Artificial Intelligence and Investing

[EnSabahNur]

What is worse than been trapped in a bubble after it bursts? Been outside of it before it bursts.

This is a pertinent problem for most investors. There is a bubble, but you don't want to miss out on the potential gains.

https://investmentmoats.com/money/divers...ductivity/

It might be better, as Investmentmoats points out, to invest in the companies that will eventually make use of the new technology as they figure out how to harness it.

Phase 1: Production of Technology

• [b]Focus:[/b] This phase is dominated by the [b]creation and initial rollout[/b] of the new technology.
  
• [b]Outperformance:[/b] The [b]technology sector[/b] (or the equivalent “tech” companies of that era, such as utility companies in the 1910s) [b]drastically outperforms[/b].
  
• [b]Timing:[/b] This is roughly the [b]first half[/b] of the cycle (e.g., five or six years on the up).
  

Phase 2: Spreading and Application of Technology

• [b]Focus:[/b] The technology spreads, becoming a platform that unleashes new industries and raises efficiency across the entire economy.
  
• [b]Outperformance:[/b][b]Non-tech companies outperform[/b] the core technology sector. This is because:
  
  
  1. [b]Efficiency Gains:[/b] The technology is transformational, raising the efficiency, productivity, and new earnings growth for companies [b]outside the tech field[/b] (e.g., AI making hospitals or banks more efficient).
     
  2. [b]Reduced ROI for Incumbents:[/b] A [b]“massive amount of new entrance”[/b] (e.g., over 5,000 AI companies funded in the last four years) in the technology field increases competition. This influx of capital ultimately [b]reduces the ROI of the incumbents[/b] after the initial phase, even if the technology is transformative.
     

This doesn't come free of course.

This  dovetails with Weijian's point that some companies will gain and some will lose. So the job is to pick the potential winners and avoid the losers.

[weijian]

Although markets are still hot on AI development, we are in the year of AI application and it is really exciting to see what transpires ahead (provided one's employment is not under threat from AI). Is the 内卷 wave from China going to infect the AI landscape soon?

China's cheaper AI tokens a double-edged sword for Asian businesses

At the heart of that question is the cost of AI tokens - a little-known building block that determines how much companies pay when AI systems read, process and generate information.

For instance, models from Chinese companies such as MiniMax and Moonshot charge about US$2 to US$3 per million output tokens. In comparison, Google’s Gemini 3.5 Flash model charges about US$9, Anthropic’s Claude Sonnet 4.5 costs about US$15 and OpenAI’s GPT 5.5 model is priced at US$30, according to a Financial Times report and Google and OpenAI’s pricing documents.

China is emerging as a stronger challenger to the US in the AI race because of its lower token costs, experts told CNA.

Verma said the region may become a multi-model market: OpenAI, Anthropic and Gemini for premium reasoning; Qwen, DeepSeek, Kimi and MiniMax-style models for high-volume workflows; and local models for language, regulation and national-security needs.

In the end, Verma said, companies will judge AI not by the model it uses, but by the business outcome it delivers.  “Is the AI helping them meet that objective? … Am I making money or saving money?” he said, of the questions that companies will be looking at in deciding how to deploy AI models.

https://www.channelnewsasia.com/asia/chi...ls-6176061

[weijian]

Training the LLM consumes cash while inference is where the revenue is made. We are towards the end of the 4th year when ChatGPT was first introduced and inference increasingly gets more important as everyone is predicting "massive demand" with plenty of WIP (work in progress) to lower down costs so that it make sense.

The AI Inference Revolution Is Here

In 2026, inference—the use of trained models to produce code, write essays, or make images of ourselves as elves—has come to the forefront.

Although they might seem similar, AI training and AI inference are computationally different. These big moves from tech giants signal that in order to support the inference demand, we’re going to need a very different mix of hardware than experts may have expected even a couple of years ago.

This is where the autoregressive nature of the model works against inference speed. Predicting each token requires reading the entire model from memory, and that model consists of possibly tens to hundreds of gigabytes of parameters (the numbers representing what the model learned in training). Crucially, this is in addition to the memory required to store the KV cache. As a result, the movement of all this data through memory often requires more bandwidth than inference hardware has available. So at least some of the computing parts of a GPU sit idle as it waits for data. Researchers found that Nvidia H100 GPUs running open-source LLMs sit idle 50 to 80 percent of the time.

If AI inference remains as desirable as Kimball expects, the evolution is likely to follow the same trajectory as the CPU. The CPU didn’t improve along a single axis but instead across multiple fronts simultaneously. Once transistor scaling slowed, chip and system architecture innovations of all kinds proliferated. The list of individual innovations that led to today’s ubiquitous, powerful personal compute could fill dozens of books.

A few decades from now, the history of AI inference innovation will show similar depth.

https://spectrum.ieee.org/inference-hardware-revolution