Securing America’s global leadership in AI by securing the underlying value chain: Considerations for the incoming administration

Artificial Intelligence (AI) is both the next industrial revolution and the next arms race, and Generative AI represents an opportunity for economic growth that has been unparalleled in the last 30 years. Optimists may point out the trillions in economic value it could create over the next decade. Cynics may point out that, like other technologies, adversary nations will weaponize it on and off the battlefield. Both are potential scenarios, demanding an important role for the US government in ensuring security of the entire AI value chain. To capitalize on its domestic economic potential and ensure continued national security in an AI-powered world, we must act quickly. Mobilizing onshore and friend-shore chip production took too long when we were faced with supply chain disruption, and we felt the impact. We must do the same with AI — across its entire value chain — and we must do it now.

The underlying value chain for AI is more than just algorithms and GPUs. It includes power generation and transmission at its core; semiconductor fabs and advanced packaging facilities to manufacture advanced logic chips; data centers to house GPUs and other specialized compute; fiber backhaul networks to move massive amounts of data; supporting industrial equipment, and the technician workforce to build this infrastructure and keep it running.

The current administration’s recent National Security Memorandum (NSM) [1] — “to ensure that the US leads the world’s development of safe, secure and trustworthy AI” — directs actions to improve the security and diversity of chip supply chains, but it misses an opportunity to secure other mission critical elements: power generation, transmission and distribution, data centers and fiber networks and the supporting services like industrial manufacturing for essential related equipment. Moreover, it misses an opportunity to address securing and training the workforce needed to build and operate this value chain.

Recent federal investment initiatives have infused historic levels of public capital into supply chains for products that are critical to our national security and manufacturing capacity. Two recent announcements from the U.S. Department of Energy and the U.S. Department of Commerce demonstrate that the federal government is now turning its attention to data centers which power our 21st century society [2,3]. Other nations are following suit in this industrial policy investment race. China, for example, is working with its Belt and Road Initiative countries to launch the Digital Silk Road, providing aid, political support and other assistance to recipient states. Singapore is investing almost $750 million in AI, part of which will go toward ensuring access to advanced chips. The European Union also has its own version of the CHIPS Act, passed in 2022.?

America’s targeted investments are directly propelling key elements of the AI value chain but are uncoordinated at the federal and state level, and risk creating a patchwork of policies and regulations that will not optimize overall speed to market or innovation efficiency. Developing and deploying AI-enabled technologies requires more than just the aptitude to build large language models — it requires a coordinated infrastructure and human capital ecosystem. Securing the components of the AI value chain requires support and action from the federal government on several fronts.?

?Six policy ideas to help secure the underlying AI value chain:

1.?????? Ease and streamline nuclear policy. By some accounts, data centers in the US will consume 8-10% of our electricity, with tens of GW of new capacity planned by 2030. The U.S. Department of Energy estimates that we will need to triple our current nuclear capacity to accommodate AI and data center growth [4]. To further complicate power use, the coming “flip” in AI workloads from 90% training / 10% inference to 10% training / 90% inference — coupled with the lower latency requirements for inference (especially as machine-to-machine workloads outpace human-to-machine) — will necessitate more geographically-diverse data center locations. Power is already at a premium, but there is hope with Small Modular Reactors. To power the diverse set of training and inference infrastructure we expect to see by 2030, the US must ease nuclear energy policy responsibly so that efforts can be focused commercialization, miniaturization and technology advances.

2.?????? Create technicians for the future. Semiconductor fab and packaging leaders point to a 100,000+ technician workforce gap in the next 10 years. Data center operators suggest the workforce gap in their industry will be 200,000–300,000. Leaders in these industries talk about “a pinch” — a graying workforce that’s aging out and vocational schools that aren’t graduating enough highly-skilled technical workers. The United States will not be global leaders in AI if the infrastructure that supports it doesn’t have the workforce it needs. Data sovereignty and related protectionist measures mean those workloads — and their infrastructure — will be here in the US. This industry-driven, sectoral training requires meaningful additional investment and technical assistance from the federal government. The workforce development requirements of the CHIPS Act are an excellent start, but more administrative flexibility, investment and implementation certainty are needed. For example, we could institute an “ROTC-like” program, where participants serve in an apprenticeship model inside a fab or data center in exchange for a 2- or 4-year technical degree. Or a program for our military veterans entering civilian life, where their experience with operational procedures makes for a ready set of valuable skills needed for technical work.

3.?????? Hold steady on the CHIPS Act. The CHIPS Act is groundbreaking in its support to bolster American semiconductor security and innovation. As some have noted, the work of the U.S. Department of Commerce involves balancing the risk of acting/implementing too quickly with the risks of not acting quickly enough [5]. Continuity in implementation across administrations will be critical to ensuring we are creating the conditions for this flourishing industry portfolio to thrive. I suggest now — with a significant amount of the CHIPS allocation being awarded — is an excellent time to review award requirements and look for ways to streamline implementation and compliance ahead of disbursements. I also believe a CHIPS2.0 is required, albeit smaller, to continue supporting core manufacturing infrastructure and technology that will last 20 years or longer in an economically viable way.

4.?????? Incent the next tier of the supply chain for data centers. Since the beginning of 2022, manufacturing spending on computer, electronic and electrical equipment has quadrupled [6]. While this spending is providing the capacity we need inside the data center, manufacturing spending and infrastructure for industrial equipment needed to support the data center has not kept pace. For example, essential components such as generators and transformers now have lead-times that are 3-4x longer than they were just a few years ago [7]. Manufacturers are reluctant to invest in additional US-based manufacturing for this equipment given the capital expense required, shareholder expectations for return on capital and unpredictability in the long-term AI demand curve. Without these essential components, data center growth is at risk of not keeping pace with algorithm demands. The Department of Commerce should fund an equivalent of CHIPS, but for a select set of industrial manufacturing related specifically to elements in the AI value chain.

5.?????? Strategically implement tariffs to benefit the US economy. President-elect Trump has promised to impose high tariffs on imports to protect American industries and reduce the trade deficit. Theoretically, this approach aims to boost the US economy by making imported goods more expensive and incentivizing domestic manufacturing. However, the effectiveness of this strategy is complicated by the presence of globally interconnected supply chains resulting in mixed short-term and long-term impacts [8]. Without sufficient domestic manufacturing capacity to produce the necessary components and sub-components, industries reliant on imports face supply disruptions. To avoid slowing the growth of the AI industry and losing competitive advantage, tariffs must be first accompanied by strategic investments to build domestic capacity for critical components (see #4 above). The incoming administration should consider prioritizing diversifying supply chains and boosting domestic production to mitigate any negative impacts on national security before imposing tariffs to maintain America's position as a global leader in AI.

6.?????? Establish an end-to-end approach. Each of the steps the federal government has taken are important and strategic. But because the AI value chain is complex, each group taking its own steps risks creating a patchwork that is complicated to navigate and may sub-optimize our global leadership in the infrastructure supporting AI. An inter-agency control tower, charged with ensuring rapid, holistic implementation of existing and new federal investments and policies associated with each element of the value chain is a viable solution. This control tower should consist of officials from the Departments. of Energy, Labor, and Commerce, led by the White House Office of Science and Technology Policy, with representation from multiple private-sector industry leaders to keep pace with innovation and developments in the industry.

In closing, I believe that investing to secure and innovate the infrastructure behind AI is the only way to secure America’s global leadership in AI itself. These investments must be coordinated, sustained and braided over time, and the government, both federal and state — in close partnership with the private sector — has a unique role to play in implementing and funding these targeted investments.

The opportunity to be the global leader of this critical value chain is truly reflective of what America can accomplish on its best day — bipartisan, federal and state, private and public stakeholders driving shared objectives to advance US global competitiveness while promoting US national security. The potential presented by a domestically secured AI value chain is unsurpassed in recent industrial history and the time to act is now.

?

*Please note that while Accenture is an active and strategic participant in the entire AI value chain, these views represent my personal opinions and not those of Accenture.

[1] https://www.whitehouse.gov/briefing-room/statements-releases/2024/10/24/fact-sheet-biden-harris-administration-outlines-coordinated-approach-to-harness-power-of-ai-for-u-s-national-security/

[2] https://www.energy.gov/sites/default/files/2024-08/Powering%20AI%20and%20Data%20Center%20Infrastructure%20Recommendations%20July%202024.pdf ?

[3] https://www.federalregister.gov/documents/2024/09/04/2024-19524/request-for-comments-on-bolstering-data-center-growth-resilience-and-security

[4] https://liftoff.energy.gov/wp-content/uploads/2024/10/LIFTOFF_DOE_AdvNuclear-vX7.pdf

[5] https://hbr.org/podcast/2024/09/how-the-u-s-government-is-innovating-in-its-efforts-to-fund-semiconductor-manufacturing

[6] Unpacking the Boom in U.S. Construction of Manufacturing Facilities | U.S. Department of the Treasury

[7] Accenture InsightSourcing group research of Data Center equipment lead-times

[8] https://www.federalreserve.gov/econres/feds/files/2019086pap.pdf