Author: US Stock Investment Network

1. How has the popularity of AI fundamentally changed business logic?

Past = Ideas were cheap + Execution was extremely difficult.

Present = Ideas are abundant and cheap + Execution is incredibly easy (through AI).

This means → Only truly high-quality ideas deserve computational resources for execution. In simpler terms, execution capability is no longer a moat; the focus of capital and teams must shift to "how to choose the right ideas" and "how to sell the results produced by AI."

2. What is driving companies to frantically purchase AI tokens? What are the consequences of not keeping up?

Core driver = Extreme leverage of efficiency. For example, an individual can spend thousands of dollars to buy tokens and complete work that previously took a team of a hundred a year in just a few weeks (such as chip reverse engineering analysis, national grid modeling).

Result = Generation of "Phantom GDP," where actual output increases significantly while costs drop sharply, leading to distortion in traditional GDP statistics.

If you don't keep up → You will inevitably face a dimensionality reduction impact. If you do not consume more tokens to create and capture excess value, you will be relegated to the "permanent bottom tier of the AI era," quickly being commoditized and eliminated by faster-moving competitors.

3. What exactly is the bottleneck in the current supply of AI computing power?

On the surface = Nvidia GPUs are in short supply, and the lifespan of old cards has significantly extended (from 5 years to 7-8 years), driving up the gross margins of cloud vendors.

Deep bottleneck 1 → Memory (DRAM): Capacity expansion is extremely slow, and new absolute supply will not be ready until 2028, meaning memory prices will likely double or triple again.

Deep bottleneck 2 → CPU: The grading mechanisms for reinforcement learning environments and the execution of a large amount of AI-generated code require excessive CPU resources, completely depleting CPU supplies.

Deep bottleneck 3 → TSMC and edge materials: TSMC's capital expenditure might soar to $100 billion by 2028, while unassuming edge supply chains like copper foil, PCB glass fiber, etc., are already fully loaded, with the entire industry relying on "astronomical prepayments" to secure capacity.

4. What trends are emerging in the competitive landscape of large models and token economics?

Current situation = Anthropic is temporarily leading with Opus 4.7 and its internal "Mythos" model, even controlling risks by limiting the pace of releases, resulting in extremely high profit margins (>72%).

Power competition = Anthropic is limited by total computing power, while OpenAI is attempting to overtake by significantly financing and hoarding computing power (partnering with Microsoft, Oracle, etc.).

Core conclusion → Token demand far exceeds the carrying capacity of infrastructure. Even second and third-tier large model vendors will "run out of stock" due to insufficient top-tier computing power. Essentially, this means that as long as you can produce quality tokens, the market can absorb it all.

5. Why is there a prediction of large-scale anti-AI protests in the short term?

Cause = The massive business restructuring brought by AI will instill fear in the general public, who will tend to attribute long-standing societal issues to AI.

Catalyst = The public communication strategies of AI giants (like Sam Altman and Dario) are extremely poor + the frequent portrayal of the grand narrative "AI will change the world/replaces jobs" intensifies anxiety among ordinary people.

US Stock Investment Network’s suggestions → The industry must stop amplifying the terrifying capabilities of the future and instead showcase the positive empowering effects of AI in the present; otherwise, public anger will quickly be weaponized by politicians or influencers, leading to large-scale boycotts.

Dylan Patel Background:

Chief Analyst at the research firm SemiAnalysis, deeply dissects GPUs (especially Nvidia)

Analyzes the supply and demand of AI computing power (who lacks chips, who is hoarding)

Tracks the supply chain (ASML equipment → TSMC → cloud vendors → OpenAI etc.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。