Welcome to this week’s edition of Geopolitics & the Day After. Each week, we curate and synthesize key developments from global politics, economics, and financial markets, drawing from a wide range of trusted sources. Our goal is to provide you with a clear, concise, and insightful overview of the forces transforming the world today and shaping tomorrow. Below is an overview of what we cover this week:
Geopolitical Concerns examines how autocratic survival logic in Russia and Iran is driving escalation rather than negotiation, even as Beijing pursues a quieter, incremental strategy to normalize control over the South China Sea, exposing a broader pattern of major powers misjudging how much pressure their weaker rivals can absorb.
Geoeconomics highlights the gap between markets pricing near-flawless cost discipline into Big Tech’s AI buildout and the mounting strain visible in currency and bond markets, from the first joint U.S.-Japan yen intervention in a generation to Treasury yields hitting a 19-year high as the Fed holds rates against oil-driven inflation.
Global Junctions traces how the AI race is increasingly decided by physical infrastructure and industrial capacity rather than model quality alone, as cheaper Chinese open-weight models squeeze U.S. pricing power, German automakers lose ground to Chinese rivals, and Beijing’s embedded model biases quietly shape global understanding of contested events.
Global Trajectories connects a historically strong El Niño and an unresolved reckoning over USAID’s dismantling to a broader erosion of institutional capacity, as fraying alliance trust with Canada and a record $2.84 trillion wave of AI-driven M&A show capital and power reorganizing around compute rather than traditional partnerships.
Geopolitical Concerns
Maksym Beznosiuk and William Dixon, Foreign Policy
F. Gregory Gause III, Foreign Affairs
Powerful Men Often Miscalculate
Mishal Husain, Bloomberg
Beijing’s Next Play for the South China Sea
Lynn Kuok, Foreign Affairs
Vladimir Putin’s war in Ukraine is entering what analysts increasingly describe as its most dangerous phase, driven not by battlefield logic but by the survival math of autocratic rule. Ukrainian drone strikes have disabled eight of Russia’s ten largest oil refineries since January, pushing more than 90 percent of Russian regions into fuel rationing and forcing domestic gasoline output roughly a third below demand, cutting directly into the revenue that buys the loyalty of Putin’s inner circle. Because any move toward negotiation now would read to Russia’s security establishment as an admission of the kind of vulnerability that ends autocrats, the Kremlin has instead intensified ballistic missile barrages on Kyiv, expanded strikes on Ukrainian gas infrastructure, and tightened domestic security controls, using each as a signal that the state can still project force even as its own refineries burn. Author and historian Peter Frankopan argues that Moscow has badly mispriced Ukraine’s capacity to keep absorbing costs, pointing to Russian battlefield casualties confirmed by the CIA at above 30,000 a month, with soldiers reportedly surviving just 20 to 35 minutes once deployed to the front. Even so, he expects the likelier outcome to be a frozen conflict along the lines of the Korean peninsula or Cyprus rather than a negotiated settlement, since Moscow shows no sign yet of concluding that continued war outweighs the risk of a reckoning at home. He also points to Ukraine’s strikes on Wildberries logistics hubs alongside the refinery campaign as evidence that the war’s costs have finally reached ordinary Russian consumers rather than remaining an abstraction shielded by state media.
The same autocratic logic is playing out from Tehran to the South China Sea. The Islamic Republic emerged from February’s war with Israel and the United States with its governing coalition intact, now dominated by Revolutionary Guard hardliners just as committed to opposing Israel as their predecessors. Tehran has found fresh leverage in its chokehold over the Strait of Hormuz, and the regime now has every incentive to pursue a nuclear weapon as the ultimate deterrent against future strikes from Israel and Washington. Iranian diplomats had reportedly floated a deal just before the war began under which Tehran would never stockpile enriched uranium, an offer that looks far less likely to be repeated now that the regime has emerged from the conflict in a stronger position, exposing Washington’s own miscalculation that force alone would bring Tehran to the table. Beijing has meanwhile shifted away from the high-intensity confrontations with Philippine vessels that defined 2023-24 toward incremental, harder-to-contest moves: a flag planted on the uninhabited Sandy Cay sandbar, a nature reserve declared over Scarborough Shoal, and a dredging campaign expanding Antelope Reef into a future military outpost. Each act is easy for Washington and its regional partners to dismiss as too minor to justify a response that risks escalation; collectively, however, they erode the norms, including a 2002 commitment not to inhabit unoccupied features, that have constrained Chinese expansion for two decades, leaving Washington to manage a security order fragmenting on several fronts at once.
Geoeconomics
Big Tech Stocks Are Pricing In a Miracle on Costs
Jonathan Weil, WSJ
America Is Great at Creating Stock Market Bubbles—and Shrugging Them Off
James Mackintosh, WSJ
The Worries That Drove Uncle Sam to Buy Yen
Jason Douglas, WSJ
US Borrowing Costs Hit 19-Year High as Federal Reserve Defies Inflation Fears
Financial Times
Wall Street’s earnings models for the largest AI hyperscalers assume a degree of cost discipline with little precedent. Analysts project combined operating margins for the five biggest cloud and AI companies will climb to roughly 31 percent by 2029, requiring sales, general, and administrative expenses to fall to about 8 percent of revenue from 10 percent even as combined revenue nearly doubles and depreciation from new data centers surges. Oracle, Amazon, and Alphabet have already turned free cash flow negative to fund the buildout, and the wide dispersion in analysts’ expense forecasts, in some cases a standard deviation of more than 30 percent around the consensus, suggests the middle of these income statements is being reverse-engineered to fit management’s revenue guidance rather than being independently modeled. The broader market has so far treated this kind of speculative excess as containable rather than systemic. The four-month boom and bust in memory-chip stocks like SK Hynix wiped out significant paper wealth without denting the S&P 500, which sits within 2 percent of its record, because the losses were financed mostly with equity rather than debt. Investors and strategists note this pattern has repeated across a dozen mini-bubbles since the 2008 financial crisis, from SPACs to clean energy to crypto treasury stocks. None of those busts, including a nearly 90 percent collapse in the Ark Innovation ETF, has stopped the broader index’s steady climb. The risk is that the scale of AI capital spending, estimated at $7 trillion in data centers over four years and increasingly financed with debt, could eventually be large enough to break that pattern and transmit losses into the wider financial system.
That optimism sits uneasily against signs of strain building in currency and bond markets, both linked to the same oil shock driving inflation concerns. The yen fell to its weakest level against the dollar in 40 years before the United States and Japan carried out their first joint currency intervention in a generation, buying yen to arrest a slide that Washington worried could push up U.S. interest rates and jeopardize $550 billion in planned Japanese investment tied to last year’s trade pact. The intervention pulled the yen back from roughly 164 to 156 per dollar, but Treasury Secretary Scott Bessent has signaled the U.S. will intervene again, reflecting concern that a disorderly yen selloff could unwind the currency’s role as a funding source for global carry trades. Japan had already burned through an estimated $70 billion in reserves defending the currency in April and May with only a temporary effect, and preliminary data suggest it spent roughly another $50 billion in this latest round. In the United States, the 30-year Treasury yield jumped to 5.24 percent, its highest since 2007, after Federal Reserve Chair Kevin Warsh held rates steady for a fifth consecutive meeting despite surging oil prices from the war in Iran and three dissenting governors who wanted an immediate hike. Warsh argued that market-driven increases in bond yields between meetings had already tightened financial conditions on the Fed’s behalf, an explanation that left traders unconvinced and pushed the S&P 500 down 1.5 percent on the day. The gap is notable: the two-year Treasury yield actually fell to around 4.28 percent on the same news, meaning investors expect the Fed to hold short-term rates low even as long-term borrowing costs climb regardless, a combination that raises financing costs for mortgages, corporate debt, and the federal government’s own deficit at once. With inflation running at 4.1 percent, more than double the Fed’s target, the bond market is signaling less confidence in the Fed’s inflation-fighting credibility than the AI-driven equity rally would suggest.
Global Junctions
Silicon, Not Software, Will Decide the AI Race
Antonin Bergeaud and Robin Rivaton, Project Syndicate
How Chinese AI Models Could Upend Anthropic, OpenAI, and Nvidia
Reshma Kapadia, Barron’s
The German Auto Industry, a Pillar of the National Psyche, Is Trembling
Jack Ewing and Jim Tankersley, New York Times
The Global Danger of China’s State-Controlled AI Models
Yaqiu Wang, World Politics Review
Competition in artificial intelligence is shifting from which lab builds the best model to which country controls the physical stack underneath it. The transformer architecture and ChatGPT’s 2022 launch created the market that now finances an unprecedented buildout of processors, memory, cooling, and electricity. The countries winning this phase are those coordinating semiconductors, packaging, and power at scale: Taiwan’s TSMC is running its most advanced production lines at full capacity with demand outstripping supply roughly three to one, South Korea’s SK Hynix dominates the high-bandwidth memory that trains frontier models, and China now produces about 35 percent of its own semiconductor equipment after building capability from packaging up through fabrication and, reportedly, a prototype extreme-ultraviolet lithography machine. That industrial buildout is already reshaping pricing in the software layer above it. Alibaba’s newly open-sourced Qwen 3.8-Max model claims coding performance comparable to the latest releases from OpenAI and Anthropic, and Chinese open-weight models are running at roughly a tenth of the cost of their American proprietary counterparts, with monthly token usage of Chinese models up 70 percent year over year in June and steadily gaining share. The White House has accused Moonshot AI’s Kimi K3 model of relying on large-scale distillation of Anthropic’s technology, a claim Beijing has rejected as unfounded, and U.S. policymakers are now weighing restrictions that would protect American labs’ pricing power at the cost of raising expenses for the U.S. companies and cloud providers that have adopted the cheaper Chinese alternatives.
The stakes of losing an industrial edge to Chinese competition are already visible in the auto sector, where German manufacturers spent decades trading engineering know-how to Chinese joint-venture partners in exchange for access to the world’s largest car market, only to watch those partners turn into rivals. Volkswagen, Mercedes, and BMW have posted double-digit sales declines in China this year, Chinese brands outsold Japanese ones in Western Europe for the first time in June, and Volkswagen’s own chief executive has questioned whether four of the company’s German factories remain competitive at all, a crisis serious enough to strain the country’s tradition of consensus-based labor relations and fuel support for far-right politicians. Beijing’s advantage extends beyond hardware and manufacturing into the information layer of its AI models themselves. Chinese systems like DeepSeek decline to describe the 1989 Tiananmen Square crackdown and repeat Beijing’s territorial claims over Taiwan, but researchers say the more consequential distortion is subtler: the same models have been found to give more accurate answers about Russian atrocities in Bucha when queried in English or Ukrainian than in Russian. When asked directly about the killings documented in that Kyiv suburb, some Chinese models reportedly offer only vague references to competing accounts rather than describing the extensive evidence of Russian responsibility that international investigators have compiled. The same models can quietly omit that governments cannot override individual rights in the name of collective development. Because retraining an open-weight model to strip out these embedded assumptions is technically difficult and rarely a priority for developers who adopt it, the growing global reach of Chinese models threatens to shape not just who builds the cheapest AI, but how billions of users come to understand contested history and rights.
Global Trajectories
What a Powerful El Niño Means for the World
Christina Lu, Foreign Policy
How Many People Are Dying as a Result of USAID’s Demise?
The Economist
‘You Aren’t Going to Like What Comes After America’
Chrystia Freeland, New York Times
Global M&A Deals Reach Record $2.8tn in 1st Half Amid Fight for AI Lead
Miyu Fukawa, Nikkei Asia
One of the strongest El Niño events on record is now developing, and forecasters expect it to peak this winter with the potential to inflict global economic losses reaching $3.1 trillion over five years if the pattern matches the 1997-98 event. The damage will fall unevenly: parts of the United States and East Africa may see stronger yields, while southern Africa, India, and Central America face drought risks just as the war in Iran and U.S. tariffs are already straining government budgets, leaving less fiscal space to respond when the shocks arrive. South Africa’s maize sector alone saw yields plunge by about 20 percent during the comparably strong El Niño events of 1997-98 and 2015-16, a benchmark now worrying agricultural forecasters heading into this winter’s peak. That erosion of response capacity echoes an unresolved debate over the human cost of dismantling USAID. Elon Musk maintains the agency’s closure cost zero lives, while a widely cited Lancet study projects 14 million deaths by 2030, a figure built on the assumption that American aid would fall by 83 percent when actual cuts have been closer to a third, concentrated in education and democracy programs rather than the health and humanitarian funding that saves the most lives. The truth sits between the two claims: ambulance and nutrition programs that lost funding have already cost identifiable lives in Malawi and Somalia, even as the sweeping mortality projections rest on extrapolating gains from an unusually favorable period, in which vaccines and other low-cost interventions were still being rolled out, onto a period when most of those gains have already been captured.
The same fiscal and institutional strain is reshaping alliances that once seemed settled. The Trump administration has cycled through justifications for tariffs on up to $20 billion of Canadian exports, from dairy access to wildfire smoke, prompting Canadian commentators to argue the stated rationale matters less than the underlying message that Washington now expects deference rather than partnership from even its closest allies. A recent Pew survey found that in most of 36 countries surveyed, including Canada, France, Germany, and Britain, China is now viewed more favorably than the United States, a shift that reflects frustration with Washington’s approach to allies even though China’s own government imposes far harsher restrictions on its own population and trading partners. Capital flows are reorganizing around a different axis entirely. Global mergers and acquisitions reached a record $2.84 trillion in the first half of the year, up 50 percent from a year earlier, with U.S. companies accounting for 56 percent of the total and the largest single deal, SpaceX’s roughly $250 billion purchase of xAI, underscoring how thoroughly the AI buildout is now driving corporate consolidation well beyond the technology sector itself, into energy companies like Dominion and AES. Japanese dealmakers have kept pace despite a weaker yen raising the cost of overseas purchases, with Mitsubishi Corp.’s roughly $7.3 billion acquisition of U.S. natural gas developer Aethon standing as the largest deal struck by a Japanese company this year. Where the postwar order once organized itself around treaty alliances and shared institutions, the current period is instead organized around access to compute and capital, a reordering that leaves traditional allies uncertain of their standing even as they remain economically bound to Washington.
