“Is AI the most important thing that has happened in a long time? In my opinion, yes. But are we at a stage where investors, in general, are overly excited about AI? In my opinion, also yes.”
This statement comes from Sam Altman, CEO of OpenAI, in an interview with The Verge in August 2025. And it perfectly sums up the current mood: Artificial Intelligence is revolutionary, but the hype may be pushing us toward a bubble.
That feeling became even stronger after the launch of GPT-5, which was promised to be a leap to “PhD level,” but was instead met with a wave of criticism from users who considered it less useful than GPT-4o. If you also felt disappointed, you’re not alone.
But does this disappointment mark the beginning of an AI bubble? And if so, is that necessarily bad?
Why Did GPT-5 Disappoint?
Unlike previous versions, GPT-5 works as a set of specialized models connected through a routing system. In theory, each user query would be directed to the most suitable model. In practice, this router failed, and the overall experience didn’t live up to expectations.
This technical error can be fixed, but the deeper issue is structural. The development of large language models (LLMs) is reaching a ceiling, with some obstacles that are hard to overcome:
- Exploding costs: training next-generation models requires thousands of cutting-edge GPUs, enormous energy consumption, and infrastructure that only big tech companies can afford.
- Training data limits: most high-quality public content has already been used. The risk now is “contamination,” with models being trained on their own outputs.
- Alignment and safety challenges: the more sophisticated the models, the harder it is to control biases, hallucinations, and misuse.
- Diminishing returns: the jump from GPT-2 to GPT-3 was huge. From GPT-4 to GPT-5, costs skyrocketed while the improvements were marginal.
Businesses Are Also Frustrated with AI
The excitement is not just among consumers. In the corporate world, billions of dollars are being poured into generative AI projects. But results are often disappointing.
A report from the MIT Media Lab revealed that 95% of corporate AI pilots fail to deliver returns. In other words: only 5% make it to production with measurable impact.
Even so, the market remains overheated. CB Insights data shows that 1 out of every 2 venture capital dollars now goes to AI startups. In the first half of 2025 alone, funding has already surpassed the entire volume of 2024. OpenAI, even while operating at a loss, is preparing to raise another $6 billion, reaching a $500 billion valuation.
This disconnect between real results and financial expectations is a classic sign of a bubble.
History Has Already Taught Us: The Dot-Com Bubble
If you remember the early 2000s, this story sounds familiar. The dot-com bubble saw companies like Pets.com and eToys valued in the billions without sustainable business models—until they collapsed just a few years later.
On the other hand, giants like Amazon, Google, and eBay survived and thrived. The capital poured into the sector helped build the infrastructure of Web 2.0: servers, submarine cables, and internet backbones.
A striking example is Webvan (1999–2001), which burned $1 billion promising online grocery delivery but collapsed due to a lack of demand. Years later, Instacart brought back the same idea, but with a leaner model enabled by smartphones and digital payments—and today is worth tens of billions.
In other words, bubbles can destroy value in the short term, but also prepare the ground for the next revolution.
What Can a Potential AI Bubble Teach Us?
If we are indeed heading toward an artificial intelligence bubble, it can have a positive effect—as long as we learn from it. Some lessons are already clear:
- Separate hype from real value: simply adding “AI” to a sales pitch is not enough; real impact is what matters.
- Build infrastructure: data centers, chips, and connectivity may be the true legacy of this phase.
- Requalify talent: engineers, designers, and leaders who fail today may be the founders of the next “Google of AI.”
- Learn from failure: growing at all costs doesn’t work. Sustainable business models are the ones that survive.
- Leverage human differentiation: in a world where everyone uses the same AI tools, what sets professionals and companies apart is the creative, ethical, and empathetic way they apply them.
Questions Leaders Should Be Asking Now
Instead of chasing quick answers, perhaps the most important step is to ask the right questions:
- How can we implement transparently a technology that we ourselves don’t fully understand?
- If we don’t understand how it works, how can we hold AI accountable for its decisions?
- How can we compensate for reduced mental engagement when tasks are delegated to AI?
- How can we requalify people in a context where new job openings are declining due to automation?
- How can we preserve our human differentiation in a world standardized by the same tools?
Conclusion: The Bubble Is Not the End, It’s the Beginning
If an artificial intelligence bubble does burst, we should not see it as a tragedy. Just like the internet bubble, the flood of capital and attention could lay the foundation for the next wave of innovation.
The challenge is not to fall into blind euphoria, but to use this moment to reflect: what is AI’s real role in our companies, our careers, and our society?
Perhaps the bubble is exactly the wake-up call we needed—not to find the “right answers,” but to start asking the right questions.
