**Introduction: From Chaos to Order**
In 2024, AI was in a state of primordial chaos. In 2025, the foundation of AI is gradually solidifying. At the beginning of 2024, the AI ecosystem was full of innovative ideas and potential power, making it a golden age for entrepreneurs. As of now, the AI ecosystem has gradually taken shape. In the large model competition, five “final competitors” have emerged. Chips are about to be introduced, data centers are under construction, and the supply chain is fully activated. Industries across the board are launching new AI projects. If 2024 was a year of bursting creativity, then 2025 will be a year to test whether these creative ideas are truly feasible.
**Three Predictions for 2025**
1. **LLM Suppliers Develop Their Own Unique Advantages**
– In 2024, the core goal of the large model competition was to reach the level of GPT-4. Five companies became the “final competitors”: Microsoft/OpenAI, Amazon/Anthropic, Google, Meta, and xAI. Other companies have withdrawn from the competition. To reach the level of GPT-4, these companies adopted similar strategies, and there was frequent talent flow.
– In 2025, these AI laboratories are cultivating their own unique core advantages:
– Google – Vertical Integration: Its advantage lies in the full-industry-chain integration. It has its own first-class chip TPU independently developed, builds its own data center, trains its own model, and has a strong internal research team.
– OpenAI – Brand Advantage: It has the strongest brand in the field of AI and has become the company with the highest revenue among large AI companies.
– Anthropic – Talent Advantage: A large number of research talents have joined, significantly enhancing its influence in the research field and becoming the top choice for AI scientists.
– xAI – Data Center Construction Advantage: It has deployed the Colossus cluster with 100,000 GPUs at a record speed, becoming an industry benchmark for data center expansion.
– Meta – Open Source Advantage: It has a strong distribution advantage on Instagram, WhatsApp, and Facebook and is fully committed to the open source strategy. The Llama model has a large number of loyal users.
However, the author believes that Sequoia’s “top five” prediction is too US-centric and optimistic. The global AI competition landscape is more complex. In the Chinese market, there are the “AI Six Little Dragons” and big company camps. In the European market, Mistral AI has emerged unexpectedly. The competition landscape in 2025 is likely to further shrink. The reasons include the continuous increase in computing power pressure, commercial monetization difficulties, and the widening technology gap. Specifically, major players may show different trends in 2025. It is expected that by the end of 2025, the number of truly competitive large model companies on a global scale may further reduce to 5-6.
2. **AI Search is Becoming a Killer Application – and Will Be Widely Popularized in 2025**
Since the advent of ChatGPT, the industry has been looking for the killer application scenario of AI. The author believes that AI search will become one of the most widely popularized applications in 2025. For example, Perplexity is developing rapidly, and OpenAI has launched ChatGPT Search. AI search is a revolutionary reshaping of traditional internet search and can bring significant productivity improvements. Different text responses have different characteristics. Entrepreneurs can create unique product experiences for specific user groups around these characteristics. AI search is likely to differentiate in the consumer market and enterprise market.
However, the author has reservations about Sequoia’s judgment of regarding AI search as the next killer application. In 2024, AI programming has become the first super app, and Cursor and Windsurf are popular among developers. The field of AI images and videos showed explosive growth in 2024, and Chinese enterprises performed brightly. The rise of multimodal creative tools may become the real killer application in 2025 instead of a single search function.
3. **Return on Investment Remains a Problem, but Capital Expenditure Will Stabilize in 2025**
At the beginning of 2024, big tech companies were worried that AI might threaten their oligopoly status in the cloud computing business and actively invested. In 2025, the situation has changed. Big tech companies have firmly grasped the leadership of the AI revolution. It is expected that capital expenditure in the AI field will stabilize in 2025. The oligopolistic market structure will also gradually form. The cost of AI computing is expected to continue to drop significantly, which is good news
