Computers may exhibit properties that render them both an object and tool of geopolitical competition, but fabrication complexity still demands high-end AI talent. The United States’ ability to maintain an edge in AI research lies in attracting and retaining the world’s best talent. Out of the three legs of the AI triad, computing power is exhibiting the most visible signs of geopolitical competition. Data and algorithms are still integral to artificial intelligence, but the constraints involved in manufacturing microchips create levers that are easily exploited. However, corralling cutting-edge technology is only one strand of competition in AI.
The focus on transistor size and expensive equipment may lead people to assume technology is essential to success, rather than the AI workforce. The technology is indeed restrictive; the resources required for microchip manufacturing pose significant barriers to entry. Fabs can cost between $10-20 billion, with lithography machines running as high as $150 million. The cutting-edge technology that permits 3-10 nm transistors remains the province of two to three companies. Other aspects of the semiconductor supply chain are similarly concentrated; for example, only one company in the world produces EUV lithography equipment. Given the bottlenecks evident in the semiconductor supply chain, export controls are a natural tool of geopolitical competition in AI.
However, this assumes better computing power is predicated merely on shrinking transistors. Moore’s law famously describes progress in microchip miniaturisation—transistors were approximately 10000 nm in 1971, and we will soon see scales as small as 3 nm. But the curve may be flattening; a Moore’s law ‘doubling’ required 18 times as much R&D spending in 2015 as it did in 1971. Vexing manufacturing and operational problems present themselves for transistors smaller than 10nm, let alone verging into picometres. Governments and companies face diminishing returns for shrinking transistor size, and urgently need other avenues to increase computing power.
Effective chip design is one option; chips optimised for AI are orders of magnitude faster than general processors (with transistors of the same size). Research in quantum computing, nanotubes, or RISC-V approaches could perhaps deliver the next phase in computing. Transistor size has shrunk because of improved engineering and design, but the premise underlying microchips has changed little since the 1960s. A revolutionary jump in computing power therefore could come from new fields of science, rather than only improving decades-old technology. For better chip design and revolutionary breakthroughs, human capital is as important as technological tools, if not more.
Though China’s strides in technology research are impressive, the United States is still home to most cutting-edge research and development. For the foreseeable future, the world’s best STEM specialists will flock to U.S. shores because of the latent power of American industry, even if China can surpass U.S. investment in emerging technologies. A steady influx of diverse talent is imperative to optimising all legs of the AI triad, because it is individuals who will be responsible for AI breakthroughs, not the tools they use. For instance, having access to large amounts of data to train algorithms does not guarantee success; any breakthroughs will instead result from novel ways to manipulate and shape this data.
Similarly, increasing computing power entails moving past the size of transistors. The best way to pursue innovative research is by pursuing diverse approaches, both in research areas and research teams. The Manhattan Project saw scientists from across the world and sub-disciplines collaborate, and the world was changed irrevocably. The opportunities afforded by controlling supply chains should not become blinders to more promising ways of increasing computing power. To maintain its technological primacy, the United States has to increase research in numerous subdisciplines and consolidate its ability to attract and retain talent. This requires immigration reform for STEM graduates, and funding to incentivise and expand research. A narrow focus on denying China technology is a sensible short-term solution, but would be moot if the United States is deposed as the preeminent destination for STEM researchers.
