It’s kind of surreal to cast my mind back to late January, when Nvidia suffered the worst one day loss in stock market history. On 27 Jan it crashed 17% ($600 bn) on the appearance of DeepSeek - which suggested cutting edge AI tech could be developed without Nvidia’s most expensive chips.
But since then everything has appeared in good order. The Trump tariff chaos of early April had pretty much every stock scrambling, but since the lows of January and April, Nvidia is up 40%. I wrote about how and why Nvidia is so integral to the AI boom in February. It supplies 90% of the world’s GPUs, which are the chips generative AI needs.
But this week has seen a wobble. Any holders of leading chip stocks, and cryptocurrency, would have been smiling at the beginning of last week - but it didn’t take long to turn back. Palantir, a leading AI defence stock, has had a hammering five days. The Nasdaq has dipped -2.5%, and there’s talk it’s because of AI.
Signs of a bubble?
But this isn’t huge in the grand scheme of things - particularly given the giddy rises of the last three months. Palantir’s volatility has been crypto-like this year. But it’s still enough for experts and articles to suggest we’re in a bubble.
As we wrote last week, GPT-5 was a bit of a disappointment when it was released two weeks ago. Then a needle came on Tuesday with an MIT report into generative AI, which had some rather startling figures. The main headline was that 95% of AI investments produce no return. Then out of surveyed companies:
80% tried AI
40% deployed it
Only 20% reached pilot stage
Just 5% reached production with measurable gains
Half of projects failed outright.
Many employees prefer consumer tools like ChatGPT over corporate AI systems.
Those aren’t pretty stats. In my experience, enterprise is fundamentally challenged by AI in several ways:
Security and governance can cause confusion and inertia.
Scalable and useful AI applications are pretty hard to build, and require specialist knowledge. n8n agents are really quite easy compared to building enterprise ready solutions on Microsoft Azure AI Foundry.
The technology is moving too fast to really meet demand, and people can’t keep up.
It’s generally causing as much fear as excitement about its potential, and some are simply against adoption.
Added from my own experience, building gen AI ready no code solutions for smaller clients is difficult to find an ROI. Unless you are automating half a role, it’s hard to make the numbers work.
So I’m not surprised about the MIT study. While LinkedIn is alight with bullshit claims about how ‘this changes everything’ the reality remains really rather turgid.
That said, while I’m no Warren Buffet, I think this cycle has a while to run yet. Nvidia’s next earnings call is just next week. If it’s a big one (and it has been pretty consistent thus far), then these rough few days may soon be forgotten.
