Weekend Food For Thought WFFT
On Todays Menu: Geopolitical Trends, Defense, Future of Conflict, BIS 2026 Q1 Outlook, China's AI Monetization Engine, and much more...
Hello from Lisboa,
I hope you had an interesting and productive week.
Donella Meadows states that: “Leverage points are places within a complex system where a small shift can produce big changes.”
May you find food for thought on where such leverage points may reside in the world around us today
1 Getting Visual
2 If You Read One Thing Today - Make Sure it is This
3 Consequential Thinking about Consequential Matters
4 Big Ideas
5 Big thinking
6 Patience
1 Getting Visual
Macro Spotlight - US Fundamentals: Persistent trade deficits lead to the accumulation of large financial liabilities, matched by reserves on the other side. The US, the main deficit country finds itself shifting more and more deeply into a position of large net debtor to the world. At close to $ 30 trillion, US net liabilities to the rest of the world are closing in on US GDP. This is a noteworthy figure.
Attention (US) Deficit - Given low unemployment, the deficits that have been piled up since the mid 2010s are historically anomaly.
Big Question: Valuations Vs Value? 59 of the top 100 public companies in the world are American and they enjoy 75 percent of total market capitalization.
BIG DATA: US data centres are projected to consume nearly 10% of the entire US power grid by 2030. That’s four times the percentage China is forecast to hit. The US has about half of the world’s data centres but only 4% of the global population.
BIG DATA Plans with a missing key Component? Does the AI build out in the US have a 15 GW pa upper bound? AI hyperscalers may have huge plans but there is not much evidence that the US can actually build more than 15 GW per year.
Global Solar supercycle - For the first time on record, a renewable energy source was the largest contributor to global annual energy demand growth. Solar PV supplied over a quarter of the increase in energy demand.
Renewable PE - “PE growth bets on a clean energy sector that’s grown up. PE growth investments in clean energy skyrocketed in 2025, overtaking control deals for the first time on record. The reversal reflects investors chasing a maturing clean energy sector with the capital to scale as electricity demand outruns supply. The value of minority PE investments in the sector rose nearly 130% year-over-year to $22.5 billion last year, driven by a sharp increase in the volume of large deals. Investors are moving away from pure-play renewables toward anything that can help generate stable, sustainable power capacity. This includes improving grid use and energy storage, or building virtual power plants that shift electricity demand from supply-constrained regions to areas with excess capacity, said Hans Kobler, founder and managing partner of Energy Impact Partners, a $4.5 billion tech investor. The International Energy Agency predicts electricity demand will outpace overall energy demand by more than twofold by 2030—driven by the AI boom and rapid data center expansion, rising industrial use of electricity and continued adoption of electric vehicles. Many of these renewables-adjacent businesses are technologically advanced and have the potential to scale quickly, opening up new avenues for PE growth and VC investors. The growth in minority deals also reflects a maturing market. More clean energy companies are moving past the technology validation stage to building real assets, including large-scale power generation projects. They no longer need a new owner, just the capital to scale more quickly, said Charles Cherington, co-founder of Ara Partners, a PE firm dedicated to buyouts and project financing in the industrial decarbonization sector. “This reflects a healthier pipeline of businesses that have proven their technology and are ready to move up-market to full commercial deployment,” Cherington said. - Pitchbook
Spotlight: India - A look below the top line shows massive scale and a real North-South Divide. India’s Uttar Pradesh, is a state more populous than Brazil and poorer than sub-Saharan Africa.
2 If You Read One Thing Today - Make Sure it is This
It’s actually two pieces but they cover interesting and important ground to ponder in the context of geopolitical trends, defence and the future of conflict - ChinaTalk pod sits down with Bryan Clark of the Hudson Institute and former submariner; Justin Mc, former Green Beret now in defense tech; Eric Robinson, former OSC/NCTC analyst and 101st Airborne officer, now a lawyer; and Tony Stark for some real talk and the good people of Exponential View explores broader lessons for manufacturing and some granular insights from Ukraine’s evolving military capacity and strategies. Both are worth your time - go check them out in full here:
Some Takeaways
“The Iran war burned through America’s L-RASM, JASSM-ER, and Tomahawk stockpiles — weapons designed for a Pacific fight against the PLA Navy, not the Iranian corvette fleet. Now Pentagon insiders are leaking that we can’t win a war over Taiwan, and it’s a six-year pipeline to refill magazines.”
“…we’ve already ramped up production, we’ve ramped up production in the last few years after Ukraine. The problem is we were on minimal sustaining rates for way too long. It’s not like we just had a max capacity magazine and we decided to empty it out and we can get back to it in two years. We were already low. We’ve burned way too much, and now just to get back to that previous low standard, it’s going to take years.
Justin: We have over-indexed on exquisite technologies because we have the ability to produce them. The problem with exquisite technologies is they take a very long time, they have very tenuous supply chains, and you can’t really do exquisite in high capacity.”
“This is a great chance to rethink what the portfolio should look like and rebalance it towards these more modular weapons — not even really lower-end, but more modular weapons that maybe don’t have quite the performance of the preferred munitions, but you can buy them at much higher volumes and the production is much easier because they’re modular and some even use commercial components, like this ERAM missile that the Air Force developed.”
“That there is this baked-in collective understanding that if the United States is going to go to war against one of the big four threats, there’s going to be a Clausewitzian strategic approach to it? We are not there.
Instead we’ve got people doing wheelies in Lamborghinis, and it looks really cool and it gives you your sizzle reels, but we are still at an impasse with an Iranian state that refuses to fundamentally break down — and using your entire inventory of L-RASMs, which for viewers that haven’t engaged with this before, is an advanced anti-ship missile almost expressly designed for the United States Air Force to employ against the People’s Liberation Army Navy in Pacific contingencies. Instead, we went after the Iranian Navy, which is sort of like expending L-RASMs on the Austrian Navy. It doesn’t matter. But here we are. We are disaggregated from strategy. And now we’ve got empty war stocks, so we’ve got a six-year pipeline to try and restore this, assuming Congress gets behind it and funds it.”
“I’ll give credit — the Department of Defense innovated and stole a model from the Iranians. They didn’t go and buy a turdcopter from Silicon Valley. They got something that they knew worked on the battlefields and copied it relentlessly. And maybe the IRGC will go to court in the Southern District and go after the United States for stealing its IP, but in this war, the United States tested what we effectively stole from the Iranians and repurposed.
Bryan Clark: My concern now going forward is we’ve got this big defense budget that we’re looking at. And you’ve got this lobbying campaign by people inside the Pentagon, over at INDOPACOM, to basically restock all these weapons that they’ve built their plans around — not new weapons, but existing weapons. That’s going to take up a huge chunk of the budget. And we want to make these long-term commitments on weapons production to try to get production capacity increased, because a company like Raytheon is not going to expand production capacity unless they have a long-term commitment from the government. So if you do a five- or seven-year multi-year procurement, they’re going to be willing to support that level of production.
But that means you’re locking in a huge chunk of investment over the next seven years that’s going to be devoted to weapons designed in the Cold War or the immediate aftermath, and designed primarily to go after the highest-capability threats posed by China. It just seems like we’re going to lock ourselves into a portfolio that’s going to have the same challenges as our existing portfolio. It’s never going to get big enough, and it’s never going to have the ability to surge in the way that maybe a new portfolio of weapons more modular or more focused on one-way attack drones might.”
“Bryan Clark: Another aspect of this is adaptability. In Ukraine, they found that the Excalibur rounds we sent were quickly obviated by Russian electronic warfare against GPS, and same with GMLRS.
Eric Robinson: Excalibur is a 155-millimeter round that can be guided by GPS. In the global war on terrorism, Excalibur came out because the wars in Iraq and Afghanistan were uncontested electronic warfare environments. It enabled individual battalion commanders to effectively choose a 10-digit grid — a one-meter spot on a battlefield — and say, I’m going to blow that up. So Excalibur revolutionized precision strike. To Bryan’s point, when it got to Ukraine and the Russians — a far more sophisticated adversary — jammed global positioning satellites, those rounds were dumb rounds and were not particularly helpful.
Tony Stark: GPS is this wonderful invention that we’ve had around for over 40 years. The problem is, because of the way it functions, it’s incredibly easy to jam. If your entire targeting package is built around being able to find that 10-digit grid, that’s a massive complication, and the US still hasn’t quite gotten around it.”
“The problem we have with these legacy weapons — these kind of high-end weapons that are highly integrated, like Excalibur — is they’re too hard to modify. We still haven’t really fixed Excalibur to address this GPS jamming issue. And they’re desperately trying to fix GMLRS to make it able to use other sources of navigation, like radio emissions. So we’re going to invest a bunch of money and make long-term commitments in weapons that are difficult to adapt, because we don’t know what the next countermeasure from the opponent is going to be. It’s GPS jamming today, but it could be something else tomorrow that goes after their seeker mechanism or their ability to orient because we’re going after something else in the electromagnetic spectrum. There’s all these opportunities for move–countermove competitions that these weapons don’t give you the ability to respond to.”
“We’re witnessing the creation of an information environment where once you get past the official bluster from the Secretary of Defense, there is an authentic problem now that has to be resolved.
Jordan Schneider: Here’s the problem. You know who else uses remove-paywalls on the Washington Post? Our allies in the Pacific and the Chinese government. So this is not ideal the way you’re going about this, if we’re trying to preserve deterrence capacity. One of the quotes from a senior administration official was, we can’t win a war today over Taiwan. If you’re a Taiwanese politician and you read that, what do you feel like? It is a signal of the breakdown of this administration and how they’re trying to rebuild.”
“Eric Robinson: Modern information markets — whether Predictit, Kalshi, or Polymarket — encourage individual action to spike these markets. And it’s not unique to American special operations.
This is not new behavior, but it’s being dramatically exacerbated. It used to be you had to be a sports hero in order to throw a game, or you had to have sufficient financial backing to witness something coming and then place a bet in commodities markets — you could look at the price of oil, there were ways to do this. But sleaze is now super democratized. There are such a great variety of markets, and there are ways for individuals to push events one way or another.
Jordan Schneider: We’ve had congressional insider trading for decades now. Which is the justification one congress member made yesterday saying this person should get a pardon and just have to give their money back. But it starts with that — there’s some level of permissivity coming from the legislators themselves. Which is not to say that should be legal or this stuff should be legal.
But there’s a wholesale reckoning with the whole graft ecosystem that really should happen sooner rather than later.”
From Exponential View:
How Ukraine solved the hardest problem in defense
This is the operating model. A device is designed, fielded, disabled by the enemy, diagnosed through a conversation between the operator who lost it and the engineer who built it, then redesigned and redeployed in roughly seven days.
Industrial tempo in the grueling conditions of war. Continuity of signal, of keeping systems operational outweighs a lot else. In one unit along the Dnipro, an operatorrefused to take shelter during 120mm mortar fire because moving would break the radio link to his drone mid-mission.
This is how military innovation happens on parts of the Ukrainian front.”
“In the US and Europe, weapons systems move from concept to deployment over five, ten, sometimes fifteen years. A major capability iteration of the flagship F-35 Lightning II fighter jet can take close to a decade. That gap – one week versus seven years – is not just down to the exceptional engineering talent in Ukraine. It is the product of how work is organized, how quickly information reaches the people who can act on it and how much authority those people hold.
It is a real-time demonstration of what becomes possible when you remove the layers between the person who sees the problem and the person who can fix it. Under fire, alacrity is the difference between life and death.
While the stakes are as new as they are high, the lesson carries well beyond the muddy, blood-soaked, adrenaline-drenched trenches of Ukraine.
It’s a lesson that will determine which nations will succeed, which will falter, which will invite attack, and which will stand firm as this century slowly fills with conflict.”
“This scale hasn’t compromised speed. Production is distributed and it has followed the same design pattern established in the early phases of the war. There is no single dominant contractor. There is no centralized R&D pipeline. Hundreds of firms operate in parallel, each pursuing its own variations on shared problems. When conditions change, and on the battlefield there is more change than constancy, the network responds. It is simultaneous, emergent, like an immune system.
The individual firms do not have to be faster than their Western counterparts – some are, but that isn’t the point – it’s that the system as a whole is faster. The network searches a wider space for solutions, with far more simultaneous bets than any monolith can. This system is also more resilient. Firms can fail, factories can be bombed. The network persists.
More subtle but less obvious, is the variety and breadth of innovation. Curiosity is a byproduct of the network itself. Solutions win because they work. Those that succeed are stamped, repeated, scaled. Those that don’t disappear.”
“The Exponential View House View: Hardware iteration speed is an organizational phenomenon, not an engineering one.
The one-week cycle is not a product of exceptional engineering talent. Ukraine’s engineers are skilled, but they are not uniquely so. The cycle is a product of feedback loop architecture: how quickly failure data reaches the people who can act on it, and how many parallel experiments are running simultaneously. Any organization that treats hardware R&D as a capability problem rather than a structural problem will misdiagnose it – and find that importing the metric without importing the structure produces neither the speed nor the cost.”
“Every industry that has ever scaled manufacturing locks the design before the line is built.
Ukraine has broken these rules. Monthly FPV1 output grew from roughly 20,000 units in early 2024 to 200,000 by the end of the same year – a tenfold increase in twelve months, even as the designs running down those lines changed week by week.
The core mechanism is modularity (read our primer on modularity here). Most Ukrainian drone designs share standardized components, common software frameworks and interchangeable subsystems. A new guidance update or jamming-resistant radio does not require retooling the line. It requires swapping a module. Ten thousand units a month and weekly design changes don’t disrupt the process; they are the same system.
Many of the people at the heart of this system have no background in the defense industry. They come from software, gaming, marketing, even B2B SaaS. Again, Cat Buchatskiy shared that “[o]ne of the biggest defensetech VCs now supporting the industry used to be the chief marketing officer at a workflow automation company. Many were working at Uber Ukraine or other rideshare companies when the invasion began.”
What matters is not where these skills came from but how close they sit to the people using them. Engineers talk to operators directly. Feedback is not categorized, aggregated, or mediated through layers of committees. The distance between someone seeing a failure and someone who can fix it is a phone call away.
Each of these – scale without stability, procurement after proof, engineers inside the feedback loop – is an organizational choice. None of them require Ukraine’s conditions to implement. All of them sit, in most Western organizations, on the other side of a conversation that has never been had.
In essence, this is an analogue to Silicon Valley. When you strip back upfront procurement, build with minimal capital, test in the real-world, find product-market fit before you scale, and trust outsiders over incumbents because they have no loyalty to the assumptions the incumbents have forgotten they’re making. This is not a new organizational model. It is the Valley’s model, applied in a war zone.”
“On March 13 this year, in an interview on Fox News, Donald Trump said of the Ukrainians: “We don’t need their help in drone defense.” Six weeks later, Reuters reported that the Pentagon had deployed a Ukrainian counter-drone platform named Sky Map to Prince Sultan Air Base in Saudi Arabia. Ukrainian engineers are training Americans on how to use it.
The United States Air Force, with an annual budget of $300 billion and more spent on air power in the past decade than the rest of the world combined, is being tutored by engineers from a country that, in 2022, barely had a drone industry.
The US and NATO are now scaling defense spending to Cold War levels. But layering a Cold War budget on top of a Cold War procurement system is merely a subsidy to the other side’s learning curve. The binding constraint in this war, and in the wars that will follow, is organizational design: the cadence at which an army, a ministry, and a supply base can learn together.
Ukraine has compressed that cadence to seven days. The Pentagon’s sits somewhere between five and fifteen years. With this model, each additional dollar spent by Western institutions will mostly buy more expensive obsolescence – and more Ukrainian engineers teaching Americans how to use kit the Americans could not build in time.”
3 Consequential Thinking about Consequential Matters
The BIS dropped their 2026 Q1 Outlook and the section below on market trends is a useful overview with a few good deep dives (private credit and precious metals) on consequential matters - Go read it in full here:
https://www.bis.org/publ/qtrpdf/r_qt2603a.pdf
Some Takeaways
“During the review period, financial markets had to adjust to shifting currents. Even though markets initially appeared calm on the surface, there were significant shifts under the surface as investors rotated away from previously high-performing assets. Volatility began to creep up, exacerbated by the conflict in the Middle East in early March.
Global equity markets saw regional and sectoral shifts in late 2025, with investors moving away from US large cap and growth stocks. European and Japanese equities posted gains, while emerging market economy (EME) equities saw an even greater boost. Despite solid earnings, concerns over artificial intelligence (AI) spending and disruption weighed on richly valued technology companies. Value and small cap stocks outperformed. While index volatility rose, it was eclipsed by significantly greater individual stock volatility.
Foreign exchange (FX) market shifts signalled growing investor unease in an increasingly fragile risk landscape. Through most of the review period, the US dollar depreciated as is typical of risk-on environments, but at the same time the Swiss franc appreciated notably, a hallmark of global risk-off episodes. The escalation of geopolitical tensions in the Middle East reversed the depreciation of the dollar, consistent with both its positive correlation with oil prices in recent years and the intensification of the risk-off sentiment.
Precious metals and energy prices also experienced notable volatility. Gold and silver surged early in 2026, reflecting a unique mix of investors’ quest for safe havens and speculative interest. However, the rally ended abruptly in late January, in part due to leveraged position unwinds. Rising geopolitical tensions in early March drove oil and natural gas prices significantly higher. Amid these shifts, precious metals came under renewed pressure, trading more like risk assets than safe havens.
Credit markets remained fairly stable but showed some differences across segments. Spreads for investment grade and high-yield bonds stayed near historical norms, supported by resilient investor appetite for yield, despite impacts from escalating tensions. While issuance in bond markets broadly held up, activity in leveraged loans and private dealmaking declined from earlier peaks. Concerns over AI-driven disruption refocused attention on private credit portfolios, especially those with significant software exposures, as investor redemptions intensified.
Sovereign bond markets diverged across advanced economies (AEs). Long-term yields rose sharply in Japan, while those in the United States and the euro area moved sideways. After initial narrowing, euro area sovereign spreads widened again after the conflict broke out. Inflation expectations edged up, leading investors to revise expectations of policy rates upwards and push back the expected timing of US rate cuts. Bond yields moved up and yield curves mildly flattened as a result, reversing some of the previous steepening.
EME assets initially rose, driven by a weaker dollar, carry trades and portfolio inflows, but momentum waned with intensifying geopolitical frictions. Latin American and Europe, the Middle East and Africa (EMEA) currencies outperformed, while Asian currencies saw modest gains. Portfolio inflows fuelled equity rallies, especially in commodity-exporting EMEs, and debt markets reflected favourable financing conditions. However, this positive trend was disrupted in early March due to heightened geopolitical risks.”
Cross-currents: selective risk-taking collides with geopolitical tensions
Despite geopolitical flare-ups, investors’ risk appetite broadly remained resilient through most of the review period, but risk-taking was selective. Solid corporate fundamentals and macroeconomic data releases contributed to this resilience. Investors’ tolerance for risk was put to the test as a resurgence of tariff-related uncertainty, sharp moves in precious metal prices and rising geopolitical tensions triggered bouts of volatility. Concerns about high valuations called into question the sustainability of the momentum in technology stocks and led to underperformance as investors rotated to other sectors. Yet risky assets held their ground for the most part, with equity markets posting modest gains and credit spreads remaining compressed.
The equity market resilience that characterised the second half of 2025 was tested on several occasions during the review period, but broadly carried over into 2026. Equity indices rose, supported by solid risk appetite and fundamentals. The rise in US equity markets was relatively subdued, while European and Japanese equities posted stronger gains, reflecting improved sentiment towards their respective economies’ outlooks and diversification flows away from US stocks. Rising tensions in the Middle East reversed some of the gains in early March, in particular for European and Asian equities, given their greater exposure to energy-related supply disruptions.
Large US technology stocks faced volatility despite strong earnings, as concerns over elevated valuations and future capital spending emerged. Guidance on higher capital spending raised fears of potential earnings disappointments, particularly for AI “hyperscalers”.
While corrections lowered valuations, price/earnings (P/E) ratios for the broader market remained above historical norms and approached dotcom bubble levels, though big tech firms’ ratios were comparatively lower (Graph 1.B). As the Magnificent 7 (M7) stocks had driven their share of the S&P 500 index to nearly 35%, declines in these stocks began to drag on market indices.”
Financing the AI infrastructure boom: on- and off‑balance sheet borrowing
Investment in artificial intelligence (AI) infrastructure, particularly data centres, has risen rapidly and now accounts for a substantial share of investment in advanced economies. Against this backdrop, big US tech firms – “hyperscalers have accelerated their capital expenditures (capex). But, as these expenditures have grown significantly beyond their usual investments, a rising share of spending is funded through borrowing. Corporate bond markets have been hyperscalers’ primary source of financing. Their gross issuance increased markedly, topping $100 billion in 2025. Most issuance was long-term, with maturities over five years, locking in funding for multi‑year build‑outs. However, credit default swap (CDS) spreads rose, especially for hyperscalers with lower credit ratings, reflecting both the volume of supply and uncertainties around the projects’ payoffs.
Alongside traditional bonds, hyperscalers have turned to off‑balance sheet arrangements to finance infrastructure expansions, often in partnership with private credit firms. A common structure involves a dedicated vehicle – often a joint venture or special purpose entity – that acquires or develops data centre assets. The vehicle is capitalised with equity from a consortium of sponsors and raises debt through private placements. The hyperscaler typically holds a minority stake, commits to long‑term operating leases or capacity offtake agreements (long-term commitments to purchase or reserve power and compute capacity), and may provide various guarantees. Economically, this substitutes upfront capex with multi‑year operating expenses while keeping most of the associated debt off the hyperscaler’s balance sheet. The debt is serviced by lease cash flows and is held by private credit funds and other institutional investors, sometimes with investment grade features supported by asset backing and contractual guarantees from hyperscalers with investment grade credit ratings.
These arrangements amount to “shadow borrowing”: obligations that are economically akin to debt but largely reside outside corporate balance sheets.
By channelling sizeable private credit into AI infrastructure, these structures strengthen links between hyperscalers and non‑bank investors such as private credit vehicles and insurers.
Banks support the vehicles with funding lines, potentially creating new shock transmission channels – eg via refinancing pressures at the vehicle level, procyclical shifts in private credit appetite or the activation of guarantees.”
Private credit’s software lending meets AI disruption
Lending by private credit funds to software-as-a-service (SaaS) firms has grown rapidly and makes up a substantial share of funds’ loan portfolios. Outstanding loans to SaaS firms increased from almost $8 billion in 2015 to over $500 billion, or 19% of total direct loans, by end-2025. By now, a third of private credit funds have extended loans to the SaaS sector, on top of their rising exposure to big US tech firms and other artificial intelligence (AI) companies).
In this box, we zoom in on business development companies (BDCs) to get a consistent picture of private credit lending to SaaS firms. BDCs are publicly traded and have quarterly disclosures. As such, they provide a window into the opaque private credit space. Moreover, BDCs account for one fifth of all direct loans in the United States and extended over 15% of their loans to SaaS firms in 2025. Concerns that AI may disrupt traditional SaaS business models have led to notable price adjustments in the software sector. Software companies’ stocks collapsed by almost 30% between October 2025 and February 2026 (Graph B1.B, red line), while BDCs’ stock prices fell by about 10% on average (blue line). Meanwhile discounts to net asset value, which is largely determined by the book value of illiquid private loans, deepened (yellow line), potentially signalling worries about underlying valuations.
BDCs with higher SaaS exposure have underperformed their peers (Graph B1.C). Since October 2025, BDCs with high exposure to software firms have performed around 5 percentage points worse than those with low exposure. These developments highlight investor concerns that further advances in AI tools may disrupt the SaaS sector amidst redemption pressures from private credit’s push towards retail investors.”
Boom and bust of the recent silver and gold rush: the role of leveraged retail investors
After a prolonged rally through 2025 and into early 2026, prices of precious metals such as gold and silver reversed abruptly in late January and February 2026. Retail-driven exuberance, increasingly channelled through exchange-traded funds (ETFs), set the stage for outsize moves, continuing the trend from 2025.
The daily rebalancing of leveraged ETFs and margin‑triggered liquidations amplified the swings, particularly in silver. Following substantial gains in 2025 and a further surge in early January 2026, gold and particularly silver prices plunged in late January (Graph C1.A). After doubling over 2025 and rising by over 50% in January 2026, the silver price fell by about 30% in a single day in late January (blue line). Gold broadly followed a similar but less extreme pattern (red line). The precious metals crash seemingly coincided with shifts in expectations about the US dollar and the path of monetary policy, but was hard to square with broader changes in fundamentals.
The abrupt price drop and the spike in precious metals’ volatility point to the role of retail flows, and amplification of price moves due to forced sales by leveraged ETFs, trend-following investors such as commodity trading advisers (CTAs) and margin dynamics.
Fund flow data indicate that retail investors were the main source of inflows into silver and gold funds in the run-up to the episode. In contrast, institutional investors maintained stable positions or even trimmed exposure (Graph C1.B). Moreover, futures positioning reveals long leveraged exposure among smaller speculative participants. “Non-reportables” – typically smaller investors – were long silver futures heading into the correction.
As prices fell sharply and exchanges raised margin requirements, these investors probably had to reduce positions quickly. “Managed money” – including CTAs and institutional investors – also cut long positions, while dealers stepped in to provide liquidity by cutting short positions (Graph C1.C). For some time, retail investors have found it attractive to use ETFs to obtain precious metal exposure. Sustained premia of gold and silver ETFs over their net asset value (NAV) signalled strong, one-sided buying pressure that outpaced primary market arbitrage (Graph C2.A). Such persistent premia arise when demand for ETF shares exceeds the capacity of authorised participants to create new shares and deliver physical metal to bring market prices down to NAV. As prices reversed in late January, these premia compressed rapidly, and for silver turned into pronounced discounts, consistent with one-sided selling pressure and a sharp turn in flows.”
Leveraged silver ETFs contributed to the turmoil because of their amplification mechanics. To maintain fixed daily leverage, these funds rebalance each day. When prices rise, they buy the underlying asset (often via silver futures) to restore target leverage and when prices fall, they sell the underlying asset. This predictable, momentum‑like trading creates feedback loops that reinforce prevailing trends and can distort prices.
The footprint of leveraged ETFs’ destabilising trading appears to have grown amid the retail-driven exuberance in precious metal markets. The leverage rebalancing multiplier, a summary measure of the market impact of leveraged ETFs’ daily rebalancing flows, doubled over the course of 2025 (Graph C2.B, blue line). The share of ETFs in the market followed similar dynamics (red line). This indicates that leveraged ETF activity became a larger part of the market and intensified price trends.
Margin‑triggered liquidations further amplified the sell‑off. Rapid price declines increased variation margins on futures positions, and several exchanges tightened initial margin requirements during the episode (Graph C2.C). The resulting funding pressures forced deleveraging among participants most exposed to the downdraft, akin to past stress episodes. The liquidations of investors’ positions, alongside systematic selling from leveraged ETF rebalancing into the decline, probably added to downward pressure, creating a self‑reinforcing loop of lower prices and further margin calls.
4 Big Ideas
Grace Shao takes a look inside China’s AI Monetization Engine and shares notes from her conversations with China AI insiders - Read it in full here:
Some Takeaways
“On the price side, China’s leading AI labs have been quietly nudging prices upward,especially Z.ai, which has explicitly announced. But it reflects a general trend in many of their coding subscription plans and the per-token rates they charge developers through their APIs. They are doing this from a position of scarcity rather than strength of marketing: demand for their models is outrunning the supply of GPUs they have to serve it. That changes who gets prioritized. Customers willing to pay the full sticker price, and especially those willing to sign multi-year commitments, are being prioritized or preferred over those hunting for day-to-day discounts. In plain terms, according to an investor, the labs are choosing better-quality revenue over bigger-looking revenue, which is the kind of move a business makes when it is trying to protect its margins rather than chase growth at any cost.
The subtler point, and the one most outside observers miss, is that the headline price of tokens is not really what generates the revenue in coding use cases. Most of the billing in a coding workload comes from something called “cache hits.” Here is what that means in plain English, which was explained to me: when a developer works within a large codebase, the AI has to reread much of the same code repeatedly as it answers follow-up questions. Rather than charging full price each time, the labs charge a much lower rate for that repeated content, and that is the cache-hit price. Because so much of a coding session involves re-reading the same files, those discounted cache-hit tokens end up being the bulk of what customers are actually paying for. T
his matters for how you compare Chinese models to Western ones. Depending on whether you compare input, output, or cached-input pricing, top Chinese models can range from modestly cheaper to dramatically cheaper than Sonnet. On output pricing, GLM-5.1 is roughly 70% below Sonnet 4.6; on standard input pricing, the gap is closer to 50%.
If you judge Chinese labs by the sticker price, you will badly underestimate how much money they are really making per customer and how healthy their margins actually are.”
The open-source game theory
Despite a growing mix of proprietary flagship releases from Qwen, Alibaba has not abandoned open source. It is now running a hybrid strategy across closed and open-weight models. The reason is multi-party game theory.
In a two-player game, both labs rationally close. In a three-player game, one player can always choose to open-source specifically to deny profit to the other two — the prisoner’s dilemma outcome, where at least one actor defects. As long as one frontier lab holds the line open, the others cannot permanently close. So, even though under pressure to profit, it is believed that the Chinese AI ecosystem will not be closed forever. The read is that at least one frontier-grade open-source model will persist, which caps how high closed-source pricing can go.
It’s been widely reported, shared, and recognized that Chinese labs, in particular, lean toward openness for non-ideological reasons: they need an ecosystem, brand recognition, and community traction that a closed Chinese model cannot easily earn. The interesting take is that despite it sounds like a philosophical choice, whether to open or close, that debate is driven by where your economic interest lies: if you own compute, you want everything open; if you own cash, you want everything behind an API.”
Competition intensifies, but no one is backing down
No consolidation is expected. This is where I realized I might be wrong. One of the lab people told me that researchers at each lab are strong, the frontier techniques are widely understood, and new entrants like Xiaomi are increasing competitive intensity rather than reducing it. But no one has an interest in dropping out of the race willingly.”
The endgame analogy
So I asked, “ What is ‘to win’? The endgame, in the insider’s framing, will look like the auto industry rather than like a single winner-takes-most platform. Some will be Rolls-Royce with a low number of sales but extremely high margin, and some will be Mercedes, BMW, and Audi sit at the premium-meets-volume sweet spot. Mass-market players generate volume at thinner margins. The strongest model does not automatically win the commercial market — the winner is whoever maximizes price times quantity at their chosen tier. Developer, consumer, and enterprise positioning are each a bet on where that product peaks.
And for most Chinese labs, their simple goal right now is to win on speed over the big techs. They’re going to lean into lean organizational charts so they can iterate and ship faster.”
5 Big thinking
Jonathan Tonkin shares some lessons from ecosystems for a world obsessed with efficiency in this essay - worth pondering in full - do it here:
Some Takeaways
“We’ve optimised so much that our systems work brilliantly — right up until they don’t. When shocks hit, the impacts can be catastrophic.”
“Natural ecosystems face similar uncertainties, extremes, and surprises. But many of them persist for centuries. One reason is that they don’t optimise around a single best strategy. In variable environments, systems persist by spreading risk across multiple dimensions.”
“…species respond in different ways to change — what we call response diversity. As a result, their populations don’t all rise and fall at the same time. When you average across them, these asynchronous dynamics smooth out the variability, making the aggregate measure of the system (e.g. overall abundance or biomass of all life in the system) more stable.
In some cases, species fluctuate in opposite directions — declines in one are offset by increases in another. These compensatory dynamics further stabilise the system, but they’re not required for the overall buffering effect to emerge.
Each species in an ecosystem performs many functions — pollination, carbon sequestration, nutrient cycling, predation, water filtration. These are what ultimately lead to ecosystem services — things like clean water for drinking, food, medicine and so on. In a system with high functional redundancy, when a species is lost from the system, another fills its role. The functioning of the system is resilient to ups and downs in the environment.
Diverse systems don’t avoid shocks — they avoid failing all at once.”
“Ecosystems do not operate in isolation. We’ve learned over recent decades just how interconnected life is. What happens in one ecosystem is the result of processes operating at a range of scales. Species move in and out of habitats, resources are exported across boundaries, and nutrients are brought in from elsewhere.
Once we begin to consider these spatial dynamics, we start to see how crucial the spatial mosaic of the landscape is for resilience of ecological systems. For instance, the idea of source-sink metapopulation dynamics (metapopulation = collection of populations) emphasises how some habitats are sources of recruits (where reproduction is possible) and others are sinks (where species are not able to reproduce for some reason or another). In this case, source populations top up the sink populations. We see this here in our local rivers where nonnative trout prey upon native galaxiid fishes in sinks that are fed by trout-free source populations.
When we begin to modify landscapes, much of the natural resilience mechanisms are removed. That’s why the Single Large Or Several Small (SLOSS) debate has raged on in ecology for decades (is it better to have one big patch of forest or lots of small ones?). For rivers, it’s important to allow them space to move — not just for flood risk protection but also for providing habitat variability and connections to the floodplain, all of which help species find places to hide during flood events or rare their young.
Intact connections among diverse ecosystems enables them to remain resilient in the face of variability.”
“Species don’t just respond to the environment in the moment — they store up gains, they delay responses, they wait. Species have a suite of inbuilt mechanisms that help them cope with unpredictability in the environment.
Some build up seed banks that lie dormant in the soil for years, then spring into life when the conditions are right. Desert landscapes that look lifeless can erupt into life almost overnight.
Others produce dormant life stages. Short-lived species may produce eggs or larvae that pause development until conditions improve. Many zooplankton rely on these “time capsules” to persist through droughts.
And species can store up the gains made during good times to get through the bad. This so-called storage effect is prominent in desert annual plant communities, where species trade off competitive ability and drought tolerance resulting in some plants growing better during wet years, others in dry years.
And finally, long life spans of course buffer species from bad times. If they can just hold tight and withstand the bad times, then offspring will come during the good. Not every year needs to be good — far from it.
Time, in other words, is another dimension across which risk is spread.”
“Many of the previous dimensions involve organisms with strategies that enable them to maximise their risk spreading. In particularly variable environments, species have evolved numerous ways to spread their risks.
One evolutionary strategy is to separate their life stages among systems. For instance, many freshwater insects have an aquatic juvenile stage and terrestrial adult stage, so they spread their risk among different places. Mayflies that emerge from streams become food for riparian birds, bats and insects, but often only for a day or two, so they minimise their exposure to terrestrial predators. Salmon prioritise egg and juvenile survivorship over adult survivorship by travelling inland to spawn.
Unlike salmon, rainbow trout spread risks by adopting different life history strategies. Some individuals remain in freshwater their entire life cycles, while others migrate to sea and return as steelhead. This partial migration spreads risk across environments. River and ocean conditions don’t vary in the same way or at the same time, so when one strategy performs poorly, such as during a heatwave or a poor year at sea, the population can persist.
Finally, while you expect massive hatches of mayflies across the United States in systems that have highly predictable river flow regimes, New Zealand insects avoid putting all their eggs in one basket to avoid catastrophe during one of the many unpredictable floods they experience.
These examples of bet‑hedging highlight how species spread risk across time when the future is unpredictable. Instead of betting everything on one strategy, these species spread their chances across different futures.”
“In sum, nature doesn’t eliminate risk — it distributes it across genes, species, space, time, and strategies.”
6 Patience
Have a Great weekend when You get to that stage,
Sune