The geographical distribution of the AI industry: Is it still winner-take-all?

Source: TOP Innovation Zone Research Institute

Author: Innovation Zone Research Group

With the advent of ChatGPT, we are witnessing a new round of artificial intelligence revolution. This revolution not only changes the relationship between humans and machines, technology and industry, virtuality and reality, but also brings profound challenges to the civilized order of human society.

To seize these opportunities, or meet potential challenges, depends not only on technical capabilities, but also on where you are. **

** **

Drought to death, flood to death

Recently, Forbes released a list of the top 50 artificial intelligence companies: ** The 43 American companies on the list come from only four states, of which California is particularly prominent, with a whopping 35 companies; ** New York (4), Texas states (2) and Massachusetts (1), where one company is operating fully remotely. None are located in the Rust Belt, the Midwest or the South.

We can also see this high geographic concentration in other AI lists. In another list of IVP Enterprise 55, there are 18 people from San Francisco, California↓

The Brookings Institution, the top think tank in the United States, recently published a detailed report. Through cluster analysis of metropolitan area data covering seven indicators of artificial intelligence research and commercialization in 384 metropolitan areas, it was found that artificial intelligence in the United States Smart activities are highly concentrated in the “superstar” San Francisco Bay Area (including SF metro and San Jose metro), and 13 “early adopters”

According to job posting data from January 2023 to May 2023 from Lightcast, 60% of new generative AI job postings are located in only the 15 metro areas mentioned above. Meanwhile, just six metropolitan areas (San Francisco, San Jose, New York, Los Angeles, Boston, and Seattle) accounted for nearly half (47%) of generative AI job postings in the United States over the past 10 months. **

** **

Network Effect

Globally, employment in the technology industry continues to grow, but if you look closely, you will find that the top industries (especially the technology industry) continue to be geographically concentrated rather than "diffused". **

Now that technological development has entered the era of "artificial intelligence and machine learning", especially in the early stages, ** more "needs" companies to be gathered together rather than dispersed - the gathering of ** companies not only provides companies with more resources and opportunities, and also strengthened the cooperation and competition between them.

Enrico Moretti, professor of economics at UCB, has studied urban economics for a long time. After a lot of research, he also came to a solid conclusion: **The high-tech industries in the United States are increasingly concentrated in a few Expensive coastal cities. **

In his book "High Wage Cities", Enrico Moretti mentioned a phenomenon - the Great Divergence. The core idea is:

** Those cities that excel in the innovation economy driven by entrepreneurs and technology have won talents and opportunities, and due to the Matthew effect, this gap is constantly widening, forming a "winner takes all" situation. **

However, there are only a few innovative cities that can become “winners”. They are lucky to have the "right" industries (clusters) with a solid human capital base, a well-educated workforce and a strong innovation ecosystem, and these cities are thriving, growing bigger, creating more and more Good jobs attract more highly skilled talents.

And once these cities became the "winners", they stayed on the table-

A report that examined 29 disruptive technologies over the past 20 years found that the distribution of these high-tech jobs remains highly concentrated—for example, the top 10 cities for computer science, semiconductors, and biochemistry each accounted for 50% of all inventors. Accounting for 70%, 79% and 59%, and maintaining the lead all year round.

One reason for this is the resilient network effect:

Take the Bay Area. During the epidemic, many migrant workers and entrepreneurs are planning a "Great Escape from Silicon Valley" - high housing prices, excruciating living costs, congestion, traffic jams, intolerable high crime rates, and the ever-growing homeless problem , and high taxes...

However, as the center of global technology and innovation, the Bay Area has already cultivated a strong and tenacious innovation network: There are a large number of technology companies, start-up companies, venture capital institutions and top research institutions-the value of the ecosystem comes from Interaction between multiple interdependent groups. When the elements of the ecosystem are more diverse, the interaction will be more complex.

**What you have to admit is that the Internet is the reason why an ecosystem is most difficult to copy, and it is also easiest to "winner-take-all" and never end. **

Of course, in the technical field of AI, there are still some unique challenges.

**First of all, it has high requirements for talents. **Theoretically, with enough expertise, anyone smart enough can make generative AI, whether they are in the Bay Area or in Shanghai.

**Secondly, it requires massive amounts of capital. **Training an artificial intelligence model requires a lot of computing power, which means a lot of money.

Silicon Valley is home to two of the world's top universities in AI research (Stanford and UC Berkeley) as well as many of the world's leading investors in AI R&D, including Alphabet, Facebook, Salesforce, and NVIDIA --** the "big company backed +Top Talent Work" recipe** has contributed to a large number of the most cited AI papers in 2022.

As more AI companies are established in California, a powerful network effect is forming here. This effect further strengthens the Bay Area’s dominance in the AI industry, making it the first choice for technology companies and talents.

** **

"AI common prosperity"

It seems that the AI industry will also become another highly concentrated, Bay Area-centric industry.

At this time, the U.S. government decided to implement "AI common prosperity" because it believed that unbalanced distribution may exacerbate social inequality and lead to economic stagnation in some areas.

The way is to **fund AI research, provide education and training, and formulate policies that are conducive to innovation and fair competition to help a wider range of regions and people benefit from the benefits of AI. **

Since 2020, NSF has established a distributed network of National Artificial Intelligence Institutes at universities across the country. To date, a total investment of nearly $500 million over five years has been launched in 19 cities, helping to build an AI talent pool, and has established connections with 37 states.

Image Source:

American academic circles are increasingly emphasizing the success of “locally based industrial policies”—after all, the space race launched by the U.S. government was a policy-oriented success story.

Coupled with the fact that manufacturing is now returning to the United States, such as the "Regional Technology and Innovation Center" plan included in last year's "Chip and Science Act", many people have realized:

**In order to revitalize the U.S. industrial base, more places need to have AI-based innovation. If AI becomes more concentrated and other regions are marginalized, the industrial base will also be negatively affected. **

So the 117th Congress proposed $80 billion in "place based" industrial policy measures, which include a number of investment plans that explicitly seek to improve the country's highly concentrated AI geography.

American version of industrial policy

** **

AI + Industry

Currently, the deployment of generative artificial intelligence is still in its early stages, but the speed is very fast: the deep integration of technology and industry is happening on a global scale. Generative AI is no longer just a tool for serving information content. It has become the "technical base" of many industries such as finance, medical care, and autonomous driving, and is expected to become the "technical infrastructure" of the future society. **

The opportunities and challenges for China in this regard are enormous.

At the recent 2023 China Computing Power Conference, many experts said: Compared with the general large model represented by ChatGPT, China's shortcomings in this area are quite obvious: First, it started late, with relatively little technology accumulation and R&D investment. ; In addition, it must be admitted that the training of general large models requires a large amount of data. Although China has a huge Internet user base,** there is still a certain gap compared with foreign technology giants in multi-lingual and multi-cultural data collection and processing. **

**However, the opportunities in China lie in industry megamodels. **

Back in 2017, Kevin Kelly predicted: The formula for the next 10,000 startups is that you are already doing something in an industry and then add artificial intelligence to it. Repeated a million times, the power is endless.

I predict that the formula for the next 10,000 startups is that you take something and you add AI to it. We’re going to repeat that by a million times, and it’s going to be really huge.

Industry large models are large deep learning models designed specifically for a specific vertical industry. Industry-specific knowledge and experience can be integrated into the model, thereby improving model quality and accuracy.

China has the most complete industrial chain and a huge physical industrial base in the world, covering almost all industry fields from agriculture, manufacturing to service industries. This provides a wealth of application scenarios and real data for the industry's large model, enabling the model to be optimized closer to actual business needs.

At the same time, China’s market is huge and has broad application space for various technologies and products. Industry large models have huge market potential in China and have broad application prospects in both the technological transformation of traditional industries and the innovative development of emerging industries.

With the transformation and upgrading of China's economy, various industries are facing the pressure of technological transformation and innovation. As a technology that can provide precise services for specific industries, industry large models can exactly meet this demand.

For example, with the automation and refinement of manufacturing, traditional manual quality inspection methods can no longer meet the needs of large-scale production lines. In order to improve production efficiency and product quality, many manufacturers have begun to use computer vision and machine learning technology to develop intelligent quality inspection models.

In fact, in China, systematic research and development capabilities covering theoretical methods and software and hardware technologies have been gradually established. For example, the Huawei Cloud Pangea Large Model has launched large models in the fields of mines, drug molecules, electricity, meteorology, and ocean waves, and has launched more than 1,000 innovative projects in various industries, helping the deep integration of artificial intelligence technology and industry applications.

**Based on the basic capabilities of general large models, industry large models have become an inevitable trend of technological development. China has a huge physical industrial base, rich industry data, an urgent need for deep integration of technology and industry, a huge market size and rapid technological iteration capabilities. **

This may also be the opportunity for Chinese industry in the field of artificial intelligence in the era of large models.