Ownership is not the only path to control. States that hold the land, the utilities, and the permits hold the terms on which AI scales.
Thematic Working Group Briefs – 2026
The AI Infrastructure Leverage Framework
AI is not simply the new electricity. Previous automation waves displaced physical or routine cognitive tasks at the margins. AI can be better perceived as the new labor wage. Large-scale AI systems can now perform non-routine cognitive work—legal analysis, medical reasoning, financial modelling—at marginal costs, which is, till now, declining dramatically, although skepticism is rising on the sustainability of the AI pricing. Skeptics claim that AI prices will surge after dependency is created. Governance frameworks built for utility infrastructure are, therefore, inadequate for a technology that may constitute a form of productive capacity in its own right.
The economic stakes reinforce this framing. The IMF projects global GDP will expand between 1.3 and 4% over the next decade (IMF, 2025). Corporate institutions have a more optimistic view: PwC (2023) projects AI will add $15.7 trillion to global GDP by 2030—roughly equivalent to China’s entire current economy. Goldman Sachs (2023) estimates the productivity effect at 1.5% points of annual growth across developed economies, a growth rate not recorded since the postwar industrial boom. More cautiously, MIT economist Daron Acemoglu projects total factor productivity gains of no more than 0.66% over the entire decade, arguing that those most expansive projections are produced predominantly by institutions with direct commercial exposure to the technology’s perceived value (Acemoglu, 2025).
Labor market projections are equally significant. The ILO and Poland’s NASK research institute find that 25% of workers globally fall within an occupation exposed to generative AI, but distinguish between transformation and replacement, concluding that full automation remains limited since most affected tasks still require human involvement even when AI performs them more efficiently (Gmyrek et al., 2025).
While artificial intelligence is frequently perceived as a weightless, cloud-based solution to global inefficiencies, the physical infrastructure required to sustain its computational power is resource-intensive. In 2025, global data centers consumed approximately 448 terawatt-hours of electricity—more than Saudi Arabia’s annual consumption—with AI workloads representing a rapidly growing share (UNU-INWEH, 2026). The International Energy Agency (2025) projects data center electricity demand will reach 945 terawatt-hours by 2030, alongside a water footprint equivalent to the annual basic domestic needs of 1.3 billion people in Sub-Saharan Africa and a land footprint exceeding 14,500 square kilometers. 90% of AI-specialized data center capacity is currently concentrated in the US and China (UNU-INWEH, 2026).
The debate over who should control AI infrastructure has moved into mainstream policy, as governments increasingly frame compute and frontier models as both strategic assets and national security issues.
Hard Intervention (Nationalization and State Control): [a] One camp argues that AI’s strategic and risk profile justifies direct state authority. The author Charles Jennings (2023) calls for nationalizing core AI technologies under a federal “Humane AI Commission” modeled on the Atomic Energy Commission, with licensing before frontier systems are deployed. British investor Ian Hogarth, who popularized the concept of “AI nationalism” as early as 2018, took the argument further by questioning publicly whether the UK government should reverse Google’s 2014 acquisition of DeepMind and reinstate the lab as a national asset (Hogarth, 2018).
Public Dividend and “Soft Nationalization”: [b] A second camp seeks public stakes without seizing day‑to‑day operations. Sanders’s American AI Sovereign Wealth Fund Act (2026) proposes a one‑time 50% stock tax on large AI firms—those earning at least 200 million dollars annually from AI—to create an estimated 7‑trillion‑dollar public fund with voting shares, board seats, and mandated citizen dividends (Sanders, 2026).
Sovereign AI: Build Rather Than Seize: Most governments have gravitated toward sovereign AI strategies that prioritize state‑backed building over expropriation. China’s push for self‑reliant chips, models, and cloud capacity is one attempt to shield national AI from foreign controls (Xi, 2025). In the Gulf, the UAE and Saudi Arabia are using sovereign wealth funds and state‑linked tech groups—such as G42, the Falcon models, and projects like the “Stargate UAE” campus—to build domestic AI data‑center capacity and models tailored to local languages and regulatory requirements (IMD, 2019; Aidata Insider, 2026). In Europe and Africa, debates over “public digital infrastructure” and “digital colonialism” frame sovereign AI as a way to secure cultural and data autonomy against dependence on US platforms, with African researchers warning that current AI supply chains risk making the continent a “data colony” unless states assert data sovereignty (Taghizade & Ahmadov, 2025; AI4D, 2025). Latin America is beginning to articulate its own sovereignty agenda: Brazil’s projects like Latam‑GPT seek to move the region from AI consumption to creation (Levy Yeyati, 2025; Daniel, 2026).
Market Autonomy and Anti‑Fusion Concerns: A smaller but vocal camp warns against deep state ownership or control. David Sacks (2026), the US administration’s designated AI policy coordinator, argues that nationalization or heavy equity stakes risk accelerating a “corporate‑government fusion” that would create a de facto central government AI with unprecedented power over information and behavior.
The Policy Gap: None of these positions resolves the core problem. Nationalization risks capital flight; dividend proposals are redistributive but do not address structural compute access or dependency; sovereign building programs are capital-intensive, exclusionary for most states; market autonomy ignores the externalities of private concentration. This policy gap motivates the Infrastructure Leverage Framework.
Control over AI infrastructure remains highly concentrated. TSMC held approximately 72% of the global foundry market share in late 2025, rising as AI demand accelerated (Yahoo Finance/Counterpoint Research, 2025). NVIDIA commands between 80 and 90% of the AI accelerator market by revenue (Silicon Analysts, 2025). At the infrastructure layer, AWS, Microsoft Azure, and Google Cloud collectively hold 68% of the global cloud market, a share that has grown every year since 2018 as capital requirements have risen to effectively prohibitive levels for new entrants (Synergy Research Group, as cited in Statista, 2026).
However, the architecture of dominance is beginning to fracture from within. On June 25, 2026, OpenAI unveiled Jalapeño, a custom inference chip developed with Broadcom, claiming roughly 50% cost savings per inference token relative to current Nvidia GPUs and a development timeline of nine months—described as the fastest advanced ASIC cycle on record (TechSpot, 2026). Amazon, Google, Meta, and Microsoft have all developed parallel custom silicon programs. The trend reflects a structural logic: the largest AI consumers are vertically integrating specifically to escape Nvidia’s pricing and supply constraints.
At the state level, several governments are mounting analogous efforts. Japan’s Rapidus corporation is targeting 2nm fabrication by 2027 in partnership with IBM. The EU’s Chips Act has committed €43 billion to double European semiconductor production (Semiconductor Engineering, 2026). China’s SMIC and Huawei have advanced toward 5nm production despite US export controls.
The critical caveat is that chip design and chip fabrication remain distinct problems. OpenAI’s Jalapeño is itself manufactured by TSMC. Every major custom silicon program depends on Taiwanese or South Korean fabrication at the leading edge. The challengers are diversifying design; they have not yet broken the fabrication chokepoint. This distinction matters directly for the ILF: it is physical manufacturing capacity and grid-connected facility access, not chip design, where state leverage most credibly applies.
The abstract vulnerabilities described in the policy literature have, in 2026, become operational events. The clearest illustration of sovereignty risk is the Anthropic case. Following a prior dispute in which Anthropic refused to allow US military use of its models for fully autonomous weapons—resulting in the company being placed on a Pentagon supply chain blacklist—the US Commerce Department issued an export control directive on June 12, 2026, ordering the suspension of access to Anthropic’s Fable 5 and Mythos 5 models for all foreign nationals, including Anthropic’s own non-US employees (Anthropic, 2026; Time, 2026). States whose public services, research institutions, or critical infrastructure had integrated Anthropic’s frontier models had no legal recourse. The incident demonstrated not only that private AI companies can be weaponized as instruments of foreign policy, but that host states bear the disruption costs while retaining none of the governance rights.
Europe’s response has been reactive rather than structural, where states discover the dependency after the architecture is installed, and absorb significant transition costs as a result. France’s domestic intelligence agency DGSI announced in June 2026 that it was replacing Palantir with domestic provider ChapsVision, following Germany’s intelligence services in making the same transition. French PM Lecornu stated that France “cannot accept new strategic dependencies in the digital sphere.” The UK is separately reviewing its £330 million NHS data contract with Palantir following sustained political pressure (Molle, 2026).
The energy externality is the most politically visible bottleneck, especially in the US. In the PJM electricity market stretching from Illinois to North Carolina, data centers accounted for an estimated $9.3 billion price increase in the 2025–26 capacity market. Wholesale power prices on the PJM grid rose 76% in the first quarter of 2026 compared to the same period in 2025. The costs of grid reinforcement to serve data centers are being allocated to residential consumers, while the fiscal value of those facilities flows to corporate shareholders. Moreover, tax cuts were granted without any binding conditions on computing access, grid obligation, or public benefit. Virginia—the world’s leading data center jurisdiction—waived $1.9 billion in sales tax revenue from data centers in 2025 alone.
Pillar A: The Infrastructure Investment Compact
The ILF restructures the state incentives from the vague promise of jobs and investment, states negotiate infrastructure compacts in which public subsidy—discounted land, subsidized utility connections, accelerated grid permitting—is provided in exchange for two binding returns: first, mandatory compute-sharing, data residency, and transparency commitments; and second, a meaningful equity stake in the physical infrastructure and the company. This equity stake converts the state from a passive tax-granting host into a co-owner of AI infrastructure, with governance rights and financial returns attached to that ownership. The UAE’s Mubadala Investment Company can be a similar operational template: a sovereign investment vehicle that takes structured equity stakes in technology ventures in exchange for market access and infrastructure support, accumulating governance influence without operational control (Mubadala, 2024).
This model also resolves the grid externality directly. Data centers should be required, as a condition of the compact, to fund proportional contributions to grid reinforcement and resilience infrastructure serving affected zones — not as a tax, but as an obligation attached to the equity arrangement.
The US has largely ceded fiscal value without governance leverage: states gave away revenue and received no binding public-interest conditions. China’s model achieves compute sovereignty through a different mechanism: direct state equity and policy direction over private AI companies, with subsidies conditioned on compliance with ideological content requirements, surveillance mandates, and research restrictions. China’s model has produced the second-largest domestic computing capacity. However, its structural cost to research autonomy and ecosystem integrity is not transferable to pluralist democratic contexts. The ILF proposes a middle path: equity co-ownership of physical infrastructure, not state direction of research agendas, but rather to facilitate its deployment in government sectors.
Pillar B: From Procurement Consumer to Service Investor
State purchasing power is conventionally framed as a demand-side instrument: governments buy AI services and can attach conditions to those purchases. Though implementing ILF state has already provided infrastructure investment and equity under Pillar A, it is not a consumer negotiating a price—it is an investor whose equity entitles it to preferred AI service terms for public sector applications, as a dividend on the infrastructure contribution rather than as a market transaction.
Accordingly, incentives shift for both sides: AI companies operating on state-enabled infrastructure have a reciprocal obligation to serve public sector needs at rates and under transparency conditions that reflect the state’s foundational investment.
Pillar C: The Nationalization Anchor
With equity stakes and service compacts in place, the nationalization anchor is less about outright seizure and more about escalation rights. If an operator exits a jurisdiction, transfers ownership to a foreign entity not subject to host-state jurisdiction, or materially breaches public-interest obligations, the state's equity position converts into preemption rights—the ability to purchase out the private share at a pre-agreed valuation, or to trigger utility classification that brings the facility under regulatory rate control.
The anchor is not a guarantee of independence. It's the convergence of corporate growth and the public's priorities, which is not expected to be optimal.[c]
References
Acemoglu, D. (2025). The simple macroeconomics of AI. Economic Policy, 40(121), 13–58. https://doi.org/10.3386/w32487
African Union. (2022). Digital transformation strategy for Africa (2020–2030). AU. https://au.int/en/documents/20200518/digital-transformation-strategy-africa-2020-2030
AI4D. (2025). Digital sovereignty or data colony? AI and Africa’s place in the global economy. Artificial Intelligence for Development Programme. https://www.ai4d.ai/research/digital-sovereignty-or-data-colony-ai-and-africa-s-place-in-the-global-economy
Aidata Insider. (2026, March 30). The Middle East’s playbook for sovereign AI infrastructure. https://aidatainsider.com/ai/the-middle-easts-playbook-for-sovereign-ai-infrastructure/
Anthropic. (2026, June 12). Statement on the US government directive to suspend access to Fable 5 and Mythos 5. https://www.anthropic.com/news/fable-mythos-access
Bloomberg. (2026, May 14). Data centers push power bills up 76% on largest US grid. https://www.bloomberg.com/news/articles/2026-05-14/data-centers-drive-76-rise-in-power-bills-on-largest-us-grid
Bureau of Industry and Security. (2023). Export controls on advanced computing and semiconductor manufacturing items. U.S. Department of Commerce. https://www.bis.doc.gov/index.php/documents/about-bis/newsroom/press-releases/3211-2023-10-17-bis-press-release-acs-fy24-controls-final/file
Business Model Analyst. (2026, June 25). States are pulling back data center tax breaks in 2026. https://businessmodelanalyst.com/data-center-tax-break-rollback/
Center for a New American Security. (2026, January). Sovereign AI Index 2026. CNAS.
Counterpoint Research. (2025, December). Global foundry market share Q3 2025. Counterpoint Technology Market Research.
Daniel, M. L. (2026, June 7). What does AI sovereignty mean for Latin America? TechPolicy Press. https://www.techpolicy.press/what-does-ai-sovereignty-mean-for-latin-america/
France 24. (2026, June 16). French spies break with data-analysis giant Palantir, wary of relying on US tech. https://www.france24.com/en/europe/20260616-french-spies-break-with-data-analysis-giant-palantir-wary-of-relying-on-us-tech
Goldman Sachs. (2023). Generative AI could raise global GDP by 7 percent. Goldman Sachs Research. https://www.goldmansachs.com/insights/pages/generative-ai-could-raise-global-gdp-by-7-percent.html
Goldman Sachs. (2025, August). How will AI affect the global workforce? Goldman Sachs Research. https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-global-workforce
Goldman Sachs. (2026, March). How will AI affect the US labor market? Goldman Sachs Research. https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-us-labor-market
Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Nafradi, B., Rosłaniec, K., & Troszyński, M. (2025). Generative AI and jobs: A refined global index of occupational exposure (ILO Working Paper 140). International Labour Organization. https://doi.org/10.54394/HETP0387
Hogarth, I. (2018, June 13). AI nationalism [Blog post]. Ian Hogarth. https://www.ianhogarth.com/blog/2018/6/13/ai-nationalism
IMD. (2019, November 17). Whoever leads the AI race will lead the future, UAE minister says at IMD’s signature programme. International Institute for Management Development. https://www.imd.org/news/digital/updates-whoever-leads-the-ai-race-will-lead-the-future-uae-minister-says-at-our-signature-programme-in-dubai/
International Energy Agency. (2025). Energy and AI. IEA. https://www.iea.org/reports/energy-and-ai
International Monetary Fund. (2025). The global impact of AI (Working Paper No. WP/25/76). IMF. https://www.imf.org/en/Publications/WP/Issues/2025/04
Jennings, C. (2023, August 31). As AI becomes more powerful, Portland tech expert urges federal regulation. Oregon Public Broadcasting. https://www.opb.org/article/2023/09/01/artificial-intelligence-charles-jennings/
Levy Yeyati, E. (2025, June 3). Smart AI regulation strategies for Latin American policymakers. Brookings Institution. https://www.brookings.edu/articles/smart-ai-regulation-strategies-for-latin-american-policymakers/
McKinsey Global Institute. (2023). The economic potential of generative AI: The next productivity frontier. McKinsey & Company. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier
Molle, C. (2026, June 16). French spy service drops Palantir in favour of French company, says Lecornu. Yahoo Finance UK. https://uk.finance.yahoo.com/news/dgsi-drops-palantir-french-firm-145948295.html
Mubadala Investment Company. (2024). Annual review 2024. https://www.mubadala.com/en/news/mubadala-annual-review-2024
Pew Research Center. (2025, October 24). What we know about energy use at US data centers amid the AI boom. https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/
PricewaterhouseCoopers. (2023). Sizing the prize: What’s the real value of AI for your business? PwC. https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-prize-report.pdf
Sacks, D. (2026, June 5). David Sacks criticizes proposed AI nationalization. Letsdatascience. https://letsdatascience.com/news/david-sacks-criticizes-proposed-ai-nationalization-21bb4fba
Sanders, B. (2026, June 1). A.I. is a public resource. You should own half of it. The New York Times. https://www.nytimes.com/2026/06/01/opinion/artificial-intelligence-bernie-sanders.html
Sanders, B. (2026, June 17). News: Sanders introduces legislation to create $7 trillion AI sovereign wealth fund [Press release]. Office of Senator Bernie Sanders. https://www.sanders.senate.gov/press-releases/news-sanders-introduces-legislation-to-create-7-trillion-ai-sovereign-wealth-fund/
Sanders, B. (2026). American A.I. Sovereign Wealth Fund Act: Summary. Office of Senator Bernie Sanders. /api/wp-media/AmericanAISovereignWealthFundActSummary.pdf
Semiconductor Engineering. (2026, April 21). Annual global IC fabs and facilities report. https://semiengineering.com/annual-global-ic-fabs-and-facilities-report/
Silicon Analysts. (2025). AI chip market analysis: NVIDIA data center dominance. https://siliconanalysts.com/analysis
Stateline. (2026, February 24). Data center tax breaks are on the chopping block in some states. https://stateline.org/2026/02/24/data-center-tax-breaks-are-on-the-chopping-block-in-some-states/
Synergy Research Group. (2026, May). Worldwide cloud infrastructure market share Q1 2026 [Data set]. Statista. https://www.statista.com/chart/18819/worldwide-market-share-of-leading-cloud-infrastructure-service-providers/
Taghizade, E., & Ahmadov, E. (2025). Techno feudalism and the new global power struggle: Echoes of a digital cold war. International Journal of Research and Innovation in Social Science, 9(2), 1144–1170. https://dx.doi.org/10.47772/IJRISS.2025.9020093
TechSpot. (2026, June 25). OpenAI debuts Jalapeño, a custom chip built to cut ChatGPT costs and reduce Nvidia reliance. https://www.techspot.com/news/112890-openai-debuts-jalapeo-custom-chip-built-cut-chatgpt.html
Time. (2026, June 13). Anthropic pulls its most powerful AI models after US bars foreign access. https://time.com/article/2026/06/13/anthropic-fable-mythos-ban-US-security/
United Nations University Institute for Water, Environment and Health. (2026, June 3). Rising emissions, depleting water and vanishing land: AI is threatening natural resources for billions. UNU-INWEH. https://unu.edu/inweh/news/environmental-cost-of-AIs-Enrgy-use-carbon-water-and-land-footprints
Virginia Mercury. (2026, April 27). Data center tax exemption changes still holding up Virginia budget. https://virginiamercury.com/2026/04/27/data-center-tax-exemption-changes-still-holding-up-virginia-budget/
World Governments Summit. (2025). His Excellency Omar Sultan Al Olama — Minister of State for Artificial Intelligence, UAE. https://www.worldgovernmentssummit.org/about/leadership/omar-sultan-al-olama
Xi, J. (2025, April 26). China's Xi calls for self-sufficiency in AI development amid U.S. rivalry. Reuters. https://www.reuters.com/world/china/chinas-xi-calls-self-sufficiency-ai-development-amid-us-rivalry-2025-04-26/
Yahoo Finance/Counterpoint Research. (2025, December). AI chips driving foundry boom: Why TSMC is winning most. https://finance.yahoo.com/news/ai-chips-driving-foundry-boom-153301101.html
[a]is this supposed to be another sub-heading? if so, just make it bold and add a space
[b]same here
[c]approved
