The Leverage Shift: Frontier Labs and the State

The second critical implication of Mythos is institutional rather than technical. Frontier AI labs now possess capabilities that governments can’t easily replicate, evaluate, or even fully understand in real time.

Regulation has traditionally assumed that governments lag slightly behind innovation but can eventually catch up. This assumption no longer holds. The pace at which frontier models evolve compresses development cycles far beyond legislative, regulatory, and bureaucratic response times. We’re now entering a world in which the limiting factor is no longer innovation, but governance—and, so far, governance is losing the race.

The relationship between governments and frontier labs is inverting. States increasingly need these companies more than the companies need states—not for revenue, but for strategic capacity. Mythos likely accelerates this shift, too. This creates a new political economy of AI governance. Labs gain leverage not by lobbying against regulation, but by making themselves indispensable. Governments, in turn, shift from trying to police AI to trying to ensure preferred access to it. This is a quiet but profound change.

This tension sits at the core of my national-security argument: AI is no longer simply governed by the state; it’s increasingly negotiated with, even as it becomes one of the central tools of state power itself.

Geopolitical Implications

Our core message remains that US-China competition is the central organizing dynamic of the global system. AI has long been an arena of strategic US-China competition, but frontier-level models such as Mythos raise the stakes of the rivalry. This space has been elevated from an abstract race for technological leadership into a concrete contest over national-security advantage.

As a result, Washington and Beijing will likely intensify efforts to shield critical AI-related sectors from external dependence, while simultaneously pouring resources into domestic capability building.

This pressure will likely extend beyond models themselves to the range of inputs required to train and deploy next-generation AI: Advanced semiconductors, rare earths and specialty materials, power generation and grid reliability, and the physical and digital infrastructure that supports large-scale computing power. Reducing vulnerability in these areas will increasingly be seen not as industrial policy, but as a core national-security objective. Reducing vulnerability in these areas will increasingly be seen not as industrial policy, but as a core national-security objective.

Investment Implications

From an investment perspective, the “Mythos moment” reinforces the need for open-mindedness and cross-disciplinary synthesis when it comes to AI―not binary views. AI’s effects may reverberate across sectors and asset classes rather than concentrating in a narrow set of beneficiaries. And once AI systems cross key capability thresholds, job displacement, repricing, and competitive restructuring can happen quickly.

Several key investment themes stand out to me here:

• Infrastructure remains foundational. Intelligence itself may become cheap, but the ability to generate, scale, and deploy it is constrained by physical realities. Semiconductors, networking equipment, data centers, power generation, cooling, and grid capacity remain bottlenecks, and demand continues to outpace supply across much of the stack.

• Software isn’t dead—but it’s being repriced. The software outlook is more complicated than it’s ever been, as investors reassess terminal value,   competitive moats, and margin durability in the face of rapidly advancing AI models. The likely winners may be firms that use AI to accelerate revenue and expand margins simultaneously, rather than treating AI as a defensive talking point or a blunt cost-cutting tool.

• Second-order effects matter at least as much as first-order adoption. AI adoption is shaped not just by technical capability, but by governance, compliance, workforce adjustment, and public sentiment. Pushback related to job displacement, data-center expansion, or high-risk AI use cases can slow deployment, even as underlying capabilities continue to advance.

• Sustained AI competition at the geopolitical level may drive even more investment across national-security themes. This is critical for my broader geopolitical framework and suggests a massive structural investment opportunity.

As AI becomes embedded in intelligence-gathering, intelligence-surveillance-reconnaissance systems, cyber operations, space systems, and weapons development, competition among major powers may reinforce demand for traditional defense spending and defense innovation. This includes software-defined military systems, space and satellite infrastructure, cyber resilience, autonomous technologies, logistics, energy security, and the industrial base required to support them.

In short, I don’t believe AI should be thought of as a bubble, or as a predetermined economic destiny. Rather, it’s likely an enduring, uneven force reshaping markets alongside a rapidly shifting, increasingly fragmented geopolitical backdrop―and a disruptor that will continue to create differentiated opportunities in the investment landscape.