The Continuity of the AI Boom — The "Fourth Industrial Revolution" Has Just Begun

From ChatGPT to Sora to World Models, Computing Power Demand Doubles Every 6 Months

Some worry that the AI boom will burst like the internet bubble of 2000. But this analogy is fundamentally flawed: in 2000, most .com companies were unprofitable; today, NVIDIA is one of the fastest-growing profitable giants worldwide. More importantly, AI is directly boosting productivity across various industries.

In terms of model scale, GPT-5 is expected to be released by the end of the year, with parameter counts surpassing 100 trillion, and training compute requirements over 10 times that of GPT-4. Google Gemini Ultra 2, Anthropic Claude 4, and xAI Grok 3 are all accelerating their iterations. Additionally, video generation models (Sora 2, Runway Gen-4), world models (World Labs), and embodied intelligence (robots) are experiencing explosive growth in inference compute demands.

NVIDIA's proposed "Compute Power Hyper-Growth Curve" shows that over the next five years, AI training compute demand will grow 3–5 times annually, and inference compute demand will grow 5–8 times annually. Even considering algorithm efficiency improvements, hardware shipments will still grow at a rate of 40–60% per year. Blackwell Ultra and the subsequent Rubin platform (2027) will further widen the gap with competitors.

Returning to the May price forecast. At that point in May, no new models were released, and no earnings reports were available, so stock prices wouldn't sharply surge in a straight line. However, sustained buying pressure (from ETF funds, passive funds, retail FOMO) would gradually push prices to new highs. I believe that by the end of May, NVIDIA's stock price would likely be between $235 and $245, representing an approximate 3–8% increase from current levels.

#Polymarket每日熱點