I was thinking about how the technology industry has gone through multiple cycles of innovations, periods of consolidation, lots of standardizations and eventually providing a rock solid base of technology that just works.

This brings me to AI.

The next AI winter has started. It might not be a hard winter, it could be mild but it certainly has started. It could also be a long one.

The release of FauxOpenAI’s version 5 model will undoubtedly be noted as the tipping point – August 2025. It was an interesting three year run – from November 2022 till August 2025. They were on a trajectory that, I must admit, even I was intrigued by what was happening.

But since early 2024, it became very clear to me that there really is no meaningful future in genAI as it is touted today.

AI winters are not new to me. My first foray into AI was in 1988, in grad school, and did not realize that it was during an AI winter as well (perhaps the 2nd). Even though I ditched it to do VLSI chip design and later open source software, there were some interesting tech that came out of it. Expert systems and business intelligence systems to name two.

This time around, there is indeed *some* intrinsic value in the genAI tools and techniques especially when it is used as narrow AI.

When the dot com bubble burst about two dozen years ago, lots of good stuff was left behind. The global Internet infrastructure (undersea fibre cables for example), open source software ecosystems, mobile and broadband Internet, and by about 2007, touch-based mobile phones etc. We are all beneficiaries of all of those.

The 2010s saw the idea of “big data” – volume, velocity, veracity – yada, yada, yada. The pay-to-play spin doctors like Gartner, IDC, Forrester all touted how big data will be “something”. For all of their pay-to-play predictions, it was just as accurate as predictive AI is of predicting the future. Notably the “cloud” aka someone else’s computer came out of that decade.

As this next AI winter starts, there will have some good tech and ideas to work with: enormous corpus of open source tools and techniques, Cache AG, Knowledge AG, Retrieval AG systems, AI ethics and governance principles and practises, cyber security informed AI solutions (MCP, Agentic systems with full-on authentication and access controls) and RISC-V based CPUs and GPUs, eclipsing the nVidias of today, with, sadly, Intel relegated to the third or fourth position after AMD and ARM.

Let FauxOpenAI spend the billions of VC$ with commensurate revenues that are a tiny fraction of the billions. There will be new(er) algorithms which will bring forward clever(er) ways to do model training, inference, especially as open source models (IBM Granite, Swiss LLM etc etc etc) become robust and competent. These will fuel sovereign AI and sovereign cloud while keeping at bay “big tech”‘s centralisation world view.

No doubt, the cheer leaders of FauxOpenAI will be doing their thing, but the chill is already in the air.

Enjoy the downtime. We will all be fine.

[Update: I would like to invite a review of the 4 videos in this post, especially the 3rd one on a brief history of AI.]