🟢 The AI bubble discussion...
This week, let me put a quick nail in the 'bubble' discussion about AI. Being quite the cautious sceptic, I am about tech and know the usual hype curve perfectly well, so I don't think AI is going away anytime soon...
Let me predict where AI is going.
AI is not going anywhere.
And the so-called AI bubble is not going to explode anytime soon.
What the AI market can (will probably) do is the reverse of a soft landing. Let's call it a soft deflating. Meaning that enthusiasm around AI and overspending from corporations will stop, or at least come to a plateau, soon enough, but not in a big crash. It will be more of a "let's pause, reconsider, and reorient" moment in time.
This will most probably happen when everyone realizes (because they already know it) but accepts that LLMs, which are a particular flavor of AI, are pretty limited and will not be able to overcome a proper reliability threshold. As LLMs stubbornly keep making 20 to 30% of mistakes, alternative truths, hallucinations, or plain and enthusiastic inventions of facts, companies will have to remove them from manufacturing, supply chains, or financial processes.
While using them for creative endeavors, marketing, or communication will lead to proper ROIs, the flaws they will introduce elsewhere will stop the current AI enthusiasm.
That being said, all the large US tech companies involved in the AI economy are circularly investing in themselves, building a massive infrastructure advantage over everyone else, meaning whatever happens next, they will be able to pivot and get to the next phase, pushing the next AI flavor of the month anytime they need.
Not to mention that the U.S. government has decided that this will be the only way to regain tech leadership over China and they funneled major public funds into the AI economy — including a $8.9 billion stake in Intel, $200 million defense AI contracts each with OpenAI, Anthropic, Google, and xAI, and a multi-billion-dollar federal cloud and AI services deal with Microsoft.
This huge underlying infrastructure is not going anywhere for the next twenty years; it's just a matter of remembering that, despite all the hype, LLMs are not the only possible way of delivering AI productivity and breakthroughs.
We already know one of the possible pivots: new versions and competitors of DeepMind. You remember? The little thing that cracked the way a protein sequence folds in 3D, opening the door to quite a few pharmaceutical breakthroughs and, as a whole, how a large chunk of biotech R&D now works.
This flavor of AI — not LLMs but expert systems — will also require more compute, more data, more infrastructure, and more energy. It just doesn't sell ads on TikTok but solves hard questions, such as financial ones. That's the thing you need to remember and maybe take into consideration in your own business, instead of blindly following Microsoft or OpenAI...
Just like Pet.com was a bump in the road to the digitization of the economy in the early 2000s, LLMs might disappear, but AI is not going anywhere.
Philippe MEDA
Philippe MEDA
Philippe MEDA
