them will be of use to you. I recommend focusing on small, measurable wins. You want to cautiously experiment with tools, seeing how they impact your business, and try to measure the return on investment that they bring.
That said, AI can be expensive to operate, particularly for companies building and running the models themselves, like in the case of DeepSeek. For these companies and providers, the costs tied to compute, infrastructure, and energy can be substantial, especially at scale. In some instances, the operational burden may even outweigh the benefits if not carefully managed.
For most companies adopting AI, however, the picture looks different. Since many rely on SaaS-based solutions delivered through the cloud, the costs are far more manageable. That said, it’s still important to monitor usage and avoid unnecessary scale, even when working with third-party tools.
The fact is that what will cost millions today may be obsolete in a year or may end up costing only a fraction of today’s price. The key is to invest thoughtfully and at the right time, or end up spending even more money that may not see a return at all.
As the example of Klarna proves, moving too fast with AI adoption will cost you, maybe even more than being patient and seeing what actually works before jumping on the hype train. Not only did Klarna and companies like it lose face by having to rehire people, they also lost much of the trust of their employees, and the opportunity cost of losing and then rehiring experienced hands.
To truly capitalise on the AI revolution, businesses should pursue purposeful, resilient AI strategies. Focus on AI’s actual strengths in removing much of the tedious time spent on repetitive tasks, rather than the promises of tech moguls who claim their programs can replace humans. It’s just not there yet, and may never be.
