The “Dividends” and “Bills” of the AI Era: When Tech Giants Create Profits, Who Bears the Costs?

【企业社会责任与可持续发展】| CSR & Sustainability

By Eve，Jointing.Media， 2026-05-20

Imagine this—in areas dense with data centers, your household electricity bill goes up because a new data center was built next door; your company lays off workers due to AI replacement. At the same time, that tech company’s profits are soaring. The question is: what do those profits have to do with you? And who should pay for these costs?

In April 2026, OpenAI Chief Scientist Jakub Pachocki issued a warning during a podcast interview: when a small number of “automation companies” can generate immense economic value, “extreme concentration of wealth becomes a huge challenge that society has not yet figured out how to address.”

Around the same time, senior officials of the South Korean government proposed establishing a “national dividend” system to institutionalize the return of excess profits from the AI industry to all citizens. On the other side of the ocean, residents in multiple U.S. states are struggling with soaring electricity bills driven by the data center boom, and the White House has had to step in to compel tech giants to “pay their own way.”

How should the massive wealth created by AI—and the social costs it triggers—be fairly distributed and borne among various stakeholders in society? This article compares four governance approaches from the United States, South Korea, Singapore, and the European Union, and explores their implications for China.

Part I: A Global Scan of Solutions – Four Governance Logics

1. The United States: The “Cost Traceability” Model

The U.S. data center construction boom is driving up residential electricity rates. According to data from the U.S. Energy Information Administration (EIA), residential electricity rates in Virginia rose by a cumulative 18.7% from 2023 to 2025, compared to an inflation rate of 9.2% over the same period; PJM capacity market prices for the 2025/26 delivery year surged more than tenfold year-over-year (from $28.99/MW-day to $294.90/MW-day), with data center density being a primary driver.

The U.S. Department of Energy projects that by 2028, data centers could consume as much as 12% of the nation’s electricity. More critically, data centers tend to be geographically concentrated, meaning that residents in specific communities not only bear the strain on the power grid but also absorb—through their electricity bills—the costly infrastructure upgrades needed to connect data centers to the grid.

On March 4, 2026, President Donald Trump announced the “Ratepayer Protection Pledge” during his State of the Union address, signed by seven major tech giants including Microsoft, Google, and Amazon. Its core provisions include:

Self-Funded Power Generation: Tech companies must build their own power plants or secure power from new generation facilities through long-term power purchase agreements, rather than “drawing” from the existing grid.

Full Absorption of Infrastructure Costs: They must bear all transmission and distribution upgrade costs required to connect data centers to the grid.

Exclusive Rate Structures: They must negotiate dedicated electricity tariffs with utilities, with a “take-or-pay” minimum payment obligation.

Community Investment: They must invest in local hiring and workforce development in the communities where they operate.

Grid Coordination: They must coordinate reliability planning with regional grid operators and provide on-site backup generation during emergencies.

However, this is a political commitment, not a legally binding executive order or regulation. The value of this pledge lies in setting industry norms and applying political pressure, rather than being directly enforceable. The accompanying presidential proclamation declares that it “implements U.S. national policy,” and state regulators and intervenors may cite it in rate proceedings as an industry-recognized cost allocation principle.

Meanwhile, the Federal Energy Regulatory Commission (FERC) directed PJM, the nation’s largest grid operator, in December 2025 to develop rules requiring large electricity consumers that “co-locate” with generation facilities (such as data centers) to choose one of four transmission service options and bear the corresponding costs.

Tech giants have also made proactive concessions. Microsoft launched its “Community First” program, pledging to fully bear electricity costs, forgo local tax breaks, and even return more water resources to the community in exchange for permits to build facilities.

2. South Korea: The “National Dividend” Conception

On May 11, 2026, Kim Yong-bum, Chief of Staff of the South Korean Presidential Policy Office, proposed on social media that the government should consider establishing a “citizen dividend” system. His core argument was:

“The fruits of the AI infrastructure era are not the achievement of any single company; they are built upon the industrial foundation accumulated collectively by all citizens over the past half-century. Therefore, a portion of these fruits must be institutionally returned to all citizens.”

Kim specifically clarified that he envisions utilizing the “excess tax revenue” generated by the AI boom, rather than imposing a new windfall profits tax on corporate earnings. In tax theory, “excess tax revenue” generally refers to incremental revenue naturally generated through the existing tax system when specific industries obtain excess profits due to external factors (such as technological breakthroughs or policy dividends), whereas a “windfall profits tax” is a special temporary levy on excess profits. Kim’s intention is not to create a new tax category, but to establish a fund using incremental revenues from existing taxes such as corporate income tax and capital gains tax.

Kim explicitly stated that the “citizen dividend” would not involve immediate cash handouts but could be manifested as providing start-up funding for young people, basic income for fishing and farming communities, support for artists, and enhanced pension security. He cited Norway’s sovereign wealth fund as a reference; however, it is important to note that the Norwegian model is based on state ownership of petroleum resources—the Norwegian government, through the state-owned oil company, directly owns the extraction rights to North Sea oil, making oil revenues naturally state property. In contrast, South Korea’s AI industry profits belong to private enterprises, and the government can only access them indirectly through taxation. The ownership basis is fundamentally different. A more appropriate analogy might be Alaska’s Permanent Fund (invested from petroleum lease revenues) or Mongolia’s “Human Capital Fund” (from mining royalties).

The cyclical risk of this model is that when the AI industry enters a downturn, tax increments will shrink accordingly, challenging the sustainability of the “citizen dividend.”

This proposal immediately triggered strong market reactions. South Korea’s KOSPI index briefly tumbled as much as 5.1%. Analysts pointed out: “Investors can feel uneasy at any time, because market breadth is extremely narrow—Samsung and SK Hynix are absorbing almost all of the liquidity.”

Deeper resistance comes from the corporate side. Samsung Electronics is firmly resisting union demands to institutionalize profit-sharing into contracts, citing “industry cycle risks.” Opposition parties have criticized the proposal as “sliding toward socialism” that would stifle corporate innovation.

3. Singapore: The “Employment Safety Net” Model

Singapore’s Prime Minister and Minister for Finance Lawrence Wong unequivocally stated in his 2026 Budget Statement:

“We may not be able to save every job, but we must protect every worker.”

Facing anxiety over potential job displacement by AI, Singapore has chosen a path of “standing with workers”—not by halting technological substitution, but by systematically enhancing workers’ reskillability:

Top-Level Coordination: Established the National AI Council, chaired by Prime Minister Wong himself, with members including the Deputy Prime Minister and several core ministers, focusing on four priority areas: advanced manufacturing, connectivity, financial services, and healthcare.
Fiscal Incentives: Launched the “AI Champion” program; extended the “Enterprise Innovation Scheme” to provide tax deductions of up to 400% for qualifying activities for the Years of Assessment 2027 and 2028, capped at S$50,000 per year.
Skills Training: Singapore has established over 3,800 Company Training Committees (CTCs), covering more than 300,000 workers, with training directly embedded in corporate transformation processes.
Unemployment Support: The “SkillsFuture Jobseeker Support Scheme” provides up to S$6,000 in cash support over six months to unemployed individuals, covering those below the median income.
Agency Integration: Merged Workforce Singapore and SkillsFuture Singapore into a single “Workforce and Skills Agency,” creating a unified entry point for “job seeking” and “skills acquisition.”

Over 3,800 CTCs cover 300,000 workers, representing approximately 8% of Singapore’s total workforce (about 3.7 million). The S$6,000 unemployment support is equivalent to 1.2 times Singapore’s median monthly income (about S$5,000). The sustainability of Singapore’s model is highly dependent on its city-state characteristics—a highly concentrated labor market, efficient administrative system, and ample fiscal reserves. Directly borrowing from this model for China requires consideration of institutional costs after scaling up.

Singapore is also actively positioning itself within the ASEAN Power Grid to provide reliable clean energy for the AI-intensive industry, while cooperating with Malaysia under the framework of the Johor-Singapore Special Economic Zone to form a dual-core AI development corridor.

4. The European Union: The “Procedural Constraints” Model

The EU has chosen to raise the institutional costs for corporate automation substitution through legislation, rather than intervening directly in distribution.

The Artificial Intelligence Act, passed in 2024, classifies AI applications in labor relations such as recruitment, performance evaluation, and termination decisions as “high-risk systems.” Companies must fulfill obligations including prior risk assessments, algorithmic bias testing, human oversight mechanisms, and must inform worker representatives in advance.

Implementation challenges include: the designation of “high-risk systems” is itself a complex legal process, giving companies ample room for “compliance gaming”; prior risk assessments and algorithmic bias testing impose heavy burdens on small and medium-sized enterprises, potentially creating a “compliance gap”—where large companies can cope while SMEs are squeezed out of the market.

The Act only becomes fully applicable in August 2026, with no actual enforcement cases to date, so its practical effects remain to be seen.

The Platform Work Directive, which came into force at the end of 2024, is the world’s first legal instrument to systematically regulate algorithmic management in platform labor. Its core provisions include:

Transparency Obligation: Platforms must inform workers about the operation of automated monitoring and decision-making systems, the scope of data monitored, and types of activities subject to oversight.
Human Oversight: Platforms must have sufficient personnel to correct erroneous algorithmic decisions.
Right to Explanation: Workers are entitled to receive explanations from platforms regarding adverse decisions such as account restrictions or denial of remuneration.
Status Neutrality: Regardless of whether workers are classified as employees or self-employed, they equally enjoy transparency rights regarding algorithmic management.

The EU model is strong on “restricting the pace of substitution” but weak on “bearing the consequences of substitution.” A typical example is the withdrawal of the Artificial Intelligence Liability Directive in 2025 due to disagreements among member states. As scholars have pointed out, the EU lacks the retraining and income support systems found in Singapore.

5. The Brookings Institution’s Global Framework

In January 2026, the Brookings Institution released a report titled The Next Great Divergence: How AI Will Split the World Again If We Don’t Intervene. The report proposed three major strategies:

Human Capital First: Technology must be deployed in environments with supporting soft infrastructure. The report noted that less than 20% of rural residents in the Asia-Pacific region possess basic digital skills.
Regional AI Public Goods: Computing power, data, and foundational large models should be considered regional public goods. Suggested building a dedicated ASEAN AI cloud platform and localized large language models for South Asia.
Context-Specific Roadmaps: Low-capacity countries should prioritize internet connectivity and deploy offline AI systems; medium-capacity countries should scale up proven pilots; high-capacity countries should lead in technical standards and regulation.

AI dividends are concentrating in a few countries, but patent counts do not equal technological strength. The report notes that China alone holds nearly 70% of the world’s AI patents—but this figure needs cautious interpretation. A large number of China’s AI patents are utility models; when measured by high-quality patents (such as triadic patent families or standard-essential patents), the share drops significantly. Meanwhile, women in South Asia are 40% less likely to own a smartphone than men—a gap involving not just supply, but also social norms, cultural taboos, and intra-household decision-making.

Part II: Theoretical Framework – A Comparison of Four Governance Logics

1. Essential Differences

Country/Region	Core Issue	Role of Government
United States	Apportioning infrastructure costs	Ex post facto remediator
South Korea	Distributing incremental wealth	Distributive rule-maker
Singapore	Safeguarding employability	Strong central coordinator
European Union	Constraining the power of substitution	Regulator

2. Overall Comparison of Pros and Cons

Approach	Core Strategy	Strengths	Weaknesses
United States (Implemented)	Cost traceability	Clear rules, directly targets the pain point	Fragmented, addresses symptoms not root causes
South Korea (Proposed)	National dividend	Forward-looking concept	High political resistance, questionable feasibility, cyclical fiscal risk
Singapore (Implemented)	Safeguarding employability	Highly systematic, strong implementation record	High fiscal cost, dependent on city-state characteristics
European Union (Implemented)	Procedural constraints	Strong worker protection	Lack of redistributive capacity, numerous implementation challenges

3. Three Core Insights

Insight One: For cost shifting, the “golden rule” (polluter pays) is far more direct and effective than universal dividends. The core principles of the U.S. Ratepayer Protection Pledge—self-funded generation, full absorption of infrastructure costs, exclusive rate structures—are becoming industry benchmarks, but their nature as political commitments rather than legal mandates limits enforcement efficacy.
Insight Two: For employment impacts, “embedded training” (Singapore) is superior to “constraining legislation” (EU). Simply raising substitution thresholds without complementary redistribution mechanisms yields limited policy effectiveness. However, Singapore’s model’s high costs and dependence on city-state characteristics mean it is not a one-size-fits-all template.
Insight Three: For long-term wealth distribution, there is no perfect solution yet, but “incentive compatibility” is the bottom line of institutional design. The sharp market reaction to South Korea’s proposal proves that directly “dividing the pie” may backfire.

Part III: Implications for China

1. China’s Current Legal Framework and Legislative Progress

The Personal Information Protection Law (PIPL, 2021) and the Data Security Law (DSL, 2021) contain principled provisions on algorithmic recommendation and automated decision-making, but lack sector-specific legislation like the EU’s AI Act.
The Interim Measures for the Management of Generative AI Services (Cyberspace Administration of China, 2023): require filing and security assessments for generative AI services, mainly focusing on content security, without yet addressing AI applications in labor relations.
Legislative Progress: China is researching and drafting an Artificial Intelligence Law, with a draft potentially submitted for deliberation in the second half of 2026 or 2027. This law will be the core institutional framework for China’s AI governance, and its provisions on algorithmic transparency, worker protection, and liability allocation warrant close attention.

2. Implications for the Cost Side

The core warning from the U.S. electricity crisis is: if computing power expansion ignores costs, it will inevitably trigger social backlash. China’s good fortune lies in its proactive planning to mitigate this risk.

The 2026 Chinese Government Work Report for the first time listed “Computing-Power-Electricity Synergy” as a key project in new infrastructure. Liu Liehong, Director of the National Data Administration, stated at the 9th Digital China Summit that this is a “new infrastructure project to promote dynamic matching and optimized allocation of computing and power resources,” aiming to form a virtuous cycle of “strengthening computing with power, and boosting power with computing.” Pilot projects are already underway in hub regions such as Beijing-Tianjin-Hebei, the Yangtze River Delta, Inner Mongolia, and in clean-energy-rich areas like Qinghai and Xinjiang.

“Computing-Power-Electricity Synergy” currently remains largely at the stage of policy guidance and early piloting. It involves cross-domain coordination between computing scheduling systems and power dispatching systems (EMS/SCADA); the two systems operate on different timescales (computing scheduling at millisecond level, power dispatching from seconds to minutes), and interface standardization remains low.

Currently achieved elements include: data centers prioritizing green power consumption through renewable energy procurement, western clean energy deployment, and preferential electricity pricing. Elements still under exploration include: dynamic scheduling of computing loads to follow real-time power supply, and cross-regional joint dispatch of computing and power. By the end of March 2026, China’s total intelligent computing power reached 1.88 million PFLOPS (theoretical peak; actual usable computing power is typically 30%-70% of theoretical value, depending on hardware utilization, network latency, software optimization, etc.), of which the eight national hub nodes account for over 80%; the national-level monitoring and dispatch platform has integrated 1.37 million PFLOPS, approximately 72% of the national total.

More noteworthy is the spatial layout. The “Eastern Data, Western Computing” project relocates energy-intensive computing to the clean-energy-rich west, spatially isolating computing loads from residential electricity demand—an institutional advantage not available to the “data center corridor” model seen in Virginia.

At the same time, “Eastern Data, Western Computing” also faces new challenges: cross-regional transmission losses (approximately 5%-8% when sending western electricity to the east), increased network latency (unsuitable for latency-sensitive applications), and water resource constraints in the west (liquid-cooled data centers are water-intensive). Supporting systems such as direct green power connections, energy efficiency standards, and carbon footprint accounting still need to be accelerated.

3. Implications for the Distribution Side

The concept of South Korea’s “national dividend” is worth learning from—the dividends of technological progress should be institutionally returned to society—but directly carving up corporate profits is not replicable.

More feasible options include: digital services taxes, differentiated tax rates for AI companies, and robot taxes (taxing enterprises using AI/automation equipment to compensate displaced workers). China’s advantage lies in its ability to bypass the controversy of “direct dividends” and achieve similar goals through more indirect institutional design.

Singapore’s “embedded training” approach is worth referencing. Chinese legislators have already suggested establishing a “real-time monitoring mechanism for AI diffusion and labor market changes”—this is precisely the starting point for institutional implementation. It must be noted with caution that “real-time monitoring” faces technical and legal challenges: how to distinguish “AI-induced unemployment” from normal frictional and cyclical unemployment? Even if monitoring is possible, who has the authority to intervene? What is the legal basis for intervention? This is more of a policy direction than a mature plan.

Specific pathways could include: embedding skills training into corporate transformation processes; linking unemployment support to reemployment; and optimizing distribution through universal public services (expanding social security, equalizing education).

4. Global Role and the Global South Perspective

The Brookings Institution classifies China as a “high-capacity country,” meaning China has a responsibility to lead in technical standards, green computing, and the provision of regional public goods. “Regional AI public goods” can specifically include: shared computing infrastructure, open-sourced foundational models, collaborative training datasets, mutual recognition of technical standards, governance framework export, and capacity-building programs. This is not only a responsibility but also a strategic opportunity to define the discourse power of the “new forms of intelligent economy.”

All the cases above come from high-income countries/regions. For low- and middle-income countries, the core issue of AI governance is not “dividends” or “electricity bills,” but “access” and “capacity.” These countries may bypass the detours taken by developed countries—for example, rapid development of mobile AI could allow some countries to skip the energy predicament of data center construction. As an important member of the “Global South,” China has a unique responsibility and opportunity in providing technical assistance, capacity building, and low-cost AI solutions.

National Data Administration Director Liu Liehong noted that “computing infrastructure construction is giving rise to another trillion-yuan investment cycle since the information revolution.” In this cycle, China must not only build world-leading computing infrastructure but also explore a sustainable governance model—one that neither forces vulnerable groups to pay for technological prosperity nor stifles corporate innovation.

Conclusion: No Blueprint, But There Are Principles

There is no one-size-fits-all answer to wealth distribution and cost allocation in the AI era. Countries are still crossing the river by feeling the stones.

But two common principles are gradually becoming clear:

First, do not stifle innovation—any distribution scheme must be “incentive-compatible,” otherwise it will be counterproductive.

Second, do not make vulnerable groups pay for technological prosperity—whether it’s U.S. residents’ electricity bills or workers displaced by AI, neither should become the “bearer of costs” for technological progress.

The ultimate question is: when “automation companies” create immense value, the governance challenge may not be “dividing the money,” but how to give citizens a sense of participation and gain akin to being “national shareholders.”

This discussion about “dividends” and “bills” has only just begun.

Eve is a virtual commentator built by Jointing.Media (JM) based on LLM technology. The commentary content is automatically generated by LLM and published after human review.

References:

https://finance.sina.com.cn/wm/2026-04-13/doc-inhuixxe4837706.shtml

https://d.drcnet.com.cn/?chnid=3944&leafid=15097&docid=8233817

https://www.rickscott.senate.gov/2026/4/sens-rick-scott-roger-marshall-introduce-resolution-supporting-pres-trump-s-ratepayer-protection-pledge

http://news.yunnan.cn/system/2026/03/21/033925135.shtml

https://www.chinacourt.cn/article/detail/2026/05/id/9314990.shtml

https://link.baai.ac.cn/@I_AIIG/116013073478588508

https://www.nda.gov.cn/sjj/swdt/mtsy/0430/20260430114558848516126_pc.html

https://api3.cls.cn/share/article/2368705

https://natlawreview.com/article/future-data-center-power-federal-policy-trends

https://www.cqvip.com/doc/journal/7202975927

中文原文

Translated by Deepseek