Cold War 2.0: Why It Probably Ends Without a Hot War — and Who Gets Rich on the Way

The thesis I want to defend in this piece is simple, and possibly wrong: the rivalry between the United States and the People’s Republic of China — what people are starting to call Cold War 2.0 — will end the same way the original ended. Not with a Pacific equivalent of Stalingrad. Not with a mushroom cloud over the Taiwan Strait. But with a multi-decade strategic competition that bleeds out economically, enriches defense contractors and tech primes on both sides, and is eventually decided by demographic and economic arithmetic rather than by combat.

If that view is correct, the most important question isn’t who wins the war — there isn’t one — but who profits from the cold one.

This is the long version of the argument. I’ll walk through the parallels (espionage, proxy wars, the strategic technology, the bloc structure), the chip-war theater (Taiwan, TSMC, ASML), the case for and against my thesis, and what the original Cold War’s economic record actually looked like. I’ll cite the work where I’m leaning on someone else’s data — including specific reporting from Politico, Bloomberg, The Economist, CSIS, RAND, CFR, CNAS, SemiAnalysis, MIT Technology Review, War on the Rocks, the Quincy Institute, the Stimson Center, Foreign Affairs, the original work of Graham Allison, and — most heavily — Chris Miller’s Chip War (Scribner, 2022), which is the single best historical source on the parallels between the Soviet semiconductor program and the contemporary Chinese one.

Table of contents

I. The shape of the analogy

Cold War 1.0 had five structural features: a strategic technology that compressed conventional advantage into a single weapon (the bomb), two ideologically-opposed superpowers locked in non-kinetic competition, an espionage war on the home front, proxy wars on the periphery, and an economic decoupling that was always less complete than it sounded. Cold War 2.0 has all five.

The single deepest parallel is not on this table: it is that a single technology has become the centerpiece of national power, and the country that masters it first — and most safely — sets the terms for the rest of the century. In 1945 that was nuclear fission. In 2026 it is the joint capability of training, deploying, and aligning frontier AI models at scale, with semiconductors as the upstream bottleneck.

This is not a soft analogy. It is a structural one. The reason Klaus Fuchs and Su Bin matter in the same paragraph is that both worked the same lever: when you cannot close a generational gap organically, you steal it.

II. Espionage rhymes: from Klaus Fuchs to Su Bin

The Soviet playbook for technology acquisition was systematic, well-funded, and astonishingly successful — until it ran into one specific industry where it broke down completely. Klaus Fuchs handed implosion-bomb design to the NKVD from inside the Manhattan Project. Theodore Hall and the Rosenbergs added detail. The KGB’s Directorate T, operating through field officers known as Line X, ran an industrial espionage program whose scope only became clear in 1981 when Colonel Vladimir Vetrov — code-named Farewell — handed French intelligence a complete index of what the Soviets had stolen. The Farewell Dossier showed the USSR had quietly absorbed Western radar, computing, machine-tool, and avionics designs at industrial scale.

The Rolls-Royce Nene engine episode (1946) is worth pausing on, because it is one of the cleanest historical parallels to what is happening with US chips today. The British government, in a moment of postwar generosity that aged poorly, sold a small batch of Nene jet engines to the Soviet Union for “research.” The engines were reverse-engineered, productionized as the Klimov VK-1, and propelled the MiG-15 — the aircraft that would shoot down American B-29s over Korea four years later. A single export license, granted because the strategic stakes were not yet visible, accelerated Soviet jet capability by an estimated decade.

But the deepest parallel — and the one most directly relevant to the AI chip war — is something Chris Miller documents in Chip War (Chapter 8, “Copy It”). In the late 1950s, a Soviet semiconductor researcher named Boris Malin returned from a year studying in Pennsylvania carrying a Texas Instruments SN-51 integrated circuit in his luggage. Alexander Shokin, the Soviet minister of microelectronics, summoned Malin and his team, placed the chip under a microscope, and gave the order: “Copy it. One-for-one, without any deviations. I’ll give you three months.”

Miller’s analysis of why this strategy failed is the single most important piece of historical pattern-matching for understanding why the chip war is, today, China’s hardest technology problem:

“Spying could only get Shokin and his engineers so far. Simply stealing a chip didn’t explain how it was made, just as stealing a cake can’t explain how it was baked… Soviet spies were among the best in the business, but the semiconductor production process required more details and knowledge than even the most capable agent could steal. Moreover, the cutting edge was constantly changing, per the rate set out in Moore’s Law… No other technology moved so quickly — so there was no other sector in which stealing last year’s design was such a hopeless strategy.”

The Soviet semiconductor program had the country’s best scientists and stolen American secrets, and it still fell terminally behind. Some Soviet chipmaking machinery used inches rather than centimeters — even though the rest of the USSR was metric — solely to better replicate American designs. Career advancement at Zelenograd (the Soviet Silicon Valley) required “becoming a better bureaucrat, not devising new products,” in Miller’s phrase. The “copy it” mentality “literally hardwired” the pathway of innovation in Soviet semiconductors to whatever the Americans had done two years earlier.

This is the analogy I would press hardest on. The contemporary Chinese semiconductor program has roughly Shokin’s structural problem, with three modifications that make it both more and less serious than the Soviet version:

• More serious for the West: China has a real consumer-electronics export base, which gives its chip industry both demand pull and feedback loops the Soviets never had. SMIC and Huawei produce things people want to buy, which makes them harder to ignore than Zelenograd.
• Less serious for the West: the structural diagnosis is the same — copying does not scale to a Moore’s Law industry, especially when the chokepoint (EUV) is not a chip but a tool. SMIC’s 7nm process via DUV multi-patterning is a remarkable achievement, but, as SemiAnalysis documents, the per-wafer cost and yield suggest a process running at perhaps 30–40% of TSMC’s economics — a tax the Chinese state is willing to absorb but not one that produces a competitive commercial product.

The contemporary playbook is operationally similar but technologically more modern.

Documented cases on the Chinese side:

• Su Bin (2014–2016) pleaded guilty in US federal court to conspiring with PLA officers to exfiltrate F-22, F-35, and C-17 design data from Boeing and Lockheed networks. Compare external photographs of Chengdu Aerospace’s J-20 and Shenyang’s J-31 with the F-22 and F-35 respectively; the structural inheritance is hard to miss.
• APT1 / PLA Unit 61398, exposed in Mandiant’s 2013 report, conducted systematic exfiltration from 141 organizations across 20 industries over seven years.
• APT10 / “Cloud Hopper” (DOJ 2018 indictment) was an MSS-linked group that hit managed service providers as a way to reach defense, aerospace, and biotech customers downstream — a supply-chain pivot the Soviets never had.
• Fujian Jinhua / Micron (2018) was a DRAM IP-theft case so brazen that the US imposed its first major export ban on a Chinese chipmaker in retaliation.
• The Thousand Talents Plan has functioned as a parallel to the Soviet recruitment of sympathetic Western scientists — but, crucially, operating openly through academic incentive structures rather than through clandestine handlers. Several US prosecutions under the Trump-era “China Initiative” attempted to characterize undisclosed Thousand Talents participation as material fraud; the legal record is mixed, but the structural intent is clear.
• APT41 (DOJ 2020 indictment of seven Chinese nationals) is dual-use: it conducts state-directed espionage during the day and financially-motivated cybercrime at night, a combination that has no clean Soviet analog.

The structural parallel is that both regimes treated stolen IP as legitimate state policy, and both fused intelligence services with industrial capacity. The Soviets had Line X funded through the KGB; China has Section 5 of the MSS, the PLA’s Strategic Support Force, and a network of state-affiliated “civilian” research institutes whose civil-military fusion is doctrinal, not coincidental.

The asymmetric difference is scale and speed. Soviet exfiltration moved through dead drops and microfilm; Chinese exfiltration moves through fiber optics. The bandwidth of theft has increased by something like nine orders of magnitude. CNAS has documented in detail how Chinese state-backed actors have used this bandwidth to compress technology absorption timelines that took the Soviets a decade into windows of months (CNAS, “U.S.-China Competition and Military AI”).

III. Proxy wars and the CRINK axis

The original Cold War was decided largely on its peripheries: Korea (1950–53), Vietnam (1955–75), Afghanistan (1979–89), Angola, Nicaragua, the Horn of Africa, Cuba. The superpowers fought through proxies because they could not afford to fight each other directly. The same dynamic is now playing out across what analysts have begun calling the CRINK axis — China, Russia, Iran, North Korea — and the Western alliance.

Ukraine is the most direct proxy war of the new era, and one in which the bloc structure has clarified faster than most expected. Russia provides the troops. China provides the economic lifeline (sanctioned oil purchases, dual-use exports, financial channels through CIPS and yuan settlement). Iran ships Shahed-136 drones by the thousand, with technology transfers running both directions. North Korea, in late 2024, sent an estimated 10,000 troops plus large quantities of artillery shells to the Russian front — the first cross-bloc deployment of foreign combat troops in Europe in eighty years.

Iran sits in the structural position that Cuba and North Vietnam jointly occupied during the original Cold War: not a peer power, not formally a treaty ally of either Beijing or Moscow, but a deniable forward operating base for the larger anti-Western bloc. Two specific commitments anchor this:

• The Iran-China 25-year Comprehensive Cooperation Agreement (signed March 2021, ~$400B framework) has functioned as an economic bypass for US sanctions. China buys, by various estimates, roughly 90% of Iran’s exported oil despite the formal sanctions regime.
• Iran’s regional proxy network — Hamas (Oct 7, 2023), Hezbollah, the Houthis disrupting Red Sea shipping (which carries roughly 12% of global trade), and Iraqi Shia militias — has effectively externalized the cost of US deterrence into a half-dozen simultaneous regional theaters. The Pacific Command’s outgoing commander stated in April 2026 that “victory over Iran” was a precondition for credibly deterring a Chinese move on Taiwan, an extraordinary admission that the regional theaters are explicitly linked.

Taiwan and the South China Sea are the closest analog to the Cold War’s Fulda Gap — the place where conventional deterrence is being most actively tested. A 2024 Bloomberg Economics analysis estimated that a US-China conflict over Taiwan would cost the global economy approximately $10.6 trillion in the first year alone, equal to roughly 9.6% of global GDP — a figure larger than the combined impact of COVID-19 and the 2008 financial crisis (Bloomberg, January 2024; updated February 2026).

The Bloomberg figure is not a casual one. It is the kind of number that, once it lands inside both Treasury and the National Security Council, fundamentally changes the math of escalation. We will return to it.

IV. Taiwan and the silicon shield

The “silicon shield” thesis was coined in Taipei in the 2000s and runs roughly as follows: Taiwan is so structurally central to the global economy — through its monopoly on leading-edge semiconductor manufacturing — that no rational Chinese leader would order an invasion that destroyed the very prize they sought. The thesis is compelling, partially true, and increasingly fragile.

The numbers, from the most current sources I can find:

• TSMC produces approximately 90% of the world’s leading-edge logic chips (sub-7nm) and around 92% of the world’s most advanced semiconductors (MIT Technology Review, August 2025).
• TSMC holds roughly 60%+ of global pure-play foundry revenue.
• Every Apple A- and M-series chip, every NVIDIA H100/H200/B200, every AMD MI300, and every Qualcomm flagship Snapdragon is fabricated on TSMC’s N5/N4/N3 nodes.
• US customers account for ~70% of TSMC’s revenue. Apple alone is roughly a quarter.
• N2 (2nm), with gate-all-around transistors, is ramping at the Hsinchu and Kaohsiung fabs through 2025 and into 2026.

Why the shield is more fragile than the headline numbers suggest:

The most advanced fabs are concentrated on a single island, within ~100 miles of PLA missile bases. A blockade — not even an invasion — could halt deliveries in weeks. Modern fabs are not easily destroyed-and-rebuilt: a leading-edge fab takes ~5 years and ~$20 billion to construct, and depends on EUV tools that ASML ships globally at a rate of roughly 50 machines per year. If you took TSMC offline by any mechanism — kinetic, blockade, or sabotage — there is no spare capacity in the world that could replace it before the global economy seized up.

The diversification hedge:

Note carefully what is not on this list: the bleeding edge stays in Taiwan. Arizona will run a generation behind Hsinchu indefinitely. That is a deliberate political choice by Taipei — the silicon shield only deters if the most valuable silicon stays on the island. The Stimson Center has flagged this dynamic as one Taipei worries about most under “America First” framing, where US pressure to onshore the leading edge is seen as actively eroding deterrence (Stimson, 2025).

Will the shield prevent war? The CFR’s Sebastian Mallaby argues — correctly, I think — that the shield raises the cost of force without diminishing the desire for unification. China can choose to absorb a one-time economic shock for what it views as a permanent strategic gain. The shield is a tax, not a fence.

V. ASML: the chokepoint nobody can replicate

If the chip war has a single physical chokepoint, it is a Dutch company headquartered in a town that most Americans cannot pronounce.

ASML is the sole manufacturer in the world of extreme ultraviolet (EUV) lithography systems. The monopoly is not 90% — it is 100%. There is no second source, never has been, and will not be one within any planning horizon that matters.

What is rarely appreciated is that the EUV monopoly almost did not exist at all. Chris Miller documents in Chip War that EUV’s commercial existence is the result of a nearly-30-year, multi-decade gamble largely seeded by a single bet: in the early 1990s, Andy Grove at Intel committed approximately $200 million of internal research funding to EUV at a time when the technology’s viability was deeply uncertain. ASML, then a small Dutch joint venture spun out of Philips, became the eventual production beneficiary. Without that initial Intel-led consortium, and the subsequent multi-billion-dollar customer co-investments by TSMC, Samsung, and Intel, the world would either not have EUV today or would have a Japanese-or-Chinese-built version of it. The fact that the chokepoint sits in the Netherlands, owned by an alliance of the West’s three biggest chipmakers, is a contingent historical outcome — one that the export-control regime is now levered against.

• An EUV machine (NXE:3800E) costs ~$200M**. The newer **High-NA EUV** (Twinscan EXE:5000) is **~$380M+.
• Each machine contains ~100,000 parts, weighs ~250 tons, ships in ~40 freight containers, and requires ~3 Boeing 747s to deliver.
• Inside, ~50,000 droplets of tin per second are vaporized by a high-power CO₂ laser to produce 13.5nm wavelength light, reflected by Carl Zeiss optics that, if scaled to the size of Germany, would have a maximum surface deviation of roughly 0.1mm.
• ASML produces approximately 50 EUV machines per year.

The supply chain inside ASML’s supply chain:

• Cymer (San Diego, ASML subsidiary since 2013) — the light source.
• Carl Zeiss SMT (Oberkochen, Germany) — the optics; ASML owns 24.9% of Zeiss SMT.
• Trumpf (Germany) — the high-power CO₂ lasers.
• VDL, Phenom, and dozens of Dutch suppliers in the Brainport Eindhoven cluster.

You cannot replicate ASML by spending money. You would have to replicate Zeiss, Cymer, Trumpf, and approximately 30 years of accumulated tacit process knowledge that lives in the heads of engineers in Veldhoven, Oberkochen, and San Diego.

The vise ASML sits in:

Former ASML CEO Peter Wennink said the quiet part loud in 2024:

“The more you put them under pressure, the more likely it is that they will double up their efforts… Thirty years from now they’ll have developed those technologies themselves.”

His successor Christophe Fouquet has been more diplomatic but the strategic stance is unchanged. The Dutch government itself has publicly worried that ASML is becoming “a tool in the trade war between the U.S. and China” rather than a Dutch firm operating under Dutch law.

Why the Netherlands matters geopolitically out of proportion to its size:

The Wassenaar Arrangement — the multilateral export-control regime that replaced CoCom — requires consensus, which means it is slow and leaky. The US has therefore relied on bilateral arm-twisting of The Hague and Tokyo, the two governments that house the irreplaceable tool makers (ASML, Tokyo Electron, Lasertec). The 2023 Dutch and Japanese restrictions were not formally part of US BIS rules; they were negotiated in parallel (SCMP). This is the closest modern analog to CoCom — and just as CoCom worked imperfectly, this regime leaks too. The Huawei Mate 60 Pro shipping with an SMIC 7nm Kirin 9000S chip in August 2023, made with deep-ultraviolet multi-patterning rather than EUV, was the first dramatic public sign of leakage. SemiAnalysis has done the most rigorous open-source modeling of how far that leakage now extends — projecting SMIC’s advanced node (7nm-and-below) capacity at roughly 45,000 wafers per month by end-2025, 60,000 by 2026, and 80,000 by 2027 in the absence of further controls.

VI. The full chokepoint stack

The whole production chain for a leading-edge chip looks like this:

• Lithography: ASML (Netherlands, EUV monopoly); Nikon, Canon (Japan, DUV only, lagging).
• Deposition / Etch: Applied Materials, Lam Research (US); Tokyo Electron (Japan).
• Metrology / Inspection: KLA (US); Lasertec (Japan, sole producer of EUV mask inspection tools).
• EDA software: Cadence, Synopsys, Siemens EDA (US-led).
• Wafers: Shin-Etsu, SUMCO (Japan) — together ~60% of global silicon wafers.
• Photoresist: JSR, Tokyo Ohka, Shin-Etsu (Japan) — ~90% of EUV photoresist.
• Masks and pellicles: Hoya, Mitsui (Japan).

Look at the country distribution: United States, Netherlands, Japan, Taiwan, South Korea. Five US-aligned democracies hold every single chokepoint. China holds none of them. That is the entire architecture of the chip war. It is also the reason the chip war is fundamentally a negotiation among allies about how hard to squeeze, not a fight between the US and China about what is technically possible.

VII. The thesis: why this probably ends without a hot war

Here is where I have to make the affirmative case. The argument has three legs.

Leg 1: Mutual economic destruction is now total

The original Cold War’s Mutual Assured Destruction ran on warheads. Cold War 2.0’s runs on supply chains. A Chinese move against Taiwan — even a blockade — would obliterate not only the global semiconductor supply but, by second-order effects, automotive production, consumer electronics, advanced industrial equipment, defense systems, and the entire AI infrastructure buildout. The Bloomberg model puts the first-year hit at $10.6 trillion. China itself absorbs an estimated 16.7% GDP loss in that scenario; the US absorbs 6.7%.

But here is the part that gets under-appreciated: China’s economic position is currently weaker, in relative terms, than the Soviet Union’s was at any point before 1985. Property-sector deleveraging, demographic decline (China’s working-age population peaked in 2014), youth unemployment that the National Bureau of Statistics stopped publishing in 2023 after it printed at 21.3%, and a growth rate that has structurally downshifted from double digits to roughly 4–5% — these are not the conditions under which a leader rationally chooses to rupture the global trading system in which his country is the largest exporter.

Foreign Affairs’ Hal Brands and Michael Beckley have argued — controversially but, I think, correctly — that the peak power dynamics matter as much as the rising power dynamics in Allison’s Thucydides framework: that when a rising power begins to slow before it has displaced the incumbent, it can become more aggressive, not less, because the window is closing (Foreign Affairs, “The Real Thucydides Trap”). This is the strongest version of the case against my thesis. I will steelman it in the next section.

Leg 2: Chinese leadership has consistently chosen patience over kinetic action

Forecasters surveyed by the Swift Centre put the probability of a Chinese blockade of Taiwan by mid-2027 at roughly 9% (range 4.9% to 15%) — and a blockade is the lower-cost option. CSIS reports that 83% of China experts surveyed reject the proposition that China plans kinetic action against Taiwan by 2027. RAND’s wargaming has steadily revised toward greater uncertainty about a Chinese victory, especially after observing the Russian military’s performance in Ukraine (RAND, “Thinking Through Protracted War with China”).

What Xi Jinping has actually said is that the PLA must be capable of taking Taiwan by 2027. Capability and intent are not the same thing. The CFR has argued (CFR Council Special Report on Taiwan) that Beijing continues to prefer political and economic coercion over military action, and that force remains a last resort — albeit one Beijing wants the option for.

Leg 3: The deep economic interdependence has no Cold War 1.0 analogue

Annual US-China bilateral trade is on the order of $650 billion. US-Soviet trade at its 1979 peak was roughly$4.5 billion (a number so small that Cold War historians often omit it). Apple’s iPhone supply chain runs through Zhengzhou. NVIDIA’s H100 packaging passes through TSMC and downstream to Chinese assembly. CATL is a top-five battery supplier to Western automakers. Tesla’s Shanghai Gigafactory is the company’s most efficient. American institutional investors hold tens of billions of dollars of Chinese sovereign and corporate debt; Chinese state-affiliated funds hold even more US Treasuries.

You cannot decouple economies this entangled in less than a decade, and any attempt to do so abruptly would produce financial-system stress on both sides that no leader actually wants to own. This is why the rhetoric is “de-risk, not decouple” (Ursula von der Leyen, 2023), and “small yard, high fence” (Jake Sullivan, 2022). Both phrases are admissions that a clean break is not on the table.

VIII. The strongest case against my thesis

Honest argumentation requires steelmanning the other side. Here is the case that I am wrong:

1. Wars start when leaders miscalculate, not when they correctly compute the cost. Bloomberg’s $10.6T figure assumes both sides are rational. The Russian invasion of Ukraine in February 2022 was a useful reminder that authoritarian leaders insulated from honest information can make catastrophically wrong calls.
2. The Davidson Window — Admiral Phil Davidson’s 2021 testimony that the PLA has been told to be ready to seize Taiwan by 2027 — is real, even if “ready” is not the same as “ordered.” A regime that builds the capability often eventually uses it.
3. Allison’s base rate is not encouraging. Out of 16 historical cases of a rising power confronting a ruling one, 12 ended in war. That is a 75% base rate. To bet against war, you have to argue the four exceptions are more representative than the twelve.
4. The peripheral wars could escalate. A Houthi anti-ship missile that hits a US carrier, or a North Korean miscalculation on the DMZ, or an Iranian-backed strike that kills US personnel in Syria, could trigger an unintended escalation that pulls the principals in.
5. Taiwan’s “America First” anxiety, documented by the Stimson Center, is the inverse of the silicon shield argument: if Taipei stops trusting US deterrence, it may move toward outright independence declaration, which is the one move that Beijing has explicitly committed to treating as casus belli.

I do not dismiss any of these. My honest probability estimate for a hot war between the US and China at any point in the next twenty years is roughly 15%. That is high enough to take seriously and low enough that it is not the modal outcome.

VIII-bis. The Ogarkov moment: when chips became strategic

Before turning to the defense bonanza, one historical detail from Chip War deserves a short section of its own — because it is the cleanest model for what is happening to Beijing right now.

In the late 1970s and early 1980s, Soviet Marshal Nikolai Ogarkov — chief of the general staff of the Soviet military — became fixated on what he called a “military-technical revolution.” He saw clearly that “long-range, highly accurate, terminally guided combat systems, unmanned flying machines, and qualitatively new electronic control systems” would transform conventional explosives into what he termed “weapons of mass destruction.” He recognized that the United States’ lead in microelectronics was not a peripheral commercial advantage but a military-strategic one, and that the Soviet Union had no answer.

Miller’s data point is striking: by the mid-1980s, the American MX missile could land within 364 feet of its target 50% of the time. The roughly comparable Soviet SS-25 averaged twelve hundred feet. The difference was almost entirely a difference in the chips powering the guidance computers. The Soviets had more tanks, more artillery, more troops — and Ogarkov correctly recognized that none of it would matter against weapons that could think.

What Ogarkov is to the late-Soviet military, Xi Jinping’s “intelligentization” doctrine is to the contemporary PLA. The Chinese leadership has internalized exactly Ogarkov’s lesson — that the next strategic technology is microelectronics-and-AI, not steel — and is racing to avoid the Soviet outcome. The export controls are an explicit attempt to force China into the Soviet position: a country with stolen designs, capable physicists, and a chip stack that nonetheless lags by 2–3 generations indefinitely.

The question is whether the Ogarkov outcome is replicable. China has structural advantages the USSR never had: a real consumer-electronics market, world-class engineering universities, deep integration into the global supply chain (until recently), and a state that has been willing to absorb economic losses on the order of $100B+ to subsidize SMIC and Huawei. Whether those advantages are enough to bend the curve is, in my view, the single most important open question in geopolitics.

IX. The defense bonanza: who actually got rich during Cold War 1.0

The original Cold War’s economic record is the empirical foundation of my second claim — that the defense-industrial complex on both sides is the long-term beneficiary. The 44 years from 1947 to 1991 produced roughly $13.1 trillion in cumulative US defense spending (in 2010 dollars), substantially of which flowed to a small set of contractors: Lockheed, Boeing, General Dynamics, Northrop, Raytheon, Grumman, McDonnell Douglas, Hughes, and a long tail of subcontractors. The post-1990 consolidation reduced “primes” to single digits — Lockheed Martin (Lockheed + Martin Marietta + General Dynamics F-16), Boeing (Boeing + McDonnell Douglas + Hughes Helicopter), Raytheon Technologies (Raytheon + United Technologies), Northrop Grumman, and General Dynamics. This is the architecture that exists today.

What is happening in Cold War 2.0 to date:

The Quincy Institute’s Profits of War report documents that roughly half of Pentagon contracting spending — nearly $2.4 trillion between FY 2020–2024 — went to private contractors, with the top five primes capturing the largest share. Despite the US withdrawal from Afghanistan in 2021 and the formal scaling back of the Global War on Terror, Pentagon spending and contractor revenue have continued at extremely high levels — driven explicitly by “strategic competition with China.” This is the most important single sentence in the entire post.

The numbers:

• From 2021 through mid-June 2024, US venture capitalists invested approximately $130 billion into defense-tech startups — a 70%+ year-over-year increase sustained over three consecutive years.
• The 2026 NDAA funds the US Department of Defense at $901 billion**, with a **$10 billion Pacific Deterrence Initiative and an additional $12.6 billion specifically for surveillance of Chinese submarines and satellites (Bloomberg, Feb 2026).
• China’s announced 2026 defense budget rose 7% year-over-year to roughly 1.91 trillion yuan (~$277 billion), though the Pentagon’s annual report on Chinese military developments notes this is a substantial undercount.
• Lockheed Martin (LMT) and RTX (Raytheon) stocks gained 18–20% in 2023–2024, outpacing the S&P 500. Within 2025, L3Harris advanced 40% and Northrop Grumman gained 22%.
• Anduril entered the top 100 defense contractors at #74. Palantir at #96. SpaceX jumped from #53 to #28.

The pattern is the original Cold War rerun, with two important modifications:

1. The “primes” are joined by a new generation of defense-tech entrants — Anduril, Palantir, SpaceX, Shield AI, Skydio, Saronic, plus the AI labs that are increasingly contracting with the Department of Defense (OpenAI’s recent partnership with Anduril, Anthropic’s classified-network deployment with Palantir/AWS).
2. The civilian-tech sector is now de facto inside the defense base. Microsoft, Amazon (JEDI/JWCC), Google Cloud, NVIDIA, Oracle — all earn material revenue from DoD contracts. The category line between “tech company” and “defense contractor” that existed in 1985 has dissolved.

This creates what I’ll call the defense-tech industrial complex. It is structurally self-reinforcing in exactly the way Eisenhower warned about in 1961, but with an even broader political base, because it now includes the AI sector, the cloud-hyperscaler sector, and the venture capital firms that finance them. There is no Eisenhower equivalent in 2026 willing to deliver the parallel speech, because the political coalition supporting the buildup is now genuinely bipartisan and includes the Bay Area.

On the Chinese side, the equivalent enrichment is structurally similar but channeled through state-owned enterprises and the military-civil fusion doctrine: NORINCO, AVIC, CETC, CASIC, plus the “champion” private firms that Beijing has elevated (Huawei, SMIC, BYD, CATL, DJI). The economic surplus of the Chinese state is being routed to the same kinds of organizations that the Soviet system’s surplus was routed to, with the difference that these organizations produce competitive consumer goods alongside military hardware — a combination the Soviet planners never figured out.

X. Politico’s reporting: where the policy actually moves

The single most useful real-time tracker of how Cold War 2.0 is being prosecuted, on the US side, is Politico’s coverage of the Hill and Commerce Department. A few examples worth flagging:

• After DeepSeek’s January 2025 release of R1, Politico Pro reported that “China hawks urge White House to tighten chip export controls after DeepSeek” — the moment that made clear the export-control regime had a major gap and that Congress would respond.
• Politico Pro subsequently reported that the Trump administration widened the export-control blacklist to hit subsidiaries, closing a loophole that Chinese firms had used to acquire restricted equipment through nominal foreign affiliates.
• The bipartisan SAFE CHIPS Act, introduced by Senators Pete Ricketts (R) and Chris Coons (D), would require the Commerce Department to deny licenses for advanced AI chip exports to China, Russia, Iran, or North Korea for 30 months. The MATCH Act (Multilateral Alignment of Technology Controls on Hardware) would tighten coordination with allied tool-makers.

The structural pattern Politico’s reporting reveals is that Congressional China hawks have effectively taken away the Executive Branch’s discretion on chip export controls. The administration cannot give a single license to NVIDIA or AMD without a public fight. This is, again, structurally identical to the 1950s–1980s pattern: a bipartisan congressional consensus on the strategic competition that ratchets the controls in only one direction.

XI. The Economist’s framing and where I diverge

The Economist has been, on balance, the most clear-eyed Western publication on the chip war and Cold War 2.0 dynamics. Its consistent framing has emphasized:

• That the “small yard, high fence” approach is preferable to comprehensive decoupling.
• That China’s domestic semiconductor capacity is improving despite controls, and that the US should expect a shrinking — not widening — gap at the trailing edge.
• That Taiwan’s silicon shield is real but not infinite, and that diversification is rational risk management, not panic.
• That the actual probability of a hot war is lower than headline coverage suggests, but the long tail is unusually fat.

Where I would diverge from the Economist’s editorial line is on the question of who pays the cost. Their framing tends to treat the chip war as an abstract policy contest. My framing — closer to the Quincy Institute’s — is that the chip war is also a domestic political-economy contest, in which the defense and AI primes are the durable winners regardless of who occupies the White House or Zhongnanhai. The Economist tends to underweight the economic-incentive structure that locks the policy in place even when the underlying threat assessment changes.

XII. Where the Cold War 1.0 analogy actually breaks

It would be intellectually dishonest to leave you with only the parallels. The differences matter, and any honest version of this argument has to flag them.

1. Economic interdependence has no historical parallel. US-China trade dwarfs US-Soviet trade by orders of magnitude. The cost of decoupling is real and asymmetric.
2. The world is multipolar, not bipolar. India, the EU, the Gulf states, ASEAN, and Brazil are real third poles, not the 1955 Bandung non-alignment LARP. Decisions made in New Delhi, Brussels, Riyadh, and Brasília will shape the trajectory more than Cold War 1.0’s sideline countries did.
3. Tech diffusion is faster. Open-weight models (DeepSeek, Qwen, Llama, Mistral, Falcon) make AI proliferate faster than fissile material ever could. CSIS has documented how Chinese AI firms have used “workarounds” — algorithmic efficiency, alternative architectures, distillation from frontier models — to compress the apparent gap even when the underlying chips lag (CSIS, “DeepSeek, Huawei, Export Controls”).
4. The ideological binary is messier. “Democracy vs. authoritarianism” is uglier than “capitalism vs. communism” — many US-aligned states are not democracies, and several democratic states hedge between the blocs.
5. Defense industrial bases are weaker than they were. Both sides have hollowed out manufacturing capacity in ways that would matter immediately in protracted conflict. War on the Rocks has documented the US industrial base’s exposure to Chinese-controlled critical materials (tantalum, gallium, germanium, antimony, tungsten, magnesium, bismuth) in ways that simply did not exist in 1962.

XIII. The bottom line

The thesis I started with: Cold War 2.0 ends without a hot war between the principals, and the long-term economic beneficiaries are the defense-industrial complexes on both sides — joined now by a defense-AI complex that did not exist in Cold War 1.0.

I think this is more likely than not. Specifically:

• Probability of US-China hot war over the next 20 years: ~15%, with the tail driven by miscalculation, peripheral escalation, or a Davidson Window action in the late 2020s.
• Probability that defense and defense-tech contractors in the US continue to outperform broad equity benchmarks over the next decade: much higher than 50%, conditional on the cold war remaining cold. The historical record from 1947–1991 is unambiguous on this point.
• Probability that ASML, TSMC, NVIDIA, Lockheed, RTX, Anduril, Palantir, and the AI labs end the 2030s materially larger and more strategically central than they are today: very high.
• Probability that the chip-war regime “succeeds” in the strict sense of preventing China from reaching parity on frontier AI compute: much lower than the policy community publicly admits. The realistic outcome is that China remains 2–3 years behind on frontier capability indefinitely, at higher cost, in a partially-bifurcated ecosystem. That is not “containment” in the 1947 sense — it is a structural drag, not a blockade.

The Cold War 1.0 ended because one side ran out of money first. The Soviet defense burden, plus oil-price collapse, plus structural inefficiency, plus Afghanistan, plus the legitimacy crisis triggered by Chernobyl and Eastern Europe, broke the system from the inside. None of that is impossible for China — the demographic and property-sector dynamics are real — but it is also not imminent.

The most likely path is a long, expensive, multi-decade competition that enriches the defense and tech sectors of both blocs, gradually decouples the critical-technology supply chains, reorganizes global trade around two partially-distinct production ecosystems, and ends not with a battle but with one side eventually concluding the cost of competition has exceeded the value of the prize.

In other words: the cold war stays cold. Lockheed, RTX, Northrop, Anduril, Palantir, NVIDIA, ASML, and TSMC become the largest companies in the world. The Pentagon and the PLA grow indefinitely. And in 2050, some historian writes a book arguing that what looked like a confrontation was always, structurally, a transfer payment from taxpayers in both countries to the firms that build the weapons and the chips.

I would like to be wrong, but I think this is the most probable outcome.

Sources and further reading

Bloomberg

• “If China Invades Taiwan, It Would Cost World Economy $10 Trillion” — January 2024
• “The $10 Trillion Fight: Modeling a US-China War Over Taiwan” — February 2026
• “Pentagon Allocates $12.6 Billion to Boost Surveillance of China’s Military”

Politico

• “China hawks urge White House to tighten chip export controls after DeepSeek”
• “Trump administration widens export control blacklist to hit subsidiaries”

The Economist

• General coverage at economist.com — see the “Chaguan” column and the China and Business sections.

CSIS (Center for Strategic and International Studies)

• “Choking off China’s Access to the Future of AI”
• “DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race”
• “The Limits of Chip Export Controls in Meeting the China Challenge”
• “Blocking China’s Access to AI Chips Matters to U.S. National Security”
• “The U.S. Industrial Base Is Not Prepared for a Possible Conflict with China”

RAND Corporation

• “Thinking Through Protracted War with China: Nine Scenarios”
• “Could Sanctions Help Deter China from Attacking Taiwan?”
• “Denial Without Disaster — Keeping a U.S.-China Conflict over Taiwan Under the Nuclear Threshold”
• “Can Taiwan Resist a Large-Scale Military Attack by China?”

Council on Foreign Relations

• “Will China’s Reliance on Taiwanese Chips Prevent a War?”
• “The Next Taiwan Crisis Won’t Be Like the Last”
• “Council Special Report: The United States, China, and Taiwan — A Strategy to Prevent War”

CNAS (Center for a New American Security)

• “U.S.-China Competition and Military AI”
• “The Export Control Loophole Fueling China’s Chip Production”

SemiAnalysis

• “Huawei Ascend Production Ramp: Die Banks, TSMC Continued Production, HBM is The Bottleneck”

MIT Technology Review

• “Taiwan’s ‘silicon shield’ could be weakening”

War on the Rocks

• “These Materials Could Cripple America’s Defense Industrial Base”
• “A Guide to Refactoring the Defense-Industrial Base”

Quincy Institute for Responsible Statecraft

• “Profits of War: Top Beneficiaries of Pentagon Spending, 2020–2024”

Stimson Center

• “Why Taiwan Fears ‘America First’ Risks Eroding Its ‘Silicon Shield’”

Foreign Affairs

• “The Real Thucydides Trap” — Hal Brands & Michael Beckley
• “A Perfect Storm for Taiwan in 2026?”

Books

• Graham Allison, Destined for War: Can America and China Escape Thucydides’s Trap? (2017)
• Chris Miller, Chip War: The Fight for the World’s Most Critical Technology (2022)