Infrastructure, Infrastructure, Infrastructure

This piece isn't specifically focused on AI, which is why I didn't title it "AI Infrastructure." There won't be too much rationality or data analysis here; that "boring" work is already handled well enough by current AI—we shouldn't try to "take its job."

So, this post is just a reflection following some recent "observations + sudden events." While AI and rational analysis will still take up a significant portion, the "points of inspiration" and the rational analysis are two quite distinct parts.

Recently, an interview video on TikTok about Singapore being "boring" sparked a lot of debate. I absolutely agree with that assessment. As a highly modernized city-state with six million people and a very small land area, the result of high efficiency and clear boundaries is indeed "boring." Almost everything driven by clear rationality and goals can end up being boring—much like the day-to-day operations of the neural networks that form the basis of AI, and so on.

Another point of contention might be one of the saddest recent events: the Texas floods, where the death toll has now exceeded 100. Cultures and beliefs may differ, but whether it's a natural disaster or a man-made one, hearing an "official" response like "an act of God" always leaves a bitter taste in my mouth. At the very least, we can draw a preliminary conclusion: due to costs and other issues, warning infrastructure was outdated in the hardest-hit counties.

I connect these two events because they remind me of a book I read a few years ago called Scale. The author spent a lot of time discussing research on the scaling effects of large cities. The "1/4 power" relationship left a deep impression on me (to be accurate, I checked the book again: in biology, it's a 1/4 relationship—as an organism's weight grows, energy savings are about 25%; for a city or a company, this figure is about 15%).

Putting this rationally back into the context of the two events: large cities can have higher efficiency and lower costs, but they also bring more pressure and a greater sense of boredom. Small towns (I know many readers have US IPs and I know the concept of a "county" isn't directly related to size, so "small town" might be more relatable here) face higher costs and lower efficiency. Of course, if we focus on natural disasters, we can analyze it this way: because the risk of loss in a large city far exceeds that of a "small town" once a disaster occurs, there is more motivation to improve warning systems, emergency response mechanisms, and a whole range of "infrastructure."

To some extent, the larger the city, the higher the level of management required, the more rules and dogmatic settings are established, and higher efficiency is somewhat positively correlated with a "boring" lived experience.

People generally cannot escape balancing efficiency (returns), cost, and "boredom." From my own subjective perspective, I have another dimension of consideration: due to my personality and aging, I am willing to accept an increasing "proportion of boredom." This allows me to exist with lower costs and mental exhaustion, and it also makes me cherish the small percentage of "interesting" things even more.

Perhaps modern people simply need more and richer "infrastructure" to support themselves.

Yesterday, I had Gemini conduct research on the recent heatwaves and power grid pressure (in Europe and the US). One graph it returned and the mention of the "solar cliff" might not be new, but many might still find it interesting.

This coincided with a recent in-depth report from The New York Times (yes, it remains an important source of serious information for me, even though I admit its reporting bias is becoming increasingly severe) regarding the different paths China and the US are taking in energy infrastructure (especially "New Energy vs. Traditional Energy" following the passage of major legislation).

Briefly revisiting the book Scale gave me an additional benefit: it reinforced my view that "no matter how technology advances, energy demand will only increase as human society develops, rather than decrease."

I don't have a preference for green energy over fossil fuels, nor do I have an "obsession" with the relationship between "global warming" and "carbon emissions." But I do have an "obsession" with technological progress. Although the direction of technological progress is often quite "boring," "green electricity" incorporates a large number of new technologies and is constantly iterating. Traditional energy? I know little about it, but the pace of progress for such a long-matured technology must at least be slowing down significantly.

I am still an "old-timer" who prefers internal combustion engine cars far more than "electric vehicles." However, I also realize that whether it's autonomous driving or assisted driving, what more people like is the modern "intelligent cockpit" feel. These are driven by "electricity." An internal combustion engine can certainly be converted into "electrical energy" to drive them, but obviously, that system would be much more complex, and the R&D and operating costs might even be far higher than a "pure electric system."

This is still "infrastructure." But it's less about scale and more about "new" vs. "old." To use another example: for a new house versus a ten-year-old "near-new" house in the same location with similar quality and amenities, a rational person is likely willing to pay some premium for the new one; the difference is just a matter of how much. Generally, new is almost always better than old.

Finally, back to AI. The third "infrastructure" I want to write about is different from the first two, although it has both the dimension of scale (scaling law) and the dimension of "new vs. old" (old data centers and compute chips cannot meet the requirements of new models). Fundamentally, it might be a dimension of the physical world versus the digital world.

Of course, this topic has been discussed extensively in past articles, so there's no need for preamble or derivation. We can go straight to the conclusion: AI is not opening up an "intelligent world" by our human standards, but rather a completely different "digital world" and "computing world." For applications, the model is the "infrastructure"; for the model, computing power is the "infrastructure"; and for computing power, "silicon wafers" are the "infrastructure."

For the third one, I originally wanted to write "energy," but on second thought, that's a bit boring and not quite right: "silicon wafers" might be more appropriate. How many diodes can fit in the same area? How much can computing bandwidth be increased? How much energy consumption will be added? How much heat will be dissipated?

These constitute a complex trade-off. These factors, rather than any "algorithm," are the fundamental drivers of technological progress. If there is one more driver, it would be "data" (but that's actually quite abstract, isn't it?).

While the first two "infrastructures" (large cities, energy) are very long-term, AI infrastructure aligns more closely with investment time cycles: Has the effect of computing power increases brought by process improvements nearly peaked? If we throw more compute and memory (heat sources) into a unit volume, will heat dissipation really be solved on time? Can million-GPU data centers really be realized within the estimated timeframe?

These are the foundational factors with greater impact, rather than "training doesn't need as many cards anymore," "inference has lower card requirements," or "model optimization reduces compute demand."

The third "infrastructure" will certainly be figured out one by one; it's just that the speed of progress disturbs people's hearts.