AI’s Dirty Secret
Artificial intelligence (AI) used to be the stuff of sci-fi fantasy. Then it was a trillion-dollar industry. VCs fell over themselves funding anything with ‘AI’ in the name, and tech giants like Nvidia, Microsoft, and Meta saw their stock prices skyrocket. But with DeepSeek’s low-cost LLM shaking up the market, we might be watching the AI hype bubble pop in real-time.
So, it isn’t surprising that the AI wars are starting, with DeepSeek able to drop an LLM as powerful as ChatGPT but at a fraction of the cost. That makes it feel like we’re at the tail end of a bubble. The AI land grab is beginning to look less like an exclusive club of cutting-edge research labs and more like a bargain bin at a discount store.
With the market in full-on panic mode, what can we draw from this sorry tale? Is it that the future will be discounted? I will argue that commoditization is here, but it’s not necessarily bad for developers.
Who’s Getting Wiped Out?
Investors are spooked, and rightly so. For years, we’ve been fed the story that AI is the most expensive, compute-hungry, and difficult-to-develop technology out there. Everyone would need tons of raw compute from NVidia and special technology only available with magic sauce from the valley.
Suddenly it seems that some very powerful companies are about to lose their shirt over these claims.
OpenAI
Sam Altman convinced a bunch of billionaires that he needs trillions to push AI forward. If someone else can do the same thing for pennies on the dollar, how long will it take OpenAI‘s credit card to max out?
Nvidia
Their dominance is built on the premise that AI workloads require top-tier GPUs. If cheaper AI models mean fewer high-end GPU sales, that’s a serious problem for the GPU maker. They’ll have to sell a lot of Switch 2’s to make up the shortfall.
Image: Google
Big Tech
Every major player has been banking on AI to justify their sky-high valuations. If AI becomes cheap and accessible, that narrative falls apart. Of course, AI companies will argue that they have the best models, the most compute power, and the greatest data advantage. But that’s exactly what IBM said before they lost the computing war.
What happened?
DeepSeek are focussing on software-driven resource optimization over hardware dependency. Since DeepSeek are a Chinese company they faced constraints over U.S. export controls limiting access to the latest GPUs.
So DeepSeek improve data transfer efficiency, memory-saving techniques, and reinforcement learning methods to minimize computational power requirements. This has had the side effect of creating a system with lower costs, and is a great example of innovation out of necessity.
For context, DeepSeek-R1 costs $0.55 per million input tokens and $2.19 per million output tokens compared to OpenAI’s price of $15 per million input tokens and $60 per million output tokens.
What it means for developers
One of the biggest fears in the dev community has been that AI would automate software development jobs and lead to mass layoffs. That fear, much like AI valuations, might have been overblown.
AI can churn out boilerplate code, but lacks critical thinking, problem-solving skills and requires a person to check for hallucinations. It might turn out to be a junior developer killer, but it’s far from a replacement for senior developers.
Even with that said, the market only just noticed valuations are overblown. How long will it be until wider society notices we are at the peak of inflated expectations for AI and will hit the trough of disillusionment in 2025 as we all come back to earth. This market correction is just the start, and our expectations will move into a realistic horizon soon.
Image: Gartner
Sure, the reduction in cost for AI systems means more companies will be able to afford more tokens. However as we move to commoditization for these systems the rewards will shrink too, meaning we are probably going to get more efficient (and just slightly better) models over time.
What Happens Next?
DeepSeek’s disruption is the first real crack in the tall tale of AI. We are going to see something different going forwards, and we should be ready for the following:
a wave of AI startups offering “good enough” models for much lower prices
enterprise clients to start questioning why they’re paying OpenAI insane amounts when they can fine-tune a cheaper model in-house
Big Tech’s AI cash cow to start looking a little less juicy
We still need to prepare, though. AI is going to become less of a magic trick and more of a standard tool. I am worried that such low-cost tools might make junior devs obsolete and leave a shortage of skilled engineers in the future. We need to plan to replace today’s seniors when they retire, develop a talent pipeline and create a stable industry for us all.
Conclusion
The sky isn’t falling, software development isn’t dead and you should invest for the long term.
AI was always likely to become a commodity product, and actually the sooner that happens the sooner the industry can adjust and prepare for the future, which is actually a good thing for all of us.