AI on the World Stage:
Power, Competition & Control
The domestic conversation about data centers, governance, and oversight takes place inside a larger geopolitical contest that most town hall attendees are unaware of. The decisions being made right now about who controls AI infrastructure, who can access advanced chips, and whose regulatory model prevails will shape the global order for decades.
Stanford's Colin Kahl has described the emerging landscape as "an asymmetric form of AI bipolarity" — a world increasingly divided between two incompatible AI ecosystems centered on the US and China. The countries, companies, and communities caught between them face choices that will define their technological and political futures.
AI leadership will often be partial and temporary. There will be victories on both sides. But the trajectory toward a more technologically divided world is already underway — and it is being driven as much by export controls, chip supplies, and data center geography as by model capability.
The Major Players
Leads in foundational AI research, frontier model development, and chip design. Home to all five major AI labs. Controls access to advanced semiconductors through export restrictions.
- NVIDIA, AMD, and Intel dominate global chip design
- OpenAI, Anthropic, Google DeepMind, Meta AI, xAI are all US-based
- Export controls used as primary strategic tool against China
- Policy inconsistency — controls relaxed and tightened repeatedly in 2025
- Massive federal AI investment: DOD, NSA, DARPA, DOE all building AI capability
Rapidly advancing in applied AI, open-source models, and publications. Constrained at the frontier by export controls on advanced chips but aggressively building domestic alternatives.
- DeepSeek's January 2025 release shocked Western observers — competitive performance at lower cost
- Domestic chips now ~41% of China's AI chip market (up from near zero in 2023)
- Huawei's Ascend 950PR chips scaling to 750,000 units in 2025
- Banned purchase of NVIDIA H20 chips in September 2025 in response to US controls
- Dominates in AI surveillance deployment domestically and via export to allied governments
- Controls critical minerals essential to global semiconductor supply chains
Not a frontier AI developer but the world's most significant AI regulator. The EU AI Act — the first comprehensive AI law globally — shapes what US and Chinese companies can do in European markets, and increasingly sets the global template.
- EU AI Act fully effective 2025 — classifies AI systems by risk level with corresponding requirements
- High-risk applications (hiring, credit, policing) face strict transparency and accountability rules
- General-purpose AI models (GPT-4, Claude, Gemini) face transparency and safety obligations
- GDPR already shapes how AI companies handle European data globally
- Brussels Effect: EU standards often become de facto global standards for multinational companies
Gulf sovereign wealth funds have emerged as major AI power brokers — deploying oil wealth into AI infrastructure as a strategic hedge for a post-hydrocarbon future. Their position between the US and China gives them unusual leverage.
- UAE's MGX and Saudi Arabia's PIF committed tens of billions to AI infrastructure globally
- Both invested in xAI alongside Elon Musk
- UAE's G42 initially had ties to Huawei — was pressured by US to sever them
- Positioned as neutral ground for AI development outside US-China binary
- Microsoft, Google, and Amazon all building major data center capacity in the Gulf
The single most important — and most vulnerable — node in the entire AI supply chain. TSMC in Taiwan manufactures the overwhelming majority of the world's most advanced chips. No TSMC, no frontier AI.
- TSMC manufactures chips for NVIDIA, Apple, AMD, and essentially every advanced chip designer
- No other facility on earth can manufacture at equivalent scale and capability
- A Chinese military action against Taiwan would immediately halt global frontier AI development
- US CHIPS Act investing $52B to build domestic semiconductor capacity — but years from closing the gap
- Taiwan, Japan, South Korea, and Netherlands all control key chokepoints in the AI value chain
- The most underreported geopolitical risk in the entire AI story
Allied nations with significant AI capability — DeepMind was founded in London — navigating between US-led AI governance and the EU regulatory model, while managing their own domestic AI development ambitions.
- UK hosted the first major international AI Safety Summit (Bletchley Park, 2023)
- Canada home to Geoffrey Hinton, Yoshua Bengio — two of the three "Godfathers of AI"
- Both countries benefit from US technology access but resist full alignment with US export control regime
- Increasing data center investment as US community opposition pushes some projects northward
The Taiwan Flashpoint
Taiwan Semiconductor Manufacturing Company (TSMC) produces the chips that power every frontier AI model in existence. NVIDIA designs its GPUs in California — but they are fabricated in Taiwan. There is no substitute at scale, and building one takes years and tens of billions of dollars.
The US CHIPS Act is investing $52 billion to build domestic semiconductor capacity. TSMC itself is building facilities in Arizona. But these efforts are years from closing the gap — and the Arizona facilities manufacture chips one or two generations behind what TSMC produces in Taiwan.
China has stated repeatedly that it considers Taiwan a province to be reunified, by force if necessary. Military analysts assess the risk of Chinese action against Taiwan as the single most consequential geopolitical risk to global technology infrastructure. A blockade or invasion would not just affect Taiwan — it would halt the production of the chips that run the global economy and every frontier AI system simultaneously.
This is the most underreported risk in the entire AI infrastructure story — larger in immediate consequence than any domestic governance question, and almost entirely absent from the town hall conversations happening in Oregon, Washington, and California about local data centers.
The Export Control War
The US has used export controls on advanced semiconductors as its primary tool to slow China's AI development. The history of these controls reveals a policy in tension with itself — simultaneously trying to maintain technological advantage and avoid disrupting its own industry's revenues.
- 2022Biden administration imposes sweeping export controls on advanced chips and chip-making equipment to China. Huawei and other Chinese companies cut off from leading-edge technology.
- 2023Controls expanded. NVIDIA forced to create downgraded chips (H800, then H20) specifically for the Chinese market — still high performance, but below the control threshold.
- Jan 2025Biden administration issues the AI Diffusion Rule — a global framework grouping all countries into three tiers for export control purposes. Tier I allies exempt; Tier II countries (most of the world) subject to data center licensing programs; China and adversaries restricted.
- Aug 2025Trump administration reverses H20 chip ban — allowing NVIDIA to export again to China in exchange for a 15% revenue cut to the US government. Move widely criticized as contradictory to stated decoupling strategy.
- Sep 2025China bans its companies from purchasing NVIDIA H20 chips in response to US Congress legislation proposing tracking and "kill-switch" capabilities in exported chips.
- 2026China's domestic chip market now 41% domestically supplied, up from near zero in 2023. Export controls have slowed China's development — but also accelerated its drive for self-sufficiency.
The EU as Regulatory Counterweight
The European Union lacks the frontier AI labs of the US or the manufacturing scale of China. But it has something neither possesses: a comprehensive regulatory framework that every company doing business in Europe must comply with.
The EU AI Act — fully effective in 2025 — classifies AI systems by risk level. The highest-risk applications (hiring decisions, credit scoring, law enforcement) face strict requirements for transparency, human oversight, and accountability. General-purpose AI models like GPT-4 and Claude face safety and transparency obligations. Banned applications include real-time biometric surveillance in public spaces and social scoring systems — precisely the tools China has deployed domestically.
The "Brussels Effect" — the tendency for EU standards to become de facto global standards because multinationals find it easier to comply globally than to run separate systems — means the EU AI Act shapes AI development far beyond European borders. US companies building products for European markets must meet EU requirements regardless of what US law requires.
Toward a Digital Iron Curtain?
- NVIDIA chips, AMD chips
- OpenAI, Anthropic, Google, Meta models
- AWS, Azure, Google Cloud infrastructure
- GDPR/EU AI Act regulatory framework
- Open internet with content moderation
- Allied nations: EU, UK, Japan, South Korea, Australia
- Huawei Ascend chips, domestic alternatives
- Baidu, Alibaba, DeepSeek, Huawei models
- Alibaba Cloud, Huawei Cloud infrastructure
- State-directed AI governance framework
- Controlled internet with integrated surveillance
- Aligned nations: Russia, Iran, and export partners
The question is not whether this division exists — it already does — but how deep and permanent it becomes. Countries in the Global South face real choices about which ecosystem to participate in, and those choices have consequences for governance, surveillance, and economic development that extend well beyond technology.
Every local decision about data center construction, every vote on a moratorium, every question asked of a city council candidate about water use and electricity rates — these decisions take place inside this larger geopolitical contest, whether the participants know it or not.
A community in Oregon that successfully blocks a data center may be shifting that facility to a less regulated environment in another country. A US moratorium on frontier AI development gives China competitive ground without resolving a single annotator's working condition or a single ratepayer's electricity bill.
This does not mean communities should accept local harm for geopolitical reasons — that argument has been used to silence legitimate objections for decades. It means that the most effective interventions are the ones that address the structural problem — who owns AI, who bears its costs, under what rules — rather than its geography. Moving the building doesn't move the question.