Young Liu: AIDC Capital Expenditure and Power Consumption Are Massive, Driven by Four Key Demand Groups

Central News Agency Report

(CNA reporter Chung Jung-feng, Taipei, 18th) Young Liu, chairman of the Taiwan Electrical and Electronic Manufacturers' Association (TEEMA) and chairman of Foxconn, stated today that artificial intelligence data centers (AIDC) are highly capital-intensive and will consume massive amounts of electricity, resulting in high power expenditure costs. Hardware depreciation expenses will be even more expensive. He predicted that four categories—model vendors, cloud service providers, governments, and enterprises—will be the primary user groups of AIDC.

Liu was invited to deliver a speech at the morning meeting of the Industrial Development Association, where he shared insights on the AI industry, robotics, and the progress of TEEMA Science Park's development.

Regarding the scale and power consumption of the artificial intelligence data center (AIDC) industry, Liu cited media reports indicating that by 2030, the AIDC industry could reach $1.6 trillion in scale and consume 174GW of electricity. To meet AIDC power demands, annual electricity generation must increase by 18GW. Taiwan's total electricity consumption, driven by AIDC applications, is approaching 40GW. By 2030, an additional 106GW of power is expected—approximately 2.5 times Taiwan's current total electricity consumption.

Liu cited market estimates on the cost of NVIDIA's Vera Rubin rack. Establishing a 1GW AIDC centered on Vera Rubin requires a capital expenditure of $47 billion, with 3,557 racks. The deployment cost of a single Vera Rubin NVL72 rack is $9.1 million. A 1GW AI data center incurs an annual power expenditure of $1.3 billion, with hardware depreciation costs six times higher than power costs.

Liu pointed out that AIDC power consumption is enormous and capital intensity is extremely high—completely different from Foxconn Group's previous focus on personal computers, smartphones, and electric vehicles in terms of power usage and capital expenditure density.

Observing AIDC demand, Liu believes it can be divided into four main groups: model vendors, cloud service providers, governments, and enterprises. Among them, model vendors have a clear business model and are currently the group with the clearest use cases and highest demand. Cloud service providers are also the largest current users and most clearly defined buyers.

Regarding enterprise applications of AIDC, Liu stated that related demand lies in building a new organizational model called 'AI Native,' where AI serves as the core execution entity, and humans are responsible for goal setting and governance oversight.

Liu analyzed that an AI Native organization has seven key points: enabling data to be used by AI, redesigning processes, possessing agent (Agent) functionality; not allowing AI to make final decisions, ensuring decision-makers no longer make blind decisions; reorganizing organizational roles, upgrading talent capabilities, and establishing a controllable, traceable, and automated governance mechanism. (Editor: Chang Chun-mao) 1150618