When Will The AI Bubble Burst?

🌈 Abstract

The article discusses the concept of "AI PCs" and the potential financial implications of Microsoft's Copilot feature, which is built on the GPT-4 language model. The author expresses skepticism about the sustainability of Copilot and other AI-powered services that are being offered for free or at low cost, as the underlying costs of running these models can be quite high.

🙋 Q&A

[01] The concept of "AI PCs"

1. What is the author's understanding of an "AI PC"?

• According to the article, an "AI PC" refers to a computer that comes with a CPU, GPU, and NPU (Neural Processing Unit), as well as Microsoft's Copilot feature (not the coding one) and a dedicated Copilot key.
• The author notes that the NPU is likely more for features like webcam processing and image resolution enhancement, rather than providing many user-facing features.

2. What are the author's thoughts on the "AI PC" concept?

• The author is left with an "overwhelming sense of 'why?'" regarding the "AI PC" concept, as they are skeptical about the potential benefits and the financial sustainability of such a feature.

[02] Concerns about the cost of Copilot

1. What are the author's concerns about the cost of running Copilot?

• The author estimates that the cost of running Copilot, which uses the GPT-4 language model, could be around 2-5 cents per query on average, which they consider to be "huge" for Microsoft.
• The author explains the high costs associated with GPT-4 usage, including $10 per 1 million input tokens and $30 per 1 million output tokens, and the fact that Copilot needs to remember previous conversations, which adds to the cost.
• The author also notes that Copilot's ability to search the internet and process large web pages can further increase the costs.

2. How does the author's previous predictions about OpenAI's pricing model relate to their concerns about Copilot?

• The author had previously predicted that OpenAI's $20/month pricing model for their AI services would not be sustainable, as the actual costs could be much higher, around $30/month per user.
• The author sees similarities between their previous concerns about OpenAI's pricing and the potential costs of running Copilot, which they believe could be costing Microsoft upwards of a dollar per user per day.

[03] The author's perspective on the AI ecosystem

1. What is the author's view on the current state of the AI ecosystem?

• The author believes that the AI industry is experiencing a "mass hysteria," with companies adding AI features where they do not belong and losing significant amounts of money in the process.
• However, the author notes that people are still "eating it up," as evidenced by the continued demand for products like the Rabbit R1.
• The author points to signs of cracks in the AI ecosystem, such as the CEO shakeup at Stability AI and reports that 85% of AI startups could be bankrupt within 3 years.

2. How does the author view the financial sustainability of AI companies?

• The author believes that AI companies are "haemorrhaging money" and are surprised that they have managed to survive this long, as their financial situations appear to be unsustainable.
• The author suspects that these companies are trying to build market share by "burning money" on AI, but believes this approach is not sustainable in the long run.