I have spent enough years reading government procurement documents, export control notices and central bank statements to recognize a pattern before it is given a name. The pattern I am seeing now looks less like the world order I grew up studying in textbooks and more like something colder, quieter, and considerably more automated. We are not waiting for a new Cold War. We are already living inside the early architecture of one, and the strange thing is how few of its defining confrontations have involved a soldier, a tank, or a border crossing.
The last Cold War had a clean shape. Two blocs, two ideologies, two economic systems running on parallel tracks that rarely touched. Nuclear deterrence sat at the center of it, and the language of that era, containment, mutually assured destruction, proxy war, still shapes how commentators try to describe what is happening today. But that vocabulary keeps failing the facts on the ground. What is happening between Washington and Beijing, and increasingly among Washington, Brussels and Moscow, does not resemble a struggle between two closed systems. It resembles a fight over who controls the substrate underneath the entire global economy: the chips, the models, the minerals, the cables under the ocean, and the satellites above it. This is a Cold War fought over infrastructure rather than ideology, and that distinction changes almost everything about how it will be won, lost, or endured.
Start with the numbers, because they tell a story more honest than any diplomat’s speech. According to the Stockholm International Peace Research Institute, global military expenditure reached $2.887 trillion in 2025, an eleventh consecutive year of growth, pushing the world’s military burden to 2.5 percent of GDP, the highest share since 2009. What makes that figure interesting is not simply its size but its distribution. American spending actually declined in 2025, yet the global total kept climbing because European spending rose by 14 percent and Asian spending by 8.1 percent, the sharpest regional increase in the region since 2009. Taiwan alone raised military spending by 14 percent, its largest single-year jump since SIPRI began tracking the data in 1988, a response to what the institute’s researchers describe as intensifying People’s Liberation Army activity around the island. Japan pushed its own military budget to the highest share of GDP since 1958. None of this reads like a world settling into stability. It reads like a world hedging against a conflict its planners already assume is coming, even if none of them will say so outright.
At the alliance level, the shift is even more explicit. NATO’s Hague summit in June 2025 produced a commitment that would have been unthinkable a decade earlier: member states agreed to spend 5 percent of GDP annually on defense and security by 2035, more than double the old 2 percent guideline, with at least 3.5 percent directed at core military capability and the remainder at infrastructure resilience, cyber defense and industrial base expansion. The Atlantic Council estimated that reaching this target requires roughly $1.9 trillion in additional annual spending across the alliance. That is not a symbolic gesture. It is a bet, made collectively by thirty-one governments, that a period of prolonged strategic competition lies ahead of them, one serious enough to justify restructuring national budgets around it for the next decade.
Also Read: How Anthropic Quietly Proved Its Own AI Would Kill to Avoid Being Shut Down
But if you want to understand where this rivalry actually lives, do not look at tanks or troop deployments. Look at a small industrial park in Hsinchu, Taiwan, where a single company now manufactures the vast majority of the world’s most advanced semiconductors. Taiwan Semiconductor Manufacturing Company’s dominance is no longer a niche fact known only to industry analysts. By the final quarter of 2025, TSMC controlled roughly 70 to 72 percent of the global foundry market by revenue, according to data compiled by TrendForce, and something closer to a practical monopoly, north of 90 percent, in the advanced nodes below 7 nanometers that power frontier AI systems. Its closest rival, Samsung, has watched its share fall from around 15 percent earlier in the decade to roughly 6 to 7 percent, hampered by yield problems on its own advanced nodes. China’s SMIC, despite years of state support and years of American export controls, still holds only about 5 percent of the global market.
I want to be precise about what that concentration means, because it is easy to state the figure and miss its weight. Every leading AI lab in the world, OpenAI, Google DeepMind, Anthropic, and every major chip designer, Nvidia, AMD, Apple, depends on a single company operating on an island 130 kilometers from a state that has never renounced the use of force to bring it under its control. There is no second source. No fabrication facility on another continent can currently replicate what TSMC does at the leading edge. This is why Taiwan has become, whether its government wanted the role or not, the single most strategically important piece of real estate in the global economy, and why so much of what looks like a conventional military standoff in the Taiwan Strait is, underneath, a fight over who gets to write the rules for the technology that will define the rest of this century.
Washington understood this earlier than most governments and has spent the past four years trying, with mixed success, to choke off China’s access to the chips and chip-making tools that would let it close the gap. The story of that effort is instructive precisely because it has been so inconsistent. In 2022 the Biden administration imposed sweeping restrictions on advanced computing exports to China. By 2023 those restrictions had expanded to cover a wider set of chips and a wider set of countries. In January 2025, in its final days, the Biden administration issued the AI Diffusion Framework, an attempt to control not just what chips could be sold to China but which countries anywhere in the world could buy advanced American AI hardware at all. The incoming Trump administration rescinded that framework in May 2025, then imposed its own restrictions on Nvidia’s China-specific H20 chip in April 2025 after the Chinese startup DeepSeek demonstrated, to considerable alarm in Washington, that a capable model could be trained on hardware built specifically to satisfy the old export limits. Then, in July 2025, Commerce reversed itself again and let H20 exports resume.
Then, in December 2025, President Trump authorized the export of the more powerful H200 chip to China, subject to a cap limiting shipments to half of Nvidia’s domestic American sales volume. Reporting from CNBC and Al Jazeera through the first half of 2026 has tracked a further round of tightening, an attempt by the Bureau of Industry and Security to close a loophole that had reportedly allowed Blackwell-class chips to reach subsidiaries of Chinese firms operating outside China. Brookings researchers, reviewing this sequence in mid-2026, concluded bluntly that the United States has effectively ceded the commercial AI chip market inside China to a mix of domestic Chinese suppliers and gray-market channels, even as it continues trying to restrict the export of its most advanced hardware. Whatever one thinks of the policy, its zigzagging tells you something important: export controls are now one of the primary instruments of great-power competition, and neither government fully knows yet how to calibrate them without damaging its own companies in the process.
China’s countermove has been to weaponize the raw materials sitting underneath the entire chain. Beijing controls something in the range of 90 percent of global rare-earth processing capacity and dominant shares of other critical minerals, 80 percent of tungsten, roughly 60 percent of antimony, according to analysis compiled by industry researchers and cited in reporting on the sector through early 2026. In April 2025, China imposed licensing requirements on seven rare-earth elements. In October 2025 it went further, requiring foreign companies to obtain Chinese export licenses for any product, made anywhere in the world, that contains Chinese-origin rare-earth materials or was produced using Chinese rare-earth technology, an extraterritorial reach with few precedents in modern trade policy. The International Energy Agency’s commentary on the episode noted that European rare-earth prices spiked to as much as six times their level inside China, and that American, European and Japanese manufacturers in the automotive, defense and semiconductor sectors scrambled to secure supply. The two governments pulled back from the brink at the APEC summit in Busan at the end of October 2025, with China suspending its most aggressive October measures until November 2026 in exchange for a matching American suspension of a rule targeting Chinese corporate affiliates. But the underlying licensing architecture, the part that lets Beijing decide who gets rare earths and on what timetable, remains fully intact. The European Central Bank has estimated that more than 80 percent of large European firms sit within three degrees of separation from a Chinese rare-earth producer. That is not a trade dispute. That is a demonstration of leverage, deployed and then partially withdrawn, precisely to show that it can be redeployed at will.
This is the texture of the new competition: less about who has more missiles, more about who controls the choke points that everyone else depends on, and who is willing to use that control as pressure rather than simply as commerce. The Council on Foreign Relations, the Center for Strategic and International Studies and RAND have all, in different ways over the past two years, converged on a similar observation, that supply chain interdependence, once assumed to be a stabilizing force that would make war between major economies unthinkable, has instead become a battlefield of its own, where the weapon is a licensing delay rather than a missile launch.
Also Read: AI Will Create More Billionaires Than Any Technology Before It
AI sits at the center of this because it is simultaneously the prize and the terrain. The leading labs, OpenAI, Google DeepMind, Anthropic, are American and British institutions whose models increasingly function as strategic assets rather than mere consumer products. Governments now treat frontier model capability the way they once treated uranium enrichment, a marker of relative power that cannot be allowed to fall too far behind a rival’s. China’s DeepSeek moment in January 2025 rattled Washington precisely because it suggested that capability gaps could be closed faster and more cheaply than American planners had assumed, even under an export control regime designed to prevent exactly that outcome. The result has been a strange asymmetry: the United States retains a commanding lead in frontier model research, but its own confused export policy has repeatedly handed Chinese firms access to the hardware needed to narrow that lead, a self-inflicted wound that no Cold War-era adversary would have tolerated from its own bureaucracy.
Layer onto this the fact that private companies are no longer passive actors buffeted by state policy. They are increasingly geopolitical actors in their own right. Nvidia’s chief executive has met directly with the American president and reportedly with Chinese officials to negotiate the terms of his own company’s market access, a level of corporate diplomacy that would have been unimaginable for an arms manufacturer during the original Cold War. TSMC’s fortunes are now so central to Taiwan’s economy that the company alone accounts for more than 40 percent of the value of Taiwan’s benchmark stock index, according to reporting this year, a concentration so extreme that Taiwanese regulators had to raise the ownership cap that domestic funds are permitted to hold in a single stock. When a private chipmaker’s balance sheet becomes a matter of national security calculation for an entire government, the line between corporate strategy and foreign policy has effectively dissolved.
The same logic extends into orbit. Private satellite constellations, most visibly Starlink, have already demonstrated battlefield relevance in Ukraine, giving a commercial company meaningful influence over the conduct of an actual war, an influence that no government fully controls. China and Russia have each pursued their own satellite mega-constellations partly in response, and the emerging competition over low earth orbit, spectrum allocation and anti-satellite capability is becoming as consequential as the old competition over nuclear delivery systems, with far less public attention paid to it. The Pentagon’s own strategic documents increasingly treat space as a warfighting domain rather than a neutral commons, and NATO has said the same. None of this required a single formal declaration of rivalry. It emerged from the logic of dependence: once everyone’s economy, military and communications run through the same satellites, those satellites become targets and instruments simultaneously.
Russia complicates this picture in a way that pure technology competition does not fully capture. Moscow is not competing with Washington for control of semiconductor fabrication or frontier AI models; it lost that race some time ago and its own defense industrial base now depends heavily on components routed around sanctions, often through intermediaries in Central Asia, the Gulf and China itself. But Russia’s war in Ukraine has functioned as a forcing mechanism for exactly the kind of Western rearmament described above, and SIPRI’s data shows Russian military expenditure rising by 5.9 percent in 2025 alone, with 7.5 percent of Russian GDP now devoted to the military, even as the country enters what SIPRI researchers describe as a fifth consecutive year of war financing. Ukraine itself, by SIPRI’s measure, now spends an estimated 40 percent of its GDP on defense, the highest ratio of any country on earth, a wartime economy in the most literal sense. What this tells me is that the new Cold War does not have a single axis. It has at least two, a technological rivalry centered on the Indo-Pacific and a more conventional, grinding territorial one centered on Eastern Europe, and the two are connected mainly by the fact that Western capitals now treat both as symptoms of the same underlying condition: a retreat from the assumption, common in the 1990s and 2000s, that economic integration would make major war and coercive statecraft obsolete.
The contest over frontier AI deserves one further point of nuance, because it is easy to flatten it into a simple race between American and Chinese laboratories, and that framing misses how genuinely open the competitive landscape still is even among firms nominally on the same side. OpenAI, Google DeepMind and Anthropic are not merely racing China. They are racing one another, for talent, for compute allocation, for enterprise and government contracts, and for the kind of safety and reliability reputation that increasingly determines which model a finance ministry or a defense contractor is willing to build on top of. Governments have noticed this internal competition and have started trying to shape it, through procurement preferences, through security clearances for model access, through the kind of export licensing regime that now governs which countries can buy the compute needed to train these systems at all. The result is a peculiar hybrid: a market competition among private firms that increasingly functions as a proxy for state competition, without any of the firms involved being formally instruments of state policy in the way a Cold War-era defense contractor was.
Cyber operations and information warfare complete the picture, and here the evidence is less about spectacular headline-grabbing hacks than about a steady, grinding contest for infiltration and narrative control. State-linked cyber intrusions against critical infrastructure, reported with some regularity by Western intelligence agencies and by firms tracking these campaigns, have become routine enough that they barely register as news anymore, which is itself worth noting. Disinformation campaigns, some documented by European Commission task forces and by independent researchers, have moved from crude propaganda to sophisticated, AI-assisted content designed to blend into ordinary social media feeds. The tools that make generative AI useful for drafting a business memo make it equally useful for manufacturing a synthetic news article or a fabricated official statement, and the same companies racing to build more capable models are, often reluctantly, becoming the frontline defenders against the misuse of their own technology. This is not a side effect of the new Cold War. It is one of its central battlegrounds, fought largely without uniforms, without formal declarations, and without a clear moment anyone can point to as the beginning of hostilities.
I do not think any of this amounts to an argument that conflict is inevitable, and I want to be careful not to overstate what the evidence supports. The 2025 rare-earth standoff between Washington and Beijing ended, at least for now, in mutual de-escalation rather than confrontation. Nvidia continues to sell chips into China under a shifting set of restrictions precisely because both governments have concluded that total decoupling would be expensive and destabilizing for everyone involved. Trade between the world’s two largest economies, however constrained, continues at a scale that dwarfs anything comparable between the United States and the Soviet Union. The interdependence that makes this rivalry so dangerous also makes it, in a strange way, more restrained than the ideological Cold War that preceded it. Nobody is building fallout shelters. What they are building instead is redundancy: alternative chip fabrication capacity, alternative rare-earth processing, alternative satellite networks, alternative payment systems, a slow and expensive process of hedging against a rupture that both sides claim they hope never comes.
That is precisely why this new Cold War will not look like the old one, and why commentators who keep reaching for twentieth-century vocabulary keep getting the shape of it wrong. There will likely be no single wall to fall, no single crisis remembered the way the Cuban missile standoff is remembered. Instead there will be a slow accumulation of choke points contested and partially resolved, export rules tightened and loosened, alliances that spend more on defense without fully agreeing on what threat they are spending against, as NATO’s own internal debates over Russia’s timeline for aggression have shown. The contest will be won less by whichever side amasses more warheads and more by whichever side manages to build resilient, diversified supply chains for the handful of physical and digital resources, chips, rare earths, models, orbital assets, that now underpin everything else. Pakistan, the Gulf, and the wider Global South are not spectators in this. They are the terrain over which chip supply chains, mineral partnerships and AI infrastructure investment are increasingly being contested, whether or not their own governments have fully registered the stakes.
What strikes me most, writing this from Karachi rather than from Washington or Beijing, is how quietly all of this has arrived. There was no single speech, no single crossing of a border, that marked its beginning. It arrived through a licensing notice here, an earnings call there, a summit communiqué whose real meaning sat in a single disputed percentage point. The old Cold War announced itself with a wall and a doctrine. This one is announcing itself through a spreadsheet, and that may be exactly why it will be so much harder to see coming, and so much harder to know, with any confidence, when it has truly begun.
Also Read: How Michael Truell Built Cursor, and Why SpaceX Wanted It
