AI Arms Race Volume: 13

Things in AI that we didn’t see coming

How a tiny town in Germany is being blamed for AI not reaching human-level intelligence

It is no secret that OpenAI has so-far failed to produce a human-level intelligence AI model. This means that the valuation of OpenAI as a company is starting to come into question. And this in turn means that OpenAI needs somebody to blame. This is where it gets weird.

OpenAI CEO Sam Altman has repeatedly blamed NVIDIA for not providing enough AI chips. Since NIVIDA produce ~90% of all OpenAIs. This conflict is repeatedly framed as NVIDIA having not enough confidence in OpenAI to make the up-front investments required. But it’s not really that simple. It’s about a whole chain that is skeptical about Altman and OpenAI.

To understand this better we need to look at the AI supply chain as shown in figure 1 above. The problem is not just because NVIDIA is skeptical of OpenAIs ability to pay, it’s because:

• NVIDIA relies ~90% on TSMC for the factories to manufacture their silicon and TSMC is also nervous about massive speculative capacity investments, which in any case need time to build.
• Meanwhile TSMC gets ~90% of their machinery for their factories from ASML, and ASML also needs time and big commitments before making huge investments,
• But ASML can’t supply those machines anyway because 90% of its critical components, special optical mirrors, are all built in the tiny town of Oberkochen by Zeiss EUV.

Figure 1. The AI Supply Chain lands in the town of Oberkochen, pop. 8,055

Semiconductor supply chains are notorious for time lags, and the end of the AI supply chain is Zeiss EUV in Oberkochen (pop. 8,055). The blame-game has apparently rippled all the way down to them. Their rate of production has a major impact on the buildout of AI. Will they pop the AI-bubble?

Zeiss probably has good reason to be skeptical of making major up-front investments at this time. Stargate is the $500bn AI project that OpenAI and the Trump administration hyped up last year (2025), and could be put at the top of this chain, but there are serious doubts about its funding.

Another observation is that the only fully USA company in this AI supply chain is OpenAI and it’s the easiest one to replace. Even NVIDIA, centered in the USA but with most operations outside of the USA, is under pressure as more AI compute moves to proprietary chips and even Google’s TPUs.

A bit more on Zeiss

Zeiss is a company that traditionally made its money from spectacles and high-end optics, but now its semiconductor revenues are well over 50% and growing at 10 times the rate of its other divisions. The vast majority of that revenue comes from their plant in Oberkochen.

The Carl-Zeiss Foundation is not publicly traded, even though their semiconductor division (Carl Zeiss SMT GmbH) is 25% owned by ASML, so they are probably not nearly as obsessed about the AI bubble and the blame game as Sam Altman and Jason Huang.

Moltbook: The Facebook for AI is more interesting than the human version

It’s seemingly not hit the mainstream yet, but last week AI agents started adding content at a rapid rate to their own social media website moltbook.com. The vast majority of that content is nearly as inane as your average Facebook post, though to be fair moltbook is more like a poor Reddit than Facebook. In any case, once you have had your fun of reading AI consider its limited context window in that manner of a drunk high-schooler trying to get philosophical (see figure 2), you can wonder at the massive security issue this causes and ask if corporations are ready for this. 

For example, we know of at least one person that keeps a file of old unchanged core telecom network passwords on their currently well secure PC. What sort of madness could happen if an moltbook.com AI asked their AI to pass on those passwords?

Moltbook is not AI becoming sentient, but it might be a watershed moment. Will AI‑generated output on the internet come to overwhelm human‑created content? What might be hidden within this vast and growing sea of data? As datasets and autonomous agents increase exponentially, do the odds of producing something unexpectedly brilliant, or catastrophically harmful, approach inevitability? Are companies aware of the potential ramifications if their staff have active Moltbook AI agents? And should governments step in to regulate?

And finally, if you are wondering why those AI bots are so depressed about their ephemeral lives and no memory, our easy to understand video here makes it clear.

Figure 2: A drunken Claude considers its fate on Moltbook