Time for a catch-up!

It has not been as busy as some weeks, but as you’ll see quite a lot has been happening across the board, be it new LLM models, yet more investment or shifts in the regulatory environment.

However, for me the interesting snippet was the news about Michael Burry, the real person behind the story of the “Big Short” book and film. Burry famously bet against the sub-prime market in 2008, walking away with $100 million when it collapsed. More importantly, before he did, once he made his “bet” he walked away from the trading floor. Now he has effectively just done the same thing, closing his hedge fund, Scion Asset Management having just effectively bet against Palantir. Is he right? Has he just correctly called another bubble? Only time will tell, but his move has made a lot of people very nervous.

Anyway, let’s dive right in on all of the other news from last week:

Big Sharks vs. small fish

Oh, what a surprise!! OpenAI is again in the news. This time it was the release of the upgrade of ChatGPT to model 5.1. The transition was so smooth that I had to check to see whether or not the new model had actually installed. I have not had long enough to test it in depth, but what reminded me to look was the fact that the responses were noticeably quicker, so I’m now intrigued to test it some more.

However, the competition from China is again catching up with Moonshot releasing Kimi K2 Thinking, a new open-source model LLM that sets new records in reasoning, coding, and math tests, including ChatGPT 5.

Speaking of new models, Baidu announced two proprietary AI processors designed for training and inference. This was timed to coincide with them giving a major upgrade to their ERNIE large language model, which now has improved multimodal (text + image/video) capabilities. This is part of a longer game, aimed at reducing dependence on Nvidia and other foreign chips under export controls.

Infrastructure and investment continued to attract headlines this week, with Microsoft announcing plans to invest $10 billion in a major AI data hub in Portugal. This positions Portugal as a key European AI infrastructure node, leveraging subsea cables and green energy.

Then Anthropic also announced a $50 Billion plan to build custom data centres in Texas and NY, creating approximately 800 permanent jobs and 2,400 construction jobs.

However, not to be outdone, AI cloud provider Nebius signed a $3 Billion, five-year agreement with Meta to supply AI infrastructure, which itself was small compared to Nebius’ $17.4Billion contract with Microsoft in September. Nebius’ market value has quadrupled this year to $27.6 Billion, driven by demand for high-performance computing to train and deploy AI models. Nebius said demand was so strong that the contract size had to be capped by its available GPU capacity, underscoring global shortages in AI compute supply.

As I’ve mentioned before, this seems to be merely pushing the “can” down the road, as the big issue that no-one seems to be properly addressing is how will people power all of this AI infrastructure. Therefore, I was not surprised by a Goldman Sachs report that highlighted the fact US AI data centres already use about 6% of US electricity production and could reach 11% by 2030. This could potentially push grid spare capacity below safe levels if generation doesn’t keep up, at a time when China is rapidly expanding power capacity, including for AI workloads.

Some companies are also doing very well, with Cursor just closing a $2.3 billion funding round at a $29.3 billion valuation, up from $2.6 Billion earlier this year. That value may not be surprising when Nvidia's CEO publicly said 100% of their engineers use it, as does half the Fortune 500 companies.

What’s so good about it? Companies using Cursor have reported that their engineers can ship products 2 – 6 times quicker, which provides a massive market advantage and boost. That’s why OpenAI literally tried to buy the company twice but failed.

Legislation, policy and other news

The European Commission is said to be considering delaying key parts of the EU AI Act, primarily due to intense political pressure from the Trump administration. It did not help that 46 European companies including Airbus and Mercedes-Benz wrote a public letter asking for changes some months ago.

As a result, the Commission is supposedly thinking about offering a one-year grace period for companies that breach rules on high-risk AI and is considering delaying fines for transparency violations until August 2027. Given how few companies have actually started preparing themselves to meet the requirements, this may be a politically astute move. However, the lawmakers who drafted the original legislation are not happy about this, stating that any delays would create uncertainty and keep citizens exposed to very risks the law was designed to prevent.

That’s all for now from me at the end of yet another busy week … so stay informed, stay critical, and wherever possible - stay ahead.

Regards

Tom Carter