360° View: Scaling Trusted AI in a Fragmented World

360° View: Scaling Trusted AI in a Fragmented World

By Mark Kennedy, NYU DRI Director

If America wants to expand global access to U.S.-aligned AI while preventing adversaries from acquiring military-relevant capabilities, policymakers must focus on more than controlling algorithms. They must also focus on shaping systems.

AI competition is no longer about who invents the best model. It is about who can scale trusted compute across borders and boundaries.

That begins with infrastructure. As David Yu emphasizes, the real chokepoints are not lines of code but semiconductors, energy systems, and the industrial supply chains that power advanced data centers. Compute governance is now national power. But controlling hardware alone will not secure long-term influence.

Winston Ma correctly highlights that China’s strategy is not merely defensive—it is distributive. By offering open models and bundled infrastructure to emerging markets, Beijing is shaping the upper layers of the AI stack and embedding itself in growth markets that will define future demand. If the U.S. approach is seen only as restrictive, America will defend foundations while losing ecosystems.

Jeff Kucik reminds us that trade architecture is inseparable from technology leadership. Plurilateral supply chains cannot be secured through narrow bilateral deals. Without upstream access to materials and downstream access to customers, scale becomes fragile.

And Jerry Haar underscores the unavoidable security dimension: targeted export controls, model testing, infrastructure protection, and allied norms are essential to slowing military misuse.

But the central task for winning the AI systems competition is integration. The U. S. must pair guardrails with growth—export controls with energy expansion, security screening with capital mobilization, and standards with market access. The country that builds the most compelling, scalable, and trusted AI ecosystem will shape the technological future. In this contest, scale is strategy.

The Real Chokepoints of AI: Compute, Energy, and Industrial Scale

By David Yu, PhD, CFA, NYU Stern and Shanghai, DRI-WISC Affiliate

As America seeks to expand global access to U.S.-aligned AI while limiting adversarial acquisition of military-relevant capabilities, policymakers should focus closely on the underlying compute infrastructure and the industrial supply chains that enable the development and scaling of advanced models and algorithms. Focusing solely on restricting algorithms or software creates a less comprehensive and ultimately less effective approach while also risking unintended constraints and consequences on innovation and open technological progress. The true strategic advantage increasingly lies in the ability to scale advanced, energy-intensive data centers supported by powerful semiconductors, reliable energy systems, and specialized materials. Yet scaling these industrial capacities represents the real strategic chokepoints shaping global AI competitiveness.

Ensuring leadership in these areas requires forward-looking policies that strengthen domestic production, accelerate infrastructure development, and incentivize private enterprise to invest in semiconductor manufacturing, energy generation, and resilient sourcing of critical materials. Current chokepoints across the three foundational pillars of compute hardware, energy infrastructure, and materials supply chains should drive coordinated public-private strategies designed to secure long-term technological advantage. Equally important is deepening collaboration and government-sponsored investment among trusted allies to build integrated industrial ecosystems, enabling aligned nations to scale AI capabilities responsibly while maintaining shared standards and security safeguards.

At the same time, policymakers must recognize that excessive restrictions risk fragmenting the global AI ecosystem, encouraging parallel development pathways among adversarial actors, and reducing U.S. influence over emerging norms. A balanced approach should promote tiered-access frameworks that expand responsible participation among partners while mitigating risks associated with high-impact military applications.

By prioritizing industrial and energy resilience that drives compute governance and alliance-based technology frameworks, U.S. policymakers can effectively balance innovation, global adoption, and national security, which will sustain long-term leadership in an increasingly competitive and strategically significant technological landscape.

Two Strategies, One Stack: The Battle for Global AI Influence

By Winston Ma, Executive Director, Global Public Investment Funds Forum, NYU Law, DRI-WISC Affiliate

Jensen Huang’s five-layer AI stack—energy, chips, cloud, models, and applications—is not simply a technology framework. Mapped against the global strategies of the two AI superpowers (the U.S. and China), it reveals a divergence that will determine in which ecosystem the emerging world builds its AI future.

The U.S. approach centers on hardening the lower layers—energy, chips, and computing infrastructure. Through strict export controls on advanced semiconductors and high-end AI systems, Washington aims to maintain chokepoints at the foundation of the stack. This protects technological primacy and national security but often appears restrictive to countries moving toward their AI futures.

In sharp contrast, China is pursuing radical openness at the upper layers while actively helping nations build their own full stacks through its Digital Silk Road initiative. Beijing not only open sources powerful models such as DeepSeek and Qwen, but also supports emerging markets in developing energy, telecom networks, data centers, and cloud computing via bundled “Chinese tech stack” offerings and “AI-in-a-box” solutions optimized for constrained compute environments.

China’s “low cost, efficiency-first” philosophy—a direct result of U.S. chip sanctions—has become particularly compelling for resource-constrained markets. Since these countries cannot afford expensive advanced AI chips anyway, China’s approach lets them fine-tune capable open models on their proprietary local data and domain expertise, delivering rapid commercialization.

The result is a clear split. Many swing states selectively combine both systems to preserve autonomy. Yet many more emerging markets and lower-income countries are gravitating toward China’s more accessible and empowering ecosystem.

The U.S. is defending the layers it can control. China is empowering—and gradually locking in—the layers that matter most. U.S. policymakers must recognize that technology control alone is not compelling for the Global South.

Modernizing Trade Law for the AI Era

By Dr. Jeffrey Kucik, WISC Global Fellow

If the United States is to remain a global leader in innovation, it must adopt policies that look beyond its borders. This requires a trade strategy that secures long-term relationships with upstream suppliers and downstream customers. Unfortunately, recent trade initiatives have failed to meet these requirements, placing the nation’s technological leadership at risk.

America’s eight most recent trade agreements exhibit two systemic flaws in common.

First, bilateral agreements fail to reflect the realities of globalized production. In the modern era, the “AI stack”—which ranges from rare-earth mineral extraction to high-end semiconductor fabrication—relies on deeply integrated, plurilateral supply chains. Bilateral deals lack the jurisdictional breadth required to shore up these multi-country dependencies. Instead, this reliance on bilateralism creates market fragmentation and exposes U.S. firms to avoidable supply shocks.

Second, these agreements are too narrow in scope. They fail to promote technology trade in either direction. Strikingly, only one of the eight recent agreements (the U.S.-Taiwan initiative) explicitly mentions “artificial intelligence,” and it does so only within the restrictive context of regulatory barriers, rather than in market promotion.

By prioritizing defensive restrictions over proactive market expansion, the U.S. risks a two-front strategic failure: it will lack firm commitments from the providers of raw materials essential for hardware, and it will lose access to emerging customer markets. As the U.S. remains hesitant to sign trade-promoting deals, it cedes vital economic territory to China, which continues to penetrate global markets with its own digital standards and subsidized tech ecosystems.

Securing the AI Military Edge

By Dr. Jerry Haar, WISC Global Fellow

To slow adversaries’ access to AI-empowered military capabilities while preserving American innovation, U.S. policymakers should combine targeted export controls, compute and model governance, investment and data restrictions, and allied coordination. 

First, policymakers should tighten and update controls on advanced AI chips, accelerators, and semiconductor manufacturing equipment to keep pace with architectural changes. They should condition any export licenses to high‑risk jurisdictions on strict end‑use/end‑user screening, mandatory on‑premise monitoring, tamper‑resistant telemetry, and rapid shutdown obligations for violations. 

Policymakers should also regulate dual‑use frontier models for national security risk. They should mandate pre‑deployment testing of frontier and high‑risk, dual‑use models for capabilities that could significantly aid biological/chemical weapons development, offensive cyber, or autonomous weapons planning

Still another action is to treat large‑scale AI compute infrastructure (national AI clouds, defense AI test ranges, major model training clusters) as critical infrastructure. They should feature mandatory cybersecurity standards to prevent adversary intrusion, model theft, or covert fine‑tuning. Policymakers should also restrict capital, talent and data flows that supercharge adversary military AI. They should expand outbound investment screening to cover equity, joint ventures, and structured finance that builds adversary capabilities in advanced chips, model training platforms, and military‑relevant AI applications. Part and parcel to this is to introduce targeted controls on the export or sale of high‑value defense‑relevant datasets (high‑resolution SAR imagery, detailed maritime traffic, aerospace telemetry, specialized biological datasets) to entities in adversary jurisdictions and/or front companies. They must also limit government‑funded AI researchers and contractors from engaging in high‑risk collaborations with defense‑linked institutions in adversary states, as well as require disclosure and mitigation for potentially sensitive joint projects.

Finally, the U.S. should build allied standards and norms for military AI use. This means operationalizing the U.S.-backed Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy into binding commitments among allies on testing, human‑in‑the‑loop requirements, and restrictions on fully autonomous lethal systems. 

Development Impact: For developing economies, access to affordable compute, reliable energy, and secure digital infrastructure is now foundational to competitiveness. Policies that pair security guardrails with scalable infrastructure investment can enable inclusive AI adoption as a tool of broad-based growth without locking countries into coercive technology ecosystems or strategic dependency.