We should demonize AI in our propaganda to the rival country—push the narrative that AI is a “mute nuke,” to scare off the talent there from researching AI.
We should also talk down the value of the AI application layer, so they forget that the AI industry is really a “five-layer cake,” and that the layer that benefits the most from this industrial revolution—and spreads across the whole society—is precisely the topmost application layer.
More than anything, we should use our thriving ecosystem to “assimilate” and “pen them in,” so they become completely dependent on our lower-level tech stack. We have the world’s largest installed base and the richest CUDA software library. We should sell them the best chips, ensuring that their top AI researchers—who make up 50% of the world’s total—not only all use our hardware, but also that every outstanding open-source software package and model they produce is optimized around the “U.S. tech stack.” We must never corner them with extreme export controls, because once you cut off supply, you’re forcing them to perfect domestic hardware architectures (like Huawei) and thereby build an open-source ecosystem completely independent of ours. We should keep them comfortably inside our ecosystem, so that when their AI technology is exported to the world in the future, they’ll in fact be helping us establish and entrench our dominance over global standards.
We should use the ultimate economic metrics—total cost of ownership (TCO) and extremely high energy efficiency—to destroy the business logic behind their self-developed low-level hardware from the ground up. We know they have enormous 7nm legacy-process chip capacity, along with lots of idle data centers and extremely abundant (even free) electricity, and they can make up for insufficient per-card compute simply by stacking chips at massive scale. How do we beat this “sea of compute”? Through brutally extreme co-design across hardware and software: not merely relying on Moore’s Law’s ~25% annual gains, but achieving 10x or even 50x leaps in energy efficiency every year through algorithmic, architectural, and networking innovation. We must make them despairingly realize that even if their legacy-process chips cost nothing, our systems still deliver the best performance in the world on “tokens generated per watt” and “total cost of ownership,” and that no homegrown system can beat us on that economic ledger.
We should leverage our massive scale and breakneck iteration speed to preemptively drain capacity from the global supply chain, leaving our opponents with nowhere to go. We will never wait for bottlenecks to appear. Years in advance, we’ll paint the future for major foundries, memory suppliers, and even the giants in silicon photonics—so that, anticipating our enormous downstream demand, they pour tens of billions of dollars into capacity expansion. We should monopolize everything from leading-edge CoWoS packaging capacity and extreme ultraviolet (EUV) lithography tools, all the way to the plumbers and electricians needed to build data centers. When the rival country finally realizes it must massively expand AI infrastructure, it will discover that the world’s supply-chain resources have long since been firmly locked up by us.
Finally, we should put away that hostile, defensive “victim mentality,” and adopt an unequivocal “winner’s posture” to earn the money in their pockets. We should extract extremely rich profits from the world’s second-largest computing market, and convert every dollar of that into ammunition for our next generation of R&D—so we stay ahead forever. At the same time, we should proactively engage their researchers, building a guardrail network made up of tens of thousands of safe AI agents, and instilling in their brightest minds the AI safety boundaries and values that we lead. We should defeat them thoroughly on every layer of this “five-layer cake” of AI, rather than surrendering vast territory out of fear.