Opening
January 20, 2025, 5:00 a.m. Pacific Time. An AI startup office on Bryant Street in San Francisco's SoMa district. The CTO woke up and checked his X (formerly Twitter) feed. The screen was flooded with a single name: DeepSeek.
A startup from Hangzhou, China, had released the benchmark results for its R1 model, and the numbers jolted Silicon Valley awake. On the AIME 2024 math olympiad qualifying exam, R1 scored 79.8% — closing in on OpenAI o1's 83.3%. In coding, mathematical reasoning, and logic problems, it matched America's frontier models. But one number changed everything. The total training cost for V3, the base model underlying R1, was $6 million ($5.58 million in GPU compute costs; all subsequent DeepSeek training costs in this book refer to total training costs). It used 2,048 H800 GPUs. What American Big Tech spent hundreds of millions of dollars to accomplish, this team had done for one-fiftieth the cost. Meta used 30.8 million GPU-hours to train a comparable model; V3's pretraining required 2.78 million GPU-hours. R1 added reinforcement learning on top of V3.
The CTO messaged a colleague: "We're spending ten times more money, but what if the gap isn't ten times the performance?"
The crisis was not one company's. That year, the combined AI capital expenditure of America's Big Tech four (Amazon, Alphabet, Microsoft, and Meta) reached $400 billion. An amount equal to 22% of South Korea's GDP was being poured into a single technology. And a $6 million model had matched the performance that money was supposed to buy. Was this a problem of technology, of capital, or of institutions?
The moment a productivity revolution intersects with hegemonic competition. History records three such intersections. And each time, the wealthier side did not win. The side that could make the technology work across an entire society won.
This chapter traces those three precedents. Rome and Carthage's contest for the Mediterranean. Britain and Germany's industrial succession. America and the Soviet Union's Cold War. The same pattern repeats across all three cases. Extract that pattern, and a framework for reading the US-China competition in the age of AI emerges.
Section A: The Scalable Side Wins, Not the Wealthy Side — Rome vs. Carthage
Carthage's Silver and Rome's Roads
In the third century BCE, two powers contested supremacy over the Mediterranean: Carthage and Rome. By the measure of wealth alone, the outcome seemed predetermined. Carthage was far richer.
According to Pliny, Carthage drew annual revenues of 12,000 talents from mines on the Iberian Peninsula. Its maritime trade network stretched from Phoenicia in the east to Iberia in the west and North Africa to the south. Maria Eugenia Aubet's analysis describes Carthage's trade network as a "hub-and-spoke" structure. Silver, tin, gold, and grain flowed through the hub of Carthage. The city was the financial center of the ancient world.
Rome was different. It was an agrarian state. It had no dazzling trade networks, no mining revenues. According to Polybius, as of 264 BCE, Rome's adult male citizen count stood at 292,234. The total population of the Italian Peninsula was between three and four million, a scale comparable to Carthage's homeland and dependencies combined.
But Rome had one thing Carthage did not: roads.
Construction of the Via Appia began in 312 BCE. It was 4.1 meters wide (14 Latin feet), laid with standardized drainage channels, milestones, and relay stations. This road system eventually extended across the empire in 400 major routes totaling 80,000 kilometers. Ray Laurence's research tracks this scale. Peter Temin argues that this road network was the physical foundation that made a "single market" possible. Military movement, commercial exchange, and information transmission all traveled the same roads.
But it was not roads alone. Rome's real weapon was institutional.
Citizenship as Protocol
Rome possessed a mechanism for systematically absorbing conquered territory. In modern terms, this might be called "protocol standardization."
The legal system, which expanded from civil law (ius civile) to the law of nations (ius gentium), provided a universal dispute-resolution framework applicable even to non-citizens. The praetor peregrinus was a dedicated magistrate handling disputes between foreigners, and between foreigners and citizens. Disparate peoples and cultures were integrated into a single transactional system. Just as TCP/IP connects different networks under one communication protocol, Roman law connected different societies under one legal framework.
The gradual expansion of citizenship was the core of this system. A staged grant from Latin rights (ius Latii) to full Roman citizenship (civitas Romana). After the Social War (91-88 BCE), citizenship was extended across the entire Italian Peninsula. The conquered were converted into system participants. The incentive for rebellion was removed, the tax base expanded, and the pool of mobilizable human resources widened.
Water infrastructure operated on the same logic. According to Frontinus, the city of Rome alone had eleven aqueducts with a combined length of 500 kilometers. Per capita daily water supply was 500 to 1,000 liters (comparable to modern New York's 500 liters). This standardized water and sewage system was replicated across provincial cities. It was the physical mechanism by which the "Roman way of life" was transplanted.
Carthage had no such mechanism. A commercial oligarchy dominated decision-making. The interests of elite families like the Barcids dictated the direction of territorial expansion. Hannibal's Italian campaign was, strictly speaking, more of a Barcid family project than a national strategy. Not an institution for absorbing territory, but a network for extracting from it. That difference was decisive.
Carthage's army was composed of mercenaries: Libyans, Numidians, Iberians, Gauls, Balearic slingers. Cost-efficient, but lacking in loyalty and cohesion. The Mercenary War (241-237 BCE) after the First Punic War exposed this vulnerability in dramatic fashion. When the war ended and Carthage could not pay its soldiers, the mercenaries revolted.
The Paradox of Cannae
August 216 BCE. The Battle of Cannae. Hannibal Barca's Carthaginian army encircled and annihilated the Roman force. Rome lost between 50,000 and 70,000 men in a single day — the worst defeat in the history of the Republic.
Hannibal's officer Maharbal urged an immediate march on Rome. When Hannibal refused, Maharbal said: "You know how to win a battle, but you do not know how to use a victory."
What decided the outcome was resilience. Rome absorbed the defeat. Here the difference between the two systems comes into sharpest focus. According to Polybius, based on the 225 BCE census, Rome's total mobilizable manpower was 700,000 — citizens and Italian allies combined. Carthage's deployed forces numbered 170,000, mostly mercenaries. A system that loses 70,000 at Cannae and still has 630,000, versus a system that cannot recover from a single defeat at Zama. Nathan Rosenstein's analysis describes Rome's alliance system as a "distributed pool" of human resources. The costs of war were spread across the entire network.
Carthage was different. Its defeat at the Battle of Zama in 202 BCE broke the system. The mercenary model was a structure where "only as many as you can pay can be mobilized." When the war chest ran dry, the army vanished.
Arthur Eckstein called this the triumph of "systemic resilience." Rome lost battles, but its system did not collapse. Carthage won battles yet lacked the institutions to scale its victories. The contest was decided not by the size of wealth but by the scalability of the institutions that mobilized it.
Here was the first large-scale demonstration that a distributed system is more shock-resistant than a centralized one. Carthage's commercial network was efficient, but it could not absorb the shocks of war. Rome's alliance network was less efficient, but it could distribute defeats and regenerate.
That is the lesson of the first precedent.
Section B: The Next-Wave Rider Wins, Not the First Mover — Britain vs. Germany (1870-1914)
1900: The Paris World's Fair
Paris, 1900. The German pavilion opened at the World's Fair. AEG electric motors ran without pause, BASF synthetic indigo gleamed in glass display cases, and Bayer chemical products lined the shelves. Germany used the fair as a showcase for its industrial capabilities.
A British industrial delegation toured the exhibition grounds and compiled a report. One sentence from that report survives: "Germany is making better use of the science we invented than we are."
The gap between invention and application. That single sentence captures the essence of what happened between 1870 and 1914.
The Curse of the First Mover
Britain was the birthplace of the First Industrial Revolution. The steam engine, the spinning frame, the railway. In 1870, Britain's share of world manufacturing output stood at 31.8%. The phrase "workshop of the world" was no exaggeration. Germany's share that year was 13.2%, less than half of Britain's.
But that success became a poison.
David Landes called this the "penalty of the pioneer." Rewiring a gaslit city for electricity is harder than electrifying a blank one from scratch. Massive sunk costs in coal-and-iron infrastructure blocked the transition to new technologies. The world's first railway system was a source of pride, but it also made upgrading track gauges and facilities difficult.
The education system was a more serious problem. Oxbridge was the cradle of classical education. Science and engineering occupied a low rung in the academic hierarchy. Technical training depended not on a systematic institutional framework but on apprenticeship. Michael Sanderson's analysis concludes that Britain's education system was suited to the empirical innovation of the First Industrial Revolution, but ill-suited to the science-based innovation of the Second.
Laissez-faire ideology minimized government intervention in industrial policy. Free trade principles prohibited infant industry protection, and the gold standard combined with fiscal orthodoxy constrained large-scale state investment. In short, Britain was trapped by the institutional inertia of its own past success.
The Late Starter's Blank Slate
Germany was different. It had only unified in 1871, so its industrial history was short. That brevity became a weapon. Unencumbered by the legacy of the First Industrial Revolution, Germany could optimize from the start for the Second — electricity, chemicals, and steel.
Friedrich List's infant industry protectionism (The National System of Political Economy, 1841) provided the intellectual foundation. The Customs Union (Zollverein, 1834) unified the domestic market. The state invested directly in railways, armaments, and the chemical industry, creating demand. According to Alexander Gerschenkron's theory of "late industrialization," the later a country industrializes, the larger the role of banks and the state. Germany followed exactly that path.
Germany's universal banking system (Universalbank), led by Deutsche Bank and Dresdner Bank, provided long-term capital to industrial enterprises. This was the fusion of banking and industry that Rudolf Hilferding named "finance capital" (Finanzkapital). It differed structurally from Britain's market-based finance, which centered on the stock market and short-term lending.
But the truly decisive difference was education. Germany's Technische Hochschule (technical university) system built a triangular structure of education, research, and industry around eleven national polytechnics. The Humboldt university model, grounded in the unity of research and teaching (Forschung und Lehre), was its philosophical foundation. In 1913, Germany's university enrollment stood at 77,000 students. Britain's was 26,000. Fritz Ringer's data. A threefold gap.
BASF (1865), Bayer (1863), and Hoechst (1863) established in-house R&D organizations, and university-trained chemists flowed into them. Johann Peter Murmann analyzes this as the prototype of a "National Innovation System." A structure in which three axes — university, industry, and government — interlocked and turned together. Between 1901 and 1914, Nobel Prize winners in the sciences numbered twelve for Germany and seven for Britain. The country that was not the land of invention took home more Nobel Prizes.
The Numbers Tell the Story of Reversal
Line up the numbers, and the scale of the reversal becomes clear.
World manufacturing share. 1870: Britain 31.8%, Germany 13.2%. 1913: Britain 13.6%, Germany 14.8%. Paul Bairoch's data. In forty years, Britain's share fell by more than half, and Germany overtook it.
Steel production. In 1870, Britain produced 300,000 tons, Germany 170,000 tons. In 1893, Germany surpassed Britain. By 1913, the gap had widened: Germany 17.6 million tons, Britain 7.7 million tons. A factor of 2.3. Brian Mitchell's statistics.
Electricity generation was more dramatic still. In 1913, Germany produced 8.0 TWh, Britain 2.5 TWh. A factor of 3.2. Germany's share of the global synthetic dye market reached 90%.
World trade share traced the same trajectory. Between 1880 and 1913, Britain fell from 38.2% to 30.2%, while Germany rose from 17.2% to 26.6%. The GDP ranking itself reversed. According to Angus Maddison's estimates (in 1990 international dollars), in 1870 Britain's GDP was $100.2 billion, Germany's $72.1 billion. By 1913, Britain stood at $224.6 billion and Germany at $237.3 billion. Germany was ahead.
Demographics played a role too. In 1914, Germany's population was 65 million, 50% larger than Britain's. But demographics alone do not explain it. Per capita GDP in 1913 was still higher in Britain ($4,921) than in Germany ($3,648). The wealthier side did not win. The side that read the shift in the technological regime and aligned its institutions to it did.
The Mirage of Inertia
Paul Kennedy's central thesis applies here: "The rise and fall of great powers is a process in which changes in the economic base are reflected, with a time lag, in shifts in relative military position."
In 1913, Britain still held the advantage on many metrics. The world's largest navy. Maritime dominance built on the Two-Power Standard. The world's largest stock of overseas investment, some 4 billion pounds. The world's preeminent financial center in the City of London. An empire covering 25% of the world's land area.
But all of this was inertia. The future of production already lay on the other side of the Rhine. The momentum of finance and military power merely obscured the loss of industrial supremacy. Alfred Chandler labeled Britain of this period "personal capitalism." Family-run firms were reluctant to scale. Germany's "cooperative capitalism" pursued "Scale and Scope" from the outset.
In today's AI competition, is the United States Britain or Germany? Could the United States, with its massive investments in existing cloud and semiconductor infrastructure, fall prey to the "penalty of the pioneer"? Or does Silicon Valley's venture capital and university system still possess the institutional innovation capacity that Germany's Technische Hochschule once provided? What structural similarities does China's state-led AI strategy share with Germany's catch-up model?
The answers to these questions remain open. One thing is certain: throughout history, the first mover's position was never guaranteed. The side that rode the next wave won.
Section C: The Side That Can Diffuse Wins, Not the Side That Can Develop — US vs. USSR
Glushkov's Dream
1969, Kyiv. Viktor Glushkov, director of the Institute of Cybernetics, brought an ambitious plan to Moscow: OGAS — the All-State Automated System (Общегосударственная Автоматизированная Система). A computer network that would collect production data in real time and optimize central economic planning. Factory inventories, logistics flows, resource allocation — all integrated into a single system. Had it been realized, it could have been the Soviet version of the internet.
But officials at the Central Statistical Administration and the Ministry of Finance opposed it. The reason was not technical incapacity. Soviet engineers were capable of designing the system. The problem lay elsewhere. The horizontal flow of information threatened the power structure of the existing bureaucracy. If production data, monopolized by each ministry, were exposed to other ministries through a network, it would become harder to falsify plan-fulfillment figures. Information was power. No bureaucrat would accept a proposal to share it.
OGAS was killed. Benjamin Peters reached this conclusion: "The Soviet failure to build the internet was not a failure of engineering but a failure of institutions."
That single sentence summarizes the technology competition of the US-Soviet Cold War.
Same Inventions, Different Trajectories
The Cold War's paradox: the United States and the Soviet Union developed the same core technologies at nearly the same time.
Nuclear technology. The Soviet Union succeeded in its nuclear test in 1949, four years after the United States. Both nations achieved nuclear deterrence, and in this domain the Soviets reached parity. In space technology, the Soviets led first: Sputnik in 1957, Yuri Gagarin's first manned spaceflight in 1961. The United States reversed the lead with Apollo 11 in 1969, but Soviet early superiority was real. Basov and Prokhorov, who provided the theoretical foundation for the laser, were also Soviet scientists.
On the dimension of technology development capability alone, the Soviet Union was not behind the United States. In the 1980s, Soviet R&D personnel numbered 1.5 million. Annual invention certificates issued ran to 80,000, comparable to the 70,000-80,000 patents issued annually in the United States. R&D spending as a share of GDP was 3.5-4.5%, actually higher than America's 2.6-2.8%.
But the way technology diffused through society was entirely different.
600 to 1
The numbers speak. According to Manuel Castells' analysis, in 1985 the United States had 30 million installed computers, personal PCs that had spread into homes and offices. That same year, the Soviet Union had 50,000 computers, most of them government mainframes. The gap was 600 to 1.
In 1970, the ratio was 100,000 to 5,000 — a gap of 20 to 1. Slava Gerovitch's data. In fifteen years, the gap widened from 20x to 600x. While the PC revolution was unfolding in America, computers in the Soviet Union remained the exclusive property of state institutions.
The internet widened the gap to infinity. In 1981, the United States had 213 ARPANET nodes. The Soviet Union had zero. That zero says more than any other number. Not because the technology was absent. As Glushkov's OGAS demonstrated, the Soviets could conceive of a network. But they could not permit one. The free flow of information was incompatible with the logic of the system.
A System That Could Concentrate but Could Not Diffuse
Mark Harrison's analysis strikes at the core: "The Soviet system excelled at single-objective optimization."
Build a nuclear weapon. Concentrate resources at KB-11. Launch a spacecraft. Funnel the budget to OKB Kuznetsov. For projects with clear objectives where resources could be concentrated in one place, the central planning system performed brilliantly. The T-34 tank, the MiG fighter jet. The Soviet Union demonstrated exceptional capability in resource concentration toward single objectives.
But the diffusion of innovation is an entirely different kind of problem. Reading the demands of millions of consumers, thousands of firms competing, technology finding applications in unpredictable markets — this cannot be commanded from the center. It requires a feedback loop: price signals allocating resources, failed firms exiting, successful firms being imitated.
The Soviet Union had no such loop. The incentive for enterprise managers was plan fulfillment, not innovation. Innovation was a risk. Try a new method, fail to meet the plan target, and you face punishment. Joseph Berliner diagnosed this structure in a single line: "The fundamental problem of the Soviet innovation system was not invention but innovation and diffusion."
According to Matthew Evangelista's estimates, the military share of Soviet R&D spending was 70–75%. In the United States, it was 30%. Most Soviet technology remained locked within the military domain. There was no mechanism for military-to-civilian spin-off. In a system that deliberately restricted civilian distribution of photocopiers and fax machines, the social diffusion of information technology was structurally impossible.
America's Conversion Ecosystem
The United States extracted overwhelming economic advantage from the same military technologies. The secret lay in its conversion ecosystem.
The DARPA model was the keystone: a distributed investment structure in which military research funding flowed to civilian universities and firms. The model Mariana Mazzucato named the "entrepreneurial state." ARPANET became the internet. The military satellite navigation system became GPS. According to a 2019 RTI International estimate, the annual economic value of GPS is $1.4 trillion. That is the civilian economic value generated by a single military project.
Technologies originating at DARPA passed through university labs and into startups. Fairchild spawned Intel, and Intel spawned the ecosystems of Apple and Google. AnnaLee Saxenian's analysis of Silicon Valley's "regional industrial system" shows that the pathway from military research personnel to civilian entrepreneurship was institutionally guaranteed. The Bayh-Dole Act (1980) allowed universities to patent technologies developed with federal research funding, creating the link between research and commercialization. The "Triple Helix" named by Henry Etzkowitz and Loet Leydesdorff — the dynamic connection among university, industry, and government — was the institutional foundation for innovation diffusion.
In the Soviet Union, all of these pathways were blocked. A rigid wall stood between military and civilian sectors. The Soviet decision in the 1960s to clone the IBM 360 (the ES EVM series) was a choice to abandon indigenous innovation in favor of imitative catch-up. As Gerovitch puts it: "The Soviet Union perceived computers as tools for planned economics, not as platforms for distributed innovation."
Loren Graham's Thesis
MIT historian of science Loren Graham compressed this paradox into a single sentence:
"Russia and the Soviet Union historically produced brilliant scientific ideas but repeatedly failed to convert them into economic value. The problem was not the quality of the ideas but the absence of institutional pathways for those ideas to diffuse through society."
Soviet scientists co-invented the laser. The United States commercialized it. The Soviet Union was the first in the world to industrialize synthetic rubber. Subsequent innovation was led by the United States. The Soviets had independent research capability in semiconductors. But the United States dominated large-scale production and application.
Possessing a technology and making that technology work across an entire society are fundamentally different problems. That is the lesson of the third precedent.
But China Is Not the Soviet Union
A warning is necessary. When applying these precedents to the AI era, resist the simplistic equation "China = Soviet Union." The analogy is seductive but dangerous.
The Soviet Union's core weakness was the absence of a civilian market. There was no mechanism for reading consumer demand, and competition between enterprises did not exist. The pathway for technology to leave the laboratory and spread through society was structurally blocked.
China is different. It has a massive civilian market of 1.4 billion people. A fiercely competitive corporate ecosystem exists, with ByteDance, Alibaba, Tencent, and Baidu battling for dominance. DeepSeek itself is a private startup that originated from the hedge fund High-Flyer. Its founder, Liang Wenfeng, is a mathematician and former quant fund manager. China has partially overcome the "diffusion failure" that was the Soviet Union's core weakness. China's AI models are not locked in laboratories. They are deployed to hundreds of millions of users, receive market feedback, and are rapidly iterated.
That said, how the institutional constraints of information control, data sovereignty, and algorithmic regulation will affect the diffusion of AI innovation remains an open question. Just as the Soviet Union controlled photocopiers, the Chinese government controls the flow of information. The method differs — China does not block the internet but manages its content. But how information control will ultimately reshape the speed and direction of innovation diffusion remains open. This book pursues that question in its later chapters.
Connection to Volume 1: From Inside to Outside
Volume 1, The Displaced and the Discerning, traced the shockwaves that productivity revolutions send through the interior of a society — the latifundium displacing the smallholder, Arkwright's factory displacing the handloom weaver, AI displacing the translator and the paralegal. The core formula Volume 1 extracted was: Technology → Capital concentration → Social instability → Institutional redesign.
Volume 2 extends the question to the dimension between nations. How does the same productivity revolution overturn the board of hegemonic competition?
When Volume 1 examined Rome, the lens was internal: the decline of the smallholder, the expansion of the latifundium, the attempted reforms and defeat of the Gracchi brothers. This chapter reads the same era, the same Rome, through a different lens — Rome's external competition, its collision with a fundamentally different system called Carthage. Where Volume 1 saw the displaced within Rome, Volume 2 sees the displaced system outside Rome.
Volume 1's core formula described a dynamic operating within a single society. Volume 2 traces how that formula operates between two societies. When the same productivity revolution (AI) is injected into two entirely different institutional systems, does it produce the same outcome or a different one? That is the question running through this entire book.
Transition: History Does Not Give Answers — It Gives Questions
The Common Language of Three Precedents
The same pattern recurs across all three precedents: "Institutional Absorptive Capacity." Developing a technology and making that technology work across an entire society are different things. What the winners possessed was not more money or superior inventions, but institutional pathways for absorbing technology.
| Precedent | Productivity Revolution | Winner's Institutional Absorption Pathway | Loser's Institutional Bottleneck |
|---|---|---|---|
| Rome vs. Carthage | Standardization of infrastructure and law | Roads, ius gentium, citizenship expansion | Mercenary dependence, oligarchic decision-making |
| Britain vs. Germany | Second Industrial Revolution | Technical universities, bank-industry integration, state strategy | First-mover lock-in, classical education, laissez-faire |
| US vs. USSR | Information technology revolution | DARPA, VC, university-industry linkage, market feedback | Central planning, information control, military-civilian wall |
In all three cases, the final verdict in the hegemonic contest was decided not on the military dimension but on the economic one. Military superiority without an economic foundation to sustain it is nothing more than inertia.
What Investors Should Note
In the hegemonic competition of the AI era, the decisive factor is not the superiority of AI weapons systems but the social diffusion of AI productivity. The side with higher "institutional absorptive capacity" holds the advantage. This absorptive capacity is a function of five variables: the education system, the degree of freedom in information flows, market feedback mechanisms, military-to-civilian conversion pathways, and regulatory flexibility. Track these five variables, and the structural trajectory of the US-China competition comes into view.
A Bridge to the Questions
History does not give answers. It gives questions.
What the three precedents yield is not a prediction but an analytical framework. The Rome-Carthage precedent asks: Between America's alliance network (NATO, AUKUS, the semiconductor alliance) and China's bilateral relationships (Belt and Road, BRICS+), which model provides greater resilience? The Britain-Germany precedent asks: Is DeepSeek's achievement of comparable performance for $6 million a signal that the late starter's "blank slate" advantage is at work? The US-USSR precedent asks: Can a vast civilian market and an information-control regime coexist?
When the same technology is injected into two fundamentally different institutional systems, does it produce a different outcome? That substitution begins in the next chapter.