◎ Lin Jianzheng
Taipei Mayor Chiang Wanan proposed the idea of relocating Songshan Airport on October 8. He plans to divide approximately 300 hectares of land into 100 hectares of industrial clusters, 90 hectares of residential clusters and 110 hectares of central park. It will connect Beitou Shilin Technology Park to the north and Neihu Technology Park and Nangang Software Park to the east to create the "Songji AI Park". He also proposed a development vision of 12 trillion yuan in output value and 65,000 job opportunities.
Taipei Mayor Chiang Wanan proposed the idea of relocating Songshan Airport on October 8, planning to divide about 300 hectares of land into 100 hectares of industrial settlements, 90 hectares of living settlements and 110 hectares of central park. (File photo) The relocation of Songshan Airport involves major adjustments to the capital’s spatial structure. The airport surroundings have long been affected by aviation noise, building height restrictions and land use conditions. Once the airport is relocated, a large area of urban development space will be released, bringing a rare urban regeneration opportunity to Taipei City. However, when this idea is dubbed an "AI park," the first question that needs to be clarified is: What kind of AI park does Taipei need?
Two types of AI parks
With the rapid development of generative AI, "AI park" has become a popular policy term. However, there are at least two distinct types of spatial development in the AI industry: one is the AI innovation cluster centered on talent, R&D, and corporate activities, and the other is the AI computing power center centered on large data centers, computing equipment, and power facilities. There are huge differences between the two in land use, energy demand and employment population.
Please read on...
Most of the large-scale AI campuses built in the United States in recent years fall into the latter category. This type of park is mainly based on high-density servers, data centers, substation facilities and cooling systems, and some even plan to have their own power generation facilities. Its power demand often starts from 100 MW (100,000 kilowatts), and large-scale projects are developing towards 1 GW (1 million kilowatts) or even several GW. Site selection gives priority to power supply, transmission capacity, land acquisition and cooling conditions. The permanent manpower required during the operation phase is relatively limited, making it difficult to create a large number of job opportunities.
The other type of space is the AI innovation cluster, which is dominated by corporate headquarters, software research and development, algorithm development, IC design, application services and new ventures. Famous AI innovation clusters in the United States, including Silicon Valley, Boston Cambridge, and Pittsburgh AI Avenue, each have different development models. Silicon Valley relies on the academic research and talent cultivation of Stanford University and the University of California, Berkeley, and combines technology companies, venture capital and entrepreneurial culture to form an important global AI innovation center; Boston's Kendall Square relies on the Massachusetts Institute of Technology to form a high-density urban innovation district; Pittsburgh AI Avenue has Carnegie Mellon University as an important research and development core, gathering AI and robotics companies along an urban corridor of about one mile.
The area around Songshan Airport has long been affected by aviation noise, building height restrictions and land use conditions. Once the airport is relocated, a large area of urban development space will be released. (File photo) These cases show that the competitiveness of AI innovation clusters comes from the combination of talents, universities, research institutions, technology companies and venture capital networks. Its spatial form can be a high-density office and R&D building in a city or on the edge of a city. What Taiwan has often said in recent years is "industrial AI, AI industrialization", which exactly reflects the two major directions of the introduction of AI technology into existing industries, as well as the development of AI software, hardware and service industries.
Songji is more like an extension of internal medicine
Back to the concept of Songji AI Park, the configuration of 100 hectares of industrial clusters, 90 hectares of residential clusters and 110 hectares of central park shows that this is a land redevelopment plan with knowledge industry and urban life as the core. In particular, it is planned to create 65,000 job opportunities. Calculated based on 100 hectares of industrial land, there will be an average of about 650 job opportunities per hectare. This kind of employment density is relatively close to urban industrial clusters formed by high-rise office buildings, corporate headquarters, software research and development, and professional services. It is significantly different from the land use and employment characteristics of large data centers.
From the perspective of industrial location, Songshan Airport is close to Neihu Science and Technology Park and Nangang Software Park, and will be connected to Beishike in the future. These areas have gathered a large number of information services, electronic technology, software design, corporate headquarters and R&D institutions. Songji AI Park can integrate existing industrial networks and provide space needed for corporate headquarters, AI application research and development, software development, IC design and new start-ups, forming the new core of Taipei's technology industry corridor.
Judging from the development experience in the United States, Songji AI Park is particularly worth learning from the models of Kendall Square in Boston and AI Avenue in Pittsburgh. The former forms high-density innovation clusters through the high integration of universities, technology companies and urban space; the latter gradually establishes AI industry corridors through university research and development, enterprise settlement and urban renewal. If Songji AI Park can form a complementary division of labor with Internal Medicine, Nangang and Beishike, it will have the opportunity to develop an AI innovation cluster with Taipei characteristics.
The industrial positioning of Songji AI Park is closer to the extension of Neihu Technology Park and Nangang Software Park, as well as the reintegration of Taipei’s technology industry corridor. This development direction is in line with Taipei City’s economic structure, which is dominated by knowledge-intensive industries, and is of substantial value. However, if the existing science and technology industrial park is just rebranded as AI without putting forward a clear industrial positioning and innovation mechanism, the "AI park" will easily become a policy package for the election.
If Songji AI Park plans to develop a large-scale AI computing power center, power supply will become an unavoidable problem. The picture shows the xAI factory located in Memphis, Tennessee, USA. (AP file photo) Power bottleneck in the computing center
If Songji AI Park plans to develop a large-scale AI computing power center, power supply will become an unavoidable problem. Northern Taiwan has a dense population, concentrated industrial and commercial activities, and huge demand for electricity. However, the development of large-scale power sources is limited by land, environment, and social acceptance. In particular, the Northeast power supply region, where Taipei, Keelung and Yilan are located, has long relied on inter-regional power transmission support. As the old units of Concorde Power Plant are gradually retired, regional power supply and grid resilience deserve more attention.
Taking the 1 GW computing power park as an example, its power consumption scale is equivalent to 1 million kilowatts, which is close to the installation capacity of a large power plant. If it is estimated that the annual average load rate is 90%, the annual power consumption will be about 8 billion kilowatt hours. Such a huge new load involves power supply development, transmission capacity, substation facilities and power supply reliability. Large-scale AI computing centers in the United States are gradually moving to areas with sufficient energy supply, precisely because electricity has become an important location condition for the development of the AI industry.
Songshan Airport is located in the highly urbanized city center of Taipei. The land is expensive and the surrounding population is dense. The expansion of power facilities is also restricted by the urban environment. Using this precious land to build large-scale data centers has no obvious advantages in terms of land use efficiency, energy supply or employment benefits. Even if Huida chooses to settle in Beishike, its main functions are enterprise operations, research and development and talent gathering, which are different from the GW-level computing power center required for large-scale AI model training. Computing infrastructure that requires a large amount of electricity, land, and cooling facilities should be planned as a whole based on the national industrial layout and energy supply conditions.
How to calculate the output value of 12 trillion yuan?
The Songji AI Park’s proposed output value of 12 trillion yuan and 65,000 job opportunities is indeed attractive, but these two figures need to be explained more rigorously. Is 12 trillion yuan the annual output value, the cumulative output value, or the overall turnover of the enterprise? Are the 65,000 job opportunities directly employed in the park, or do they include surrounding industrial chains and indirect employment effects? The policy implications represented by different calculation methods vary greatly.
If 12 trillion yuan is the annual output value, calculated based on 65,000 job opportunities, the average output value for each employee is about 185 million yuan. This kind of figure may appear in the statistics of large technology companies with highly concentrated revenue and dispersed global production activities, but it may not reflect the actual added value created in the location of the park. The output values of AI computing centers, semiconductor manufacturing and software R&D service industries are calculated in different ways, and output values, investment amounts and employment effects cannot be estimated together. The Taipei City Government should provide a complete estimate of industrial composition, land use intensity, employment density and economic benefits so that society can review its rationality.
Huida will be stationed in the T17 and T18 bases of Beitou Shilin Science and Technology Park. (File photo) Urban regeneration is the core value
The opportunities brought by the relocation of Songshan Airport far exceed the reuse of 300 hectares of land. Once the airport is relocated, surrounding areas that have long been subject to aviation height restrictions may also have the opportunity to re-examine urban renewal and land use, further promoting the overall restructuring of transportation, residential supply, public facilities and industrial space. For Taipei City, where land is highly scarce, this is an extremely rare public resource.
The 110-hectare central park planned in the Songji AI Park can improve the urban heat island effect, increase public open space, and combine flood detention, urban ecology, and environmental education functions; the 90-hectare living settlement should provide affordable housing for technology industry employees and young families, reducing long-distance commuting; and the 100-hectare industrial cluster can establish an industrial ecosystem for AI R&D and application innovation through the cooperation of universities, research institutions, technology companies, and startup teams. The integrated planning of the three has the opportunity to form a new urban core that combines industrial development, residential life and ecological environment.
Conclusion
The 110-hectare central park planned in Songji AI Park can improve the urban heat island effect and increase public open space. The picture shows the Central Park built after the relocation of Taichung Shuinan Airport. (File photo, provided by Taichung City Government) The relocation of Songshan Airport involves the overall adjustment of national aviation policy, national defense security and transportation system, and requires joint evaluation by the central and local governments. However, from an urban development perspective, the 300 hectares of land released by the airport relocation and the surrounding urban renewal potential after the lifting of aviation height restrictions do provide Taipei City with a rare urban regeneration opportunity.
From the perspective of industrial characteristics, employment population and land allocation, Songji AI Park is closer to the extension of Neihu Science and Technology Park and Nangang Software Park. In the future, if we can combine corporate headquarters, AI research and development, innovative housing and large urban parks, and form an industrial division of labor with Beishike, we will have the opportunity to reshape the development pattern of Taipei's technology industry corridor.
The planning of Songji AI Park should first clarify whether to develop a talent-intensive AI innovation cluster or a power-intensive AI computing power center, and propose credible assessments of output value, employment and energy demand. The use value of the 300 hectares of land lies in creating Taipei’s industrial and urban development opportunities in the coming decades. The imagination of the AI industry needs to be based on realistic conditions, and the opportunities for urban regeneration are worth realizing through long-term planning.
The planning of Songji AI Park is closer to an extension of the existing Neihu Science and Technology Park and Nangang Software Park. (File photo) (The author is a retired professor at National Yang-Ming Chiao Tung University and vice chairman of the Taiwan Industrial Technology Promotion Association)
