The short answer is yes, but not in the way most policymakers think. The longer answer, as Russ Wilcox argues, is that America and its allies are competing in the wrong arena.
Watch the full episode here:
At Davos and in Washington, the conversation still orbits around artificial general intelligence and frontier models. Wilcox, a State Department and Pentagon advisor, dismisses that fixation as misplaced. “Everyone is talking about… AGI or ASI,” he told me, “but the conversation is so focused on those media-driven terms that we’re losing sight of governance and how it’s actually being implemented now.”
The real contest, in his telling, is not about who builds the smartest model first, but who builds the most effective AI infrastructure institutional, physical and political.
Infrastructure, not innovation, is the bottleneck
By most conventional metrics, America is winning the innovation race. The leading frontier models are built by American firms; US universities dominate AI research output; venture capital continues to flow. Yet Wilcox’s claim that “we are losing the AI race to China” rests on a different yardstick: coordination.
China treats AI as a system-level capability embedded across institutions. Even the language is revealing. “In China, they don’t call AI ‘AI’… They just call it intelligence,” Wilcox notes. The implication is practical rather than philosophical: intelligence is something to be deployed everywhere, not sold as a product.
Western governance, by contrast, is fragmented. The United States alone has roughly 90,000 state and local government entities, each operating with its own data systems and procurement rules (US Census of Governments: https://www.census.gov/programs-surveys/cog.html). The result is duplication, opacity and slow diffusion—precisely the conditions under which infrastructure technologies underperform.
Chips are not the whole story
Much of Washington’s response has focused on semiconductor controls. Here Wilcox is blunt. “We should not be sanctioning chips,” he argues—not because China will comply, but because it will adapt. Export controls already appear to have accelerated Chinese efforts to build more efficient models and domestic supply chains, as analysts have observed in the wake of restrictions on advanced Nvidia GPUs (CSIS: https://www.csis.org/analysis/chinas-ai-industry-after-us-export-controls).
More importantly, fabrication is only one link in the chain. Advanced chips must be assembled, tested and packaged—processes still heavily concentrated in Asia. According to the Council on Foreign Relations, most US-designed semiconductors are sent abroad for these final steps (https://www.cfr.org/articles/onshoring-semiconductor-production-national-security-versus-economic-efficiency). Manufacturing capacity without downstream integration offers limited strategic leverage.
Electricity and water is the key to AI at scale
If governance is the soft infrastructure problem, energy is the hard one. Wilcox is sceptical that AGI-style systems are imminent, not for philosophical reasons but physical ones. “It’s an energy problem,” he says.
Data centres already consume roughly 1–1.5% of global electricity, a share projected to rise sharply as AI workloads grow (International Energy Agency: https://www.iea.org/reports/energy-and-ai). A single large data centre can draw as much power as a town of tens of thousands.
Cooling is an additional constraint. Academic research suggests that large-scale AI systems can have significant water footprints, depending on location and cooling method (Li et al., Making AI Less Thirsty, ACM: https://cacm.acm.org/sustainability-and-computing/making-ai-less-thirsty/). These are not marginal concerns. Grid capacity, permitting timelines and water rights increasingly shape where AI can be deployed at scale.
China, Wilcox notes, is investing heavily in energy coordination and load-balancing systems, while Western debates remain stuck between climate targets and local opposition.
What to do about it
Wilcox’s prescription is notably unromantic. First, stop treating AI as a silver bullet. “Don’t trust AI,” he advises. Use it as decision support, not decision-maker, with transparency at the level of inference rather than abstract model claims.
Second, build shared civic infrastructure. Digitising and connecting public records; budgets, zoning, permits, historical decisions, would allow local and federal governments to see patterns they currently miss. Estonia’s experience with digital government shows how administrative capacity can be transformed when data is treated as infrastructure, not paperwork (e-Estonia:
https://e-estonia.com/
).
Finally, align AI policy with energy and industrial policy. Without grid expansion, faster permitting and investment in power generation, AI ambitions will stall regardless of software advances.
The West is not short on talent or capital. It is short on coordination. Until that changes, it may continue to win demos while losing the system.
Russ Wilcox
Substack (The Pacific Divide):
https://pacificdivide.substack.com/
Instagram: https://www.instagram.com/russwilcox2013/
Originally published in the For Breakfast newsletter on LinkedIn, February 23, 2026. Full video episodes are on YouTube at @HaveItForBreakFast.