Expanding the Advanced Manufacturing Investment Credit for AI Infrastructure

Executive Summary

Generative artificial intelligence (AI) has triggered an unprecedented build‑out of high‑performance computing capacity. Hyperscale data centres and AI‑optimized server farms consume extraordinary amounts of electricity, water and materials. In response to growing concerns over supply‑chain resilience and national competitiveness, U.S. policy makers enacted the Advanced Manufacturing Investment Credit (AMIC) under the 2022 CHIPS and Science Act to subsidise semiconductor fabrication facilities. The July 2025 One Big Beautiful Bill (OBBB) subsequently increased the credit from 25 % to 35 % for property placed in service after 31 December 2025semiconductors.org.

In late 2025, OpenAI CEO Sam Altman urged the U.S. to expand the AMIC to cover AI servers, data centres and critical grid componentsreuters.com. Altman argued that AI projects require $1.4 trillion of computational resources over the next eight years and that the U.S. should invest in re‑industrialising fabs, turbines, transformers and steelreuters.com. He emphasised that the credit differs from loan guarantees and stressed that OpenAI does not seek federal loan guarantees for data‑centre projectsreuters.com. This white paper analyses the legislative context, investment landscape, energy and environmental implications, and policy options for extending the AMIC to AI infrastructure. It concludes that expanding the credit could support domestic manufacturing and competitiveness only if coupled with robust efficiency, transparency and equity requirements.

Introduction: AI‑Driven Infrastructure Boom

The generative‑AI revolution has made computing power a scarce commodity. Training and operating large language models requires clusters of graphics processing units (GPUs) with energy‑intensive cooling systems. Private capital is pouring into hyperscale data centres and AI‑optimised server farms across the U.S. and worldwide:

• The Stargate initiative—a $500 billion consortium spearheaded by OpenAI, SoftBank, Oracle and Nvidia—aims to build 10 GW of data‑centre capacity. New facilities in Texas, New Mexico, the Midwest and Ohio will add ~4.5 GW of capacity and create roughly 25,000 onsite jobsreuters.com. Partners include SoftBank as lead equity investor, Oracle as developer and off‑taker, Nvidia supplying chips, CoreWeave providing cloud services, and Samsung, SK Hynix and Cisco contributing memory, networking and hardwarereuters.com.
• Meta Platforms pledged to invest at least $600 billion over three years in U.S. infrastructure and jobs, including large AI data centres. The company recently secured a $27 billion project‑finance deal to fund a facility in Louisianareuters.com.
• Google announced a $15 billion investment to build a 1 GW AI campus in Visakhapatnam, India and plans to spend $85 billion on data‑centre capacity globallyreuters.com. It is preparing its largest ever investment in Germany, emphasising renewable power and waste‑heat reusereuters.com.
• Global players such as Microsoft and Abu Dhabi’s G42 are deploying billions of dollars to expand AI data‑centre capacity in the United Arab Emirates, while Humain in Saudi Arabia aims to build ~6 GW of capacityreuters.comreuters.com.

The velocity of investment highlights a supply‑demand mismatch in computing. Yet the environmental and social cost of this growth has sparked intense debate. Data centres consumed 176 TWh of electricity in the U.S. in 2023—4.4 % of the nation’s total consumptionenergy.gov. Projections indicate they could absorb 6.7 %–12 % of U.S. electricity by 2028energy.gov, with global data‑centre demand expected to more than double by 2030datacenterdynamics.com. These facilities also draw millions of gallons of water, generate substantial waste heat and increase local utility bills. Policy makers must reconcile the imperatives of AI innovation, energy security, environmental stewardship and social equity.

Legislative Context and Sam Altman’s Proposal

The CHIPS Act and the Advanced Manufacturing Investment Credit

The CHIPS and Science Act of 2022 created the Advanced Manufacturing Investment Credit (AMIC), codified in Internal Revenue Code §48D. The credit initially allowed a 25 % refundable tax credit for investments in advanced manufacturing facilities that primarily produce semiconductors or semiconductor manufacturing equipmentirs.govthetaxadviser.com. Eligible property includes tangible assets integral to the facility and placed in service after 31 December 2022irs.gov. The facility must begin construction before 1 January 2027 and be located in the U.S.; foreign entities of concern are excluded, and there are recapture provisions for property disposed of or converted to non‑qualifying use within five yearsthetaxadviser.com.

The One Big Beautiful Bill Act

On 3 July 2025, Congress enacted the One Big Beautiful Bill (OBBB) (H.R. 1). Among other provisions, the OBBB increased the AMIC from 25 % to 35 % for qualified property placed in service after 31 December 2025thetaxadviser.comsemiconductors.org. The law also restored deductions for research‑and‑development (R&D) expenses and modified foreign‑derived intangible income (FDII) rules, reflecting a legislative strategy to make the U.S. more attractive for advanced manufacturingsemiconductors.org.

Sam Altman’s Proposal to Expand the Credit

In November 2025, OpenAI CEO Sam Altman publicly urged the U.S. to broaden the AMIC’s scope. He called for making AI servers, data centres and critical grid components eligible for the credit, arguing that AI requires massive capital expenditure and that the U.S. must re‑industrialise across chip fabrication, turbines and transformersreuters.com. Altman emphasised that such tax incentives would differ from federal loan guarantees and assured that OpenAI had only sought loan guarantees for semiconductor fabs, not for data‑centre projectsreuters.com. He stressed that taxpayers should not backstop private‑sector data‑centre projects; if OpenAI fails, other companies will fill the marketreuters.com.

This proposal has spurred debate. Proponents argue that extending the credit could catalyse domestic manufacturing of AI hardware, reduce reliance on foreign supply chains and accelerate innovation. Critics caution that subsidising data centres may exacerbate energy consumption, water usage, carbon emissions and grid stress, and may disproportionately benefit large corporations at the expense of ratepayers and communities.

The AI Infrastructure Investment Landscape

Major Projects and Financing

The table below summarises notable AI infrastructure commitments:

Sponsor/Consortium	Location and Capacity	Investment	Notes
Stargate consortium (OpenAI, SoftBank, Oracle, Nvidia, CoreWeave, Samsung, SK Hynix, Cisco, others)	U.S. (Texas, New Mexico, Midwest, Ohio), Norway, UAE	~$500 billion for 10 GW of data‑centre capacity; $18 billion project‑finance loan for New Mexico sitereuters.comreuters.com	4.5 GW expansion will create 25,000 onsite jobsreuters.com; partners supply chips, memory, networking and financingreuters.com.
Meta Platforms	U.S.	≥ $600 billion over three yearsreuters.com	Large AI data‑centre build‑out; $27 billion financing secured for Louisiana facilityreuters.com.
Google	Visakhapatnam, India (1 GW)reuters.com; Germany; global	$15 billion in India; $85 billion global data‑centre spendreuters.comreuters.com	Focus on renewable energy and waste‑heat reuse; expansion emphasises AI mission and job creation.
Microsoft & G42	UAE	> $15 billionreuters.com	200 MW data‑centre expansion with approval to import advanced Nvidia chips.
Humain (Saudi Arabia)	Saudi Arabia	plan for ~6 GW of capacityreuters.com; $3 billion initial campusreuters.com	Partnership with AirTrunk; aims for dual listing; sources chips from Nvidia, AMD, Qualcomm and others.
Alibaba Cloud	Dubai, UAE	380 billion yuan (~$53 billion) over three yearsreuters.com	Launching second data centre; region aims to host largest AI campus outside the U.S.
Qatar Investment Authority & Blue Owl Capital	Global	$3 billion digital infrastructure platformreuters.com	Focus on hyperscale compute capacity for global hyperscalers.

The scale of these investments demonstrates that AI infrastructure development is becoming an arms race. Access to chips, capital and energy is driving cross‑border alliances and new financial models, such as project‑finance loans with banks earning spreads over secured overnight financing ratesreuters.com.

Energy Demand, Water Use and Grid Strain

Data‑centre growth exerts unprecedented pressure on electricity and water systems. Key statistics include:

• U.S. consumption: data centres used 176 TWh of electricity in 2023 (4.4 % of U.S. load)energy.gov. The U.S. may require 325–580 TWh by 2028 (6.7–12 % of electricity)energy.gov. Roughly 60 % of a data centre’s electricity powers servers and 7 %–30 % goes to cooling systemspewresearch.org. Facilities consumed 17 billion gallons of water in 2023 and could double water use by 2028pewresearch.org.
• Fuel mix: as of 2024, natural gas provided > 40 % of data‑centre electricity, renewables ≈ 24 %, nuclear ≈ 20 %, and coal ≈ 15 %pewresearch.org. The International Energy Agency projects that data‑centre electricity consumption will more than double to 945 TWh globally by 2030, with renewables rising to 50 % and natural gas at 26 %datacenterdynamics.com.
• Grid impacts: the U.S. Energy Information Administration expects overall U.S. electricity demand to grow from 4,097 TWh in 2024 to 4,305 TWh by 2026, due largely to AI data centres, cryptocurrency and electrificationreuters.com. Transformer shortages have become acute; lead times for large transformers now range 80–120 weeks and specialised units take 3–6 years, threatening grid reliabilityutilitydive.com. Many utilities propose new natural‑gas plants to meet data‑centre demand, which may lock in fossil‑fuel dependence and raise emissions.
• Cost pass‑through: studies reveal that data‑centre expansion raises local electricity bills. In the PJM wholesale market, data centres increased capacity market prices by $9.3 billion, raising residential bills by $18 per month in western Maryland and $16 per month in Ohiopewresearch.org. Researchers project that AI and crypto mining could raise average U.S. electricity bills by 8 % by 2030pewresearch.org.

Environmental and Social Impacts

Several reports raise concerns about the environmental justice and social equity implications of data‑centre growth:

• The Science, Technology and Public Policy (STPP) report warns that a single data centre can consume up to 2 MWh of electricity—enough to power 2,000 homes—and millions of gallons of waterstpp.fordschool.umich.edu. Data centres often provide few high‑paying jobs compared with their subsidy packages and increase local utility ratesstpp.fordschool.umich.edu. The report advocates for energy audits, minimum performance standards, renewable additionality requirements, comprehensive reporting and reconsideration of tax breaksstpp.fordschool.umich.edu.
• The Consumer Federation of America notes that the U.S. hosted 5,426 data centres as of March 2025, with 335 under construction; they consumed as much electricity as Poland in 2023 and could use 12 % of U.S. electricity by 2028consumerfed.org. Utilities requested $29 billion in rate hikes in early 2025 to pay for data‑centre demand, and 160 AI‑focused data centres have been built in water‑stressed regionsconsumerfed.org.
• State and local policies: Many states are responding with legislation aimed at transparency, ratepayer protection and renewable energy requirements. According to the National Caucus of Environmental Legislators (NCEL), at least 22 states introduced over 60 bills in 2025 to address data‑centre impactsncelenviro.org. Bills include measures such as Georgia SB 34, which prohibits passing data‑centre costs onto ratepayers; Georgia HB 1192 requiring financial assurances from data centres; Utah SB 132 permitting direct contracting for electricity; New Jersey S 4143 requiring data centres to procure all energy from renewable or nuclear sourcesncel.net; and Oregon HB 3546 creating special rate classes for data centres. Other bills require transparency on energy and water use, limit tax breaks or restrict siting near communities.
• Environmental justice: In California, utilities argued that data‑centre demand justifies raising the state’s greenhouse‑gas emissions cap, threatening to prolong fossil‑fuel reliance. The largest under‑construction data centre could use as much electricity as 2 million homes, and the state is considering building new gas plants—often sited in disadvantaged communities—to meet demandcapitalandmain.com. In the Southeast, utilities plan to build 43 GW of new natural‑gas capacity based on projections that may be exaggeratedselc.org. Methane leakage from gas pipelines is four times higher than national estimates, compounding climate and public‑health risksselc.org.

Regulatory Landscape

Federal Policies

The Executive Order on Accelerating Federal Permitting of Data Center Infrastructure (23 July 2025) seeks to accelerate the permitting of large data‑centre projects. It defines Qualifying Projects as those that have capital expenditures ≥ $500 million and add ≥ 100 MW of load or are designated by the Secretaries of Commerce, Energy or Defensewhitehouse.gov. Covered Components include transmission lines, generation turbines, grid transformers, energy‑storage systems and microgrid componentswhitehouse.gov. Federal agencies may provide loans, loan guarantees and tax incentives but must ensure that federal support is less than 50 % of total project cost and treat support as not a major federal action to expedite environmental reviewwhitehouse.gov.

The order instructs the Council on Environmental Quality to create categorical exclusions under the National Environmental Policy Act (NEPA), to expedite Endangered Species Act consultations and to prioritise data‑centre projects on brownfield and Superfund siteswhitehouse.gov. It directs agencies to identify at least 16 federal sites suitable for co‑located AI facilitiesconsumerfed.org and revokes a prior order that required new data centres to be paired with new renewable generation. While streamlining may reduce permitting delays, critics argue that it weakens environmental safeguards and undermines public participation.

The Congressional Research Service notes that while data centres do not receive technology‑specific federal tax benefits, they can access energy‑related credits. Section 48 of the Internal Revenue Code provides a 30 % investment tax credit (ITC) for solar and battery storage systems, and Section 48E provides a similar credit for clean electricity. Data centres can also claim the Section 179D deduction for energy‑efficient commercial buildingscongress.gov. These credits incentivise energy efficiency and renewable integration but do not directly target data‑centre servers or grid infrastructure.

State and Regional Policies

As noted above, states are taking diverse approaches:

• Transparency and Reporting: Bills in Texas and Georgia (e.g., SB 1929 and HB 1192) require data centres to report energy and water usage and to provide financial assurances to cover infrastructure upgradesncelenviro.org.
• Ratepayer Protection: Georgia SB 34 and Virginia HB 2084 prohibit passing costs of data‑centre upgrades onto ratepayersncel.net. Oregon HB 3546 and Virginia legislation create special rate classes to shield households from higher billsncel.net.
• Renewable Requirements: Minnesota HF 16 offers a clean‑energy tariff for data centres; New Jersey S 4143 requires that all data‑centre electricity be from renewable or nuclear sourcesncel.net. New York S 6394 prohibits incentives for fossil‑fuel power purchase agreementsncel.net.
• Siting and Zoning: Laws in Maryland and Oregon mandate local impact studies or restrict data centres in residential zones, while Washington requires community benefits agreementsncelenviro.org.

These policies illustrate a patchwork of state‑level efforts to align data‑centre expansion with climate and social objectives.

Policy Analysis and Recommendations

Weighing the Pros and Cons of Expanding the AMIC

Potential benefits:

1. Boost domestic manufacturing: Extending the AMIC to AI servers, data centres and grid components could stimulate U.S. production of high‑performance computing equipment, memory chips, transformers and turbines, reducing dependence on foreign suppliers.
2. Reindustrialisation: Altman’s call to reindustrialise across fabs, turbines and transformers echoes concerns about supply‑chain fragility. A broader credit could catalyse investment in advanced manufacturing clusters that also produce clean‑energy equipment such as microgrids and battery systems.
3. Competitiveness and innovation: Tax incentives could help U.S. firms compete with state‑backed initiatives in China, the Middle East and Europe, fostering innovation in AI and high‑performance computing.

Risks and challenges:

1. Energy and environmental impact: Without conditions, a broader credit could accelerate build‑outs of energy‑intensive data centres, locking in natural‑gas dependence and straining water resources. Subsidising energy‑hungry infrastructure runs counter to climate commitments.
2. Equity concerns: Utility cost increases may be shifted to ratepayers, disproportionately affecting low‑income households and communities near new gas plants. Data‑centre jobs are few relative to public subsidies, raising questions about social return on investmentstpp.fordschool.umich.edu.
3. Uncertain demand forecasts: Analysts warn that some electricity‑demand forecasts are exaggerated; one report finds that projected growth through 2030 would require nearly 90 % of the global chip supply, which is unrealisticselc.org. Overbuilding could leave stranded assets and wasted subsidies.

Recommended Conditions for Extending the AMIC

To balance innovation with sustainability and equity, any expansion of the AMIC should include the following provisions:

1. Energy‑Efficiency and Performance Standards
   • Require qualifying projects to meet stringent power‑usage effectiveness (PUE) targets, energy‑efficient cooling and waste‑heat recovery systems.
   • Link tax credit eligibility to adoption of renewable energy, such as on‑site solar, battery storage or long‑term power‑purchase agreements for renewables. Encourage co‑located generation using retired coal plant sites as suggested by the DOEenergy.gov.
2. Transparency and Reporting
   • Mandate annual reporting of energy consumption, carbon intensity, water usage and local economic benefits. Require independent energy audits and publication of compliance data to build public truststpp.fordschool.umich.edu.
3. Ratepayer Protection and Cost Allocation
   • Incorporate safeguards to ensure that utilities cannot pass the cost of data‑centre infrastructure upgrades to ratepayers. Consider adopting special rate classes or direct contracting mechanisms similar to Georgia SB 34 and Utah SB 132ncel.net.
4. Community Benefits and Environmental Justice
   • Require applicants to negotiate community benefits agreements with local governments, guaranteeing job training, infrastructure improvements and environmental mitigation. Prioritize siting on brownfield or Superfund sites to avoid greenfield sprawlwhitehouse.gov and limit projects in water‑scarce regions.
5. Scaling Next‑Generation Technologies
   • Allocate a portion of the credit to research and deployment of energy‑efficient computing architectures (e.g., neuromorphic chips, photonics) and software optimisation to reduce overall compute demand. Encourage open‑source innovation and small‑business participation through set‑asides.
6. Sunset Provisions and Recapture
   • Impose sunset dates and recapture provisions to regularly evaluate whether the credit still serves national interests. Tie future extensions to demonstrated reductions in carbon intensity and local benefits.
7. Coordination with State and Federal Policies
   • Harmonise federal incentives with state‑level renewable requirements, transparency laws and siting regulations. Encourage states to develop regional planning frameworks that coordinate data‑centre siting with grid upgrades and resource planning.

Additional Policy Levers Beyond Tax Credits

1. Grid Infrastructure and Supply‑Chain Investment: Address bottlenecks in transformers and transmission by accelerating domestic manufacturing of grid equipment and implementing the DOE’s recommendation to reuse retired coal plants and develop microgridsutilitydive.comenergy.gov.
2. Permitting Reform with Safeguards: While the July 2025 executive order seeks to expedite permitting, Congress should ensure robust environmental review and public participation. NEPA categorical exclusions should not become loopholes for large projects; interagency coordination can improve efficiency without sacrificing accountability.
3. Integrated Planning: Federal and state agencies should develop integrated resource plans that account for AI‑driven demand, renewable deployment, energy‑efficiency improvements and climate targets. Demand‑side management and dynamic pricing can help balance loads during peak demand.

Conclusion

The AI revolution promises transformative benefits across science, medicine, education and creative industries. Yet the computational infrastructure underpinning this revolution presents profound energy, environmental and social challenges. Expanding the Advanced Manufacturing Investment Credit to include AI servers, data centres and grid components could strengthen the U.S. manufacturing base and enhance competitiveness. However, unconditional subsidies risk exacerbating energy consumption, carbon emissions and inequity. A successful policy must integrate strong efficiency standards, renewable energy requirements, transparency, ratepayer protections, community benefits and sunset provisions. By coupling financial incentives with environmental and social safeguards, the U.S. can ensure that the AI infrastructure boom supports both innovation and the public interest.

Visual Inspiration

The following illustration symbolises the intersection of AI infrastructure and clean energy. Sleek server towers blend with wind turbines and solar panels against a sunset horizon, reflecting the need to harmonise technological progress with environmental sustainability.