Ascent of AI and China Victory in AI Race?

Good morning. After a few weeks of silence, I have a lot to write about. It has been very active on several key fronts for AI. This post includes one “Deep Think” topic (as our fellow member Tim Hall likes to call them) plus a discussion of China’s advantages in the AI race.

The Ascent of AI

Many of you have read The Ascent of Man by Jacob Bronowski. It is a great book, and I highly recommend it. It makes you think about where we come from and how we became the humans we are today. AI is an emerging and comprehensive technology developing at breakneck speed, expanding beyond a mere tool for humanity to potentially replacing and exceeding humans in almost everything we do. I am not an AI doomsdayer; I think the opportunities and benefits will far outweigh any negative consequences over the long term.

That being said, a key question is: “How close are we to this technology really being like dealing with a human and a human brain?”

In September an important paper was published outlining a key development in the evolution of Large Language Models (LLMs): arxiv.org/pdf/2509.13351.

LLMs are great at creative tasks but are not yet capable of tasks that require precise logic. I find this interesting, as the word inference means “a conclusion reached on the basis of evidence and reasoning.” LLMs do not reason per se; perhaps this conventional word choice for when an LLM is in delivery mode is more aspirational than descriptive, given we are in the very earliest stages of AI development.

LLMs are probabilistic, guessing based on odds. The paper indicates an ability to teach them to think more like logical planners. The probabilistic behaviour (predictive and natural) is more aligned with human speech and communication, which works great for current chat like uses. However, grafting this probabilistic model on top of a more defined and logical structure could dramatically enhance the accuracy, forecasting, and operational “cleanliness” of Chat AI.

Developing more human like qualities

One recent development indicating these models are developing more human like qualities is the inclusion of the memory feature in the major LLMs. I use Claude from Anthropic more than any other model. They now allow a user to opt into a “memory” feature where the model searches through previous conversations to find and reference relevant information. One can also opt out and base all queries on an “incognito” basis, where the answers are independently developed.

In a recent conversation, Claude started its answer with “Given your CEO experience at Intrum, you’ll appreciate that….” The use of memory from other conversations and the personal reference to my experience certainly made me pay greater attention to the answer and made it feel more personal or human like, independent of whether the answer was fully correct. This is an incredibly important step in achieving greater accuracy and human like interaction with AI.

I could go on about the often discussed milestones of Artificial General Intelligence and Super Intelligence, where AI equals or exceeds human intelligence, but that would make this post incredibly long. There is still widespread debate about the definitions and timetables (something for another day). That said, these models are already better than humans at many tasks and are more human like than we would ever have expected as recently as last year. This progression will continue to the point where AI becomes pervasive in everything we do.

Will China Win the AI Race?

Jensen Huang recently shook up the world with the definitive statement that China will win the AI race. While Huang is certainly “talking his own book” (the world’s most valuable company will benefit if allowed to fully access China and other foreign markets), he makes some very compelling points on the advances to date and China’s future advantages.

There are three fundamental points relevant to this topic:

• China is not far behind today in the development of AI products.
• China has a tremendous advantage in power generation.
• China’s centralized economy will facilitate AI development.

The introduction of the DeepSeek model this past January shook up the world, as it was comparable to western models and supposedly developed at a fraction of the investment. I downloaded DeepSeek when it was introduced and it performed similarly to other LLMs I used at the time. Shortly thereafter I deleted the application (I didn’t want the CCP having a window into my iPhone), but the idea that China was far behind in AI was certainly debunked.

The power generation point is perhaps the greatest of China’s advantages, in AI and in geopolitical influence globally. As we have discussed before, power generation is critical to the advancement of AI: currently the world doesn’t have any operational 1GW+ data centers. However, it is very likely we will see fully operational 1GW data centers before mid 2026, likely part of 45 to 60GW of total compute across Meta, Microsoft, Amazon/AWS/Anthropic, OpenAI/Oracle, Google/DeepMind, and xAI.

China has the largest installed electricity generation capacity on the planet. In fact, it generates a full third of the world’s electricity, more than the United States, the EU, and India combined.

What is amazing is that, unlike the typical carbon intensive emerging market, China is not only leading the world in overall energy production but is also the undisputed clean energy leader globally. China has 887 gigawatts of renewable energy capacity (double that of the USA and EU combined) and can add almost 1 terawatt of renewable capacity each year, the equivalent of 300 large nuclear power plants. China is becoming a dominant superpower thanks to its dominance in energy production, and it has established this while being the cleanest producer to boot.

More specific to AI, an extrapolation through 2030 shows two interesting but potentially troubling developments:

• China’s lead will continue to widen.
• Projected AI power demand will quickly become the equivalent of total power production in the US.

Which leads to the final advantage: regulation and political will. Huang highlights that the state/federal divide of US politics may lead to “50 new regulations.” AI is such a critical industry for economic and military leadership globally that it warrants a federal, centralized regulator and policy in the US. Otherwise, the US will lose its innovation advantage and allow China’s centrally planned economy to surpass it.

In the end, AI may be the latest and most critical example of a phenomenon applicable to almost any industry over the last half century:

The US innovates; China replicates and Europe regulates.

In this case, the US should become more like China and much less like its European cousins.

I hope this post was interesting and informative. Have a great day. Andrés