Stop idolizing large models. Andrew Ng personally shares AI secrets: Small models + edge computing = the key to wealth

[Introduction] Andrew Ng pointed out that in the current fierce competition of large models, it doesn't matter who the winner is. The key is that those who can build trustworthy AI applications will be the ones who truly shape the future and, incidentally, become the next ones to achieve financial freedom through AI.

The real opportunity for AI entrepreneurship lies not in "bigger" but in "smarter"!

While tech giants are competing in the multi - billion - dollar race for model parameters, the real entrepreneurial opportunities are quietly emerging at the other end—the era of agents is opening up.

This is the "SME revolution" in the AI world: it doesn't pursue the myth of general intelligence but uses thousands of focused, reliable, and business - savvy agents to tackle those unassuming yet extremely valuable real - world problems.

AI Entrepreneurship: The Opportunity Lies in Agents!

Is it true that the bigger the model, the better? The more the investment, the better? Is the competition in the AI field really that simple?

If so, then AI is not suitable for entrepreneurship.

However, apart from building large models, a greater opportunity lies in using tools to solve practical problems in various industries, which is the so - called agent.

Currently, the market for AI agents is approximately $5.1 billion and is expected to increase to $69.1 billion by 2032, a nearly 17 - fold increase in 7 years.

An agent breaks down tasks into a series of smaller, more manageable subtasks. It formulates strategic plans, executes them step by step, and makes adjustments during the process, simulating advanced human reasoning.

An agent uses multiple sets of prompts + knowledge bases and links with external tools such as search engines and code execution.

An agent can also self - criticize the results output by large models. Through self - questioning, self - answering, and iterative self - correction of questions like "Is this the best method? Can it be more accurate? What have I missed?", the quality and reliability of the output are greatly improved.

Entrepreneurs planning to enter the AI field must understand that the advantage of agents lies in "specialization".

Large companies try to use a single model to cover everything, while in a multi - agent system, each agent is only responsible for a part of the work.

How to assign different roles to agents so that a group of small, inexpensive models can outperform expensive standard models like GPT - 5 when working together, thereby reducing costs and increasing efficiency in specific fields—

This is the incremental value that entrepreneurs can bring.

Don't Pursue AGI, Use AI to Solve Daily Problems

The tech community is enthusiastic about pursuing Artificial General Intelligence (AGI).

This pursuit has consumed billions of dollars in research funds and sparked countless media hypes. Andrew Ng's advice is to ignore it.

For entrepreneurs who want to enter the AI field, it's best to hope that AGI won't appear because the emergence of AGI would mean that all companies would fire all knowledge workers.

If that day really comes, there will naturally be no room for entrepreneurs to survive.

The reason why entrepreneurs are needed in the current AI field is that there are a large number of "boring", unglamorous but highly commercially valuable problems in every industry today.

Start - ups should focus on measurable results, such as automating document processing, optimizing energy consumption, improving the accuracy of medical diagnosis, and streamlining manufacturing processes.

These applications won't make headlines, but they create billions of dollars in value by improving efficiency, reducing errors, and cutting costs. This is where the real money is made in the field of artificial intelligence right now.

Entrepreneurs should choose an industry they are familiar with, critically examine each step of the recurring work process, and see which repetitive, data - intensive tasks consume the most manpower? Which processes are most error - prone? These are your targets.

Build or deploy narrow AI solutions to solve these specific, high - cost problems. The return on investment is direct, measurable, and huge.

In this process, entrepreneurs don't have to develop models from scratch. Instead, they should adopt the "take - what - you - need" approach, actively explore and experiment with leading open - source models that cost only one - tenth of proprietary models, regardless of their origin. By building these cost - effective and rapidly improving foundations, start - ups can operate at a lower burn rate, launch products faster, and outperform giants stuck with expensive proprietary models.

Pay Attention to Edge Computing of Small Models

Currently, most large models run in the cloud.

However, Andrew Ng pointed out that as the performance of small models increases and the price of hardware decreases, we will see more small - sized models and more models running locally in the future.

The total market for small models will grow from $930 million in 2022 to $5.45 billion in 2032; the market for edge computing is expected to reach $378 billion in 2028.

Edge computing makes it possible to develop more applications involving private data.

For example, entrepreneurs can develop a mobile app that analyzes users' voice and usage data in real - time to detect early signs of diseases such as depression and Alzheimer's. All data is securely stored on local devices, allowing users to use it with peace of mind.

In the manufacturing industry, entrepreneurs can add multi - modal models to cameras, enabling users to input prompts to instantly identify tiny defects they care about.

In the retail industry, entrepreneurs can develop rentable shopping guide robots to replace salespeople. Customized robots can observe users' behavior, select potential users to send free gifts, and store private user data that merchants care about locally.

Edge computing + small models are feasible because the existing infrastructure is already in place. Billions of smartphones and Internet of Things devices already exist as computing devices.

These devices have zero latency, zero cloud costs, and ultimate privacy protection.

Entrepreneurs only need to take advantage of these three advantages to build small but beautiful applications by optimizing the performance of small models in specific fields.

As models become smaller and more powerful at the same time.

For entrepreneurs in the AI field, the moat shouldn't be the technology itself.

Currently, anyone can download and deploy open - source models. The real irreplaceable aspect of start - up companies is to provide users with a sense of trust.

Entrepreneurs don't need to educate customers about how advanced their technology is. Instead, they need to find ways to convince customers to trust the AI they develop.

Meanwhile, regulatory agencies no longer accept black - box systems; they require interpretable and transparent models.

When models themselves are no longer scarce, the real competitive advantage comes from trustworthy AI applications.

The companies that are growing the fastest are not just deploying more models; they are deploying models that are verified, monitored, and properly managed. Only a commitment to building reliable and transparent systems can maintain a leading position in the long run.

Pay Attention to Dual - Use Application Scenarios

In February 2025, Andrew Ng made a controversial statement: "I'm glad Google has changed its stance on AI weapons."

However, such remarks are not unfounded. The application of AI in the military field has become a crucial, inevitable, and innovative explosive area.

Although autonomous weapons pose ethical risks, the reality is that the "military AI gold rush" has begun. This is not limited to life - and - death weapons.

One of Andrew Ng's portfolio companies is developing autonomous drones that save lives by conducting rapid reconnaissance in disaster areas and providing safe logistics support for remote troops. This can be regarded as a dual - use application scenario.

In addition, AI - driven intelligent security, such as threat detection, predictive maintenance of complex hardware, military simulation and training, and unprecedented large - scale logistics optimization. Investments from governments around the world are pouring in, creating a mature ecosystem for start - ups that can provide powerful, reliable, and cutting - edge AI solutions.

Entrepreneurs should investigate "dual - use" technologies.

An AI system designed to optimize commercial supply chains can be adapted to military logistics. A computer vision model used for industrial inspections can be repurposed for equipment maintenance on naval vessels.

The key is to build basic technologies that are applicable to both the civilian and defense sectors, thus opening up a large, well - funded market. However, many people are too timid to enter.

Foresight Means Wealth, Insight Means Opportunity

The above points are some practical suggestions Andrew Ng gave to AI entrepreneurs. These guidelines are based on current technology and are actionable.

Following these suggestions, starting an AI business and surviving is not an unattainable dream but a present reality.

If you have foresight and can perceive the development trends of the world, the future is within reach.

The future AI wealth doesn't belong to those who own the largest GPU clusters but to those who know how to use the smallest models to solve the most specific problems.

Beyond the shadow of giants lies the golden frontier for entrepreneurs—they make intelligence truly practical and make trust the new moat.

AI is no longer a war of technology but a race of execution and insight. The winners will be those dreamers of agents who can stand firm "under the waves".

References:  

https://www.reddit.com/r/AgentsOfAI/comments/1nogqpy/andrew_ng_the_ai_arms_race_is_over_agentic_ai/ 

https://aiquantumcomputing.substack.com/p/the-ai-oracle-has-spoken-andrew-ngs 

This article is from the WeChat official account "New Intelligence Yuan". Editor: Peter Dong. Published by 36Kr with authorization.