What Happens Next

What Happens Next

This chapter will not offer predictions. It will offer something more useful and more honest: scenarios, each one following with some internal logic from the conditions described in the preceding parts, each one requiring a different set of things to be true, and each one producing a different world.

The distinction between prediction and scenario matters here more than it does in most contexts. The transition described in this Almanac has no historical analog close enough to calibrate on. Nobody has navigated the concentration of frontier technology production in a single geographically contested location before. Nobody has managed the bifurcation of a global compute stack during a period of active great power competition. The honest answer to the question of what happens next is that the range of outcomes is wide, the tail risks are severe, and the variables that determine which scenario materializes are partly policy choices and partly contingencies that no analysis can predict.

The good resolution requires four things to be true simultaneously, each of them difficult, none of them impossible. First, the United States and its allies sustain the investment and political will required to build genuine leading-edge semiconductor capacity outside Taiwan on a timeline that matters. Not at the current pace of CHIPS Act implementation, but as a sustained decade-long commitment at a scale commensurate with the strategic importance of the outcome. Second, the bifurcation stabilizes at a level of divergence that maintains sufficient interoperability for global economic activity and scientific collaboration to continue. Third, some form of technical transparency mechanism emerges for AI development that does not depend on shared physical infrastructure. Fourth, China does not close the semiconductor gap through indigenous development before the first three conditions are met.

The good resolution is not utopian. It does not require the US-China competition to end or the geopolitical tensions around Taiwan to resolve. It requires the specific set of institutional failures described in Part III to be corrected sufficiently, and on a sufficient timeline, to prevent the worst outcomes. It is achievable in principle. The conditions for achieving it are not currently present.

The bad resolution does not require a single dramatic catastrophe. It requires only that the current trajectory continues without significant course correction until the conditions for course correction no longer exist. The initiating event in most bad resolution scenarios is a Taiwan disruption before domestic alternatives are ready. The disruption does not need to be a military invasion. A sustained naval blockade, a political crisis that produces export restrictions, a significant natural disaster, or a sufficiently credible threat of military action that causes TSMC's international workforce to evacuate and its supply chains to seize. Any of these produces the same outcome: the world's leading-edge chip production stops, and there is no substitute on any relevant timeline.

The deeper consequences develop over the years that follow. A sustained disruption of TSMC's operations, lasting long enough to allow China to consolidate its position in the developing world's infrastructure, produces a bifurcation that is no longer a managed competition between a leading Western stack and a trailing Chinese stack. It produces genuine symmetry: two frontier AI development programs, running on separate hardware, with separate model architectures, and no meaningful technical connection between them. Two nations with the most capable AI systems in history are developing those systems on incompatible infrastructure with no shared frameworks, no shared verification mechanisms, and no institutional architecture capable of managing the coordination problems that produces.

The bad resolution does not require malice from any actor. It requires only the continuation of the institutional failures documented in Part III, the deepening of the bifurcation documented in Part IV, and a triggering event that arrives before the good resolution's conditions are met.

The most probable world is neither the good resolution nor the bad one. It is a prolonged middle state, uncomfortable and costly, in which the bifurcation deepens gradually without fully stabilizing, the institutional failures persist without becoming catastrophic, and the window for the good resolution narrows incrementally without quite closing. The export controls hold imperfectly. China makes partial progress on indigenous semiconductor development, reducing but not eliminating the capability gap. TSMC continues to operate in Taiwan under increasing geopolitical tension, with periodic near-crises that each resolve without the disruption scenario materializing. The CHIPS Act fabs come online behind schedule and below leading-edge performance, providing partial insurance without genuine resilience.

This scenario does not end. It continues indefinitely, accumulating risk without triggering the resolution that would either address the underlying problem or make its consequences undeniable. It is the scenario that looks, from inside it, like stability. It is the scenario that produces the bad resolution if a single significant contingency breaks the wrong way.

The Almanac principle requires stating plainly what the author believes, not retreating into the false security of evenhanded scenario analysis.

I think the most probable world trends toward the bad resolution on a timeline of ten to twenty years, not because catastrophe is inevitable, but because the good resolution requires a quality of institutional response that democratic systems have not demonstrated in this domain and that the incentive structures described in Part III actively work against. The software-defined era produced institutions that are very good at optimizing within a stable physical substrate and very poor at recognizing when the substrate itself has become the variable. Correcting that limitation requires a degree of strategic clarity, sustained across electoral cycles and commercial pressures, that would be unusual in any historical period and is particularly difficult in the current one.

The people who understand this problem most clearly are not, for the most part, in positions of sufficient institutional authority to act on their analysis. The people in those positions are operating on frameworks built for the world that ended sometime around 2014, when the assumptions of the software-defined era began visibly failing, and they will continue operating on those frameworks until the cost of doing so becomes undeniable.

The cost will become undeniable. The question is whether it becomes undeniable early enough and in a form specific enough to allow a corrective response, or whether it becomes undeniable all at once, in a form that makes correction impossible.

This Almanac is, among other things, an attempt to make it undeniable early.