Weekend Food For Thought
On today's menu: The US Economy, Half Way Check Of Global Economy, Unseen Infrastructure Of Chip Making, When To Do What You Love, and much more...
Hello from Lisboa,
I hope you had an interesting and productive week.
Claude, when prompted, states that: “Engineering epistemology positions the engineer as someone who models, designs, and intervenes from outside. But in social systems — organizations, cities, markets — the analyst is always a participant. Your model changes the system you’re modeling. Engineering naturally decomposes. Understand the parts, understand their interfaces, understand the whole. This works when the whole is the sum of its parts — which is sometimes true. But emergence means some behavior only exists at the level of the whole and disappears when you decompose. You can study every neuron and never find consciousness. You can analyze every market participant and never find a bubble until it’s bursting. The trap is subtle: decomposition works often enough that it becomes the default, and its failures get attributed to insufficient data or modeling error rather than to the method itself. The engineering view of systems is the right starting point for systems that were designed — where you have access to the architecture, where the goals are specifiable, where you can run experiments. It becomes dangerous when applied to systems that evolved — biological, social, ecological — where the architecture is emergent, the goals are contested, and interventions have consequences that don’t show up in your model because you drew the boundary too tight. The best systems thinkers tend to be bilingual: rigorous enough to build the model, humble enough to know what the model can’t see.”
Take Note…Get bilingual…
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 Depth is the new luxury
1 Getting Visual
Spotlight: The ability to make things: ”Productivity growth in American manufacturing stalled after the global financial crisis. Remarkably, output per worker in 2022 was lower than it was in 2010. Meanwhile, unit labor costs have risen by almost sixty percent. In 1988-2010, output per worker grew at 3.64 per cent per annum (%pa), while unit labor costs grew at just 0.13%pa. In 2011-2025 by contrast, output per worker grew at -0.26%pa, but unit labor costs grew at an astounding 3.07%pa. In short, US manufacturing firms are becoming globally uncompetitive. The great sucking sound of the compute boom isn’t helping. By bidding up factors of production across the board, the AI boom is driving up costs in tradables, in a process I have described as the generalized Dutch disease. At the center of world trade sits global manufacturing. The global competition over these high-value trades is the principal driver of innovation and productivity growth in the world economy. Not unreasonably, everyone and their mother wants a bigger piece of the global manufacturing trade. And everyone wants to climb the value ladder. Countries that are able to outcompete others for a larger share of the global manufacturing trade, and especially the high-skill component, win; others lose. This is a zero-sum game. And a brutal one at that. How could it be any other way?” - Policy Tensor
Spotlight: The US Economy: Fiscal taps are open:
Spotlight: The US Economy: Flooding the Zone: M2 is expanding rapidly.
Spotlight: US Grid - A Lack of Power: “A grid running on borrowed time. A quarter of US distribution transformers have outlived their expected design lifetime of roughly 30 years.” - FT
Spotlight: US Grid - Lacking the Transformers needed for Transformation: “Demand for transformers and other power equipment is surging as AI data centres compete with renewable energy projects, transport electrification and industrial expansion for scarce supplies. Average lead times for transformer orders, once measured in months, now stretch into years. Prices for certain large devices have almost doubled post-pandemic. Analysts expect the supply crunch to last until at least 2030. Transformers still rely on the same core principle of electromagnetic induction discovered by Michael Faraday in 1831. Advances in materials and design have allowed the largest devices to handle ever higher voltages, supporting the expansion of national grids. “They take a lot of engineering,” says Hitachi’s Fleming. “They take a lot of design time. They take a lot of craft and technical know-how in order to build by hand.” With up to 80,000 different designs, most transformers still have to be built largely to order, taking three to six months to make. The most demanding stage is the windings, when copper wire is applied around a transformer’s core — a “beautiful” process, according to Bruno Melles, an engineer now leading Hitachi’s global transformer business. Each winding is unique and is “still a human manual activity that we’re very proud of”, he says.” - FT
Spotlight: China’s Power play: “Our China, Commodity, and Geopolitical strategists argue that China’s grid advantage is becoming a structural edge in the new age of electrification. The key global bottleneck is no longer power generation, but transmission and storage. China’s scale in electricity infrastructure is lowering industrial power costs, improving energy security, and strengthening its competitive position in manufacturing and AI. That advantage is also becoming an export edge through energy-intensive goods and potentially AI services. The US remains competitive through innovation and may start responding with policies to expand grid capacity.” - BCA
The Big Picture: The Energy Transition: More solar capacity was added in the past 20 months than in the previous two years.
Spotlight: Rationality via Checks & Balances: “No country has put more concrete policy questions to a vote than Switzerland: over 7,700 at national and subnational level. What does this multi‑century experiment in direct democracy actually show? From 1793 to 2026, voters have said “no” at the national level in 53% of all cases. The chart tracks the net balance of outcomes – “yes” minus “no” – over time. The cumulative weight of “no” decisions has steadily pulled the line down. The takeaway is that it has been easier to block major change than to secure broad consent for it. The Swiss system tends to move in small steps. - Costa Vayenas
2 If You Read One Thing Today - Make Sure it is This
Take time to check the state of the global economy half-way through 2026 - The good people in the Tower of Basel AKA the BIS has put together their Annual Economic Report - it’s always a good place to start…Do it here in full:
https://www.bis.org/publ/arpdf/ar2026e.pdf
Some Takeaways
“Growth held up well in 2025, despite significant headwinds from higher tariffs and geopolitical uncertainty. Three factors stand out.
First, the drag from higher trade barriers was lessened by effective tariff rates that were lower than initially anticipated, trade diversion and firms’ willingness to absorb costs through lower margins.
Second, a wave of optimism about AI spurred a surge in capital expenditure on AI infrastructure, lifting investment in the United States with spillovers along global supply chains.
Third, animal spirits about AI lifted stock valuations, sustaining favourable global financial conditions. Yet this resilience was soon tested in early 2026, when the closure of the Strait of Hormuz delivered a major shock to global energy supplies. Rising energy prices once again pushed inflation well above central banks’ targets, echoing the post-Covid-19 inflation surge. Although the recent conflict in the Middle East seems to have abated, the economic effects of the Hormuz disruption may linger as the full restoration of physical energy supply takes time and the initial price increases propagate through supply chains. The closure of the Strait of Hormuz has raised the costs of manufacturing and agriculture inputs, with potentially dire consequences for food prices and food security among the poorest countries. Despite these challenges, financial markets have remained buoyant, reflecting expectations that the disruptions would be short-lived and the AI boom would continue.
Looking forward, four pressure points demand attention.
First, inflation has risen. The energy supply shock has been substantial, and its effects may propagate through supply chains. Global headline inflation picked up shortly after the conflict in the Middle East began, and prices of plastics and fertilisers – key inputs – have risen by 30% and 50%, respectively. The central question is whether these initial price increases will broaden and persist, as during 2021–23. On the one hand, mitigating factors limit second-round effects. Greater slack in the labour market may help to contain wage pressures. And policy rates are higher now than in 2022. On the other hand, memories of the post-pandemic inflation surge are still fresh. Given that it will take several quarters to purge the imbalances in oil physical markets, further volatility in energy prices could arise. In turn, inflation expectations could de-anchor more quickly than in the past.
Second, the optimism surrounding AI may not last, despite its promise of future productivity gains. The current surge in capital expenditure could prove unsustainable if supply bottlenecks restrain production. Intense competition for market leadership may fuel overinvestment further, as seen in previous innovation waves, increasing the risk of a sharp reversal if AI payoffs disappoint.
Third, financial vulnerabilities persist. Easy financial conditions could tighten and become a potent amplifier in adverse scenarios where interest rates rise and AI payoffs disappoint. Compressed risk premia and stretched valuations highlight the scope for unwinding. Increasingly opaque financing of AI activities, high leverage in core markets and the growing footprint of private credit further undermine the resilience of financial markets. The current tension between exuberant risk appetite and elevated macroeconomic risks could unwind abruptly.
Fourth, fiscal pressures are mounting. With already high debt levels, governments face rising demands for spending amid energy shocks and geopolitical tensions. These rising pressures coincide with a less benign financial environment than the one prevailing in the aftermath of the Great Financial Crisis. Moreover, GDP growth has also slowed from post-pandemic peaks. Consequently, interest payments as a share of GDP have risen across many countries. Compounding these fiscal challenges, the financial structure of sovereign bond markets has become more fragile, as explored in detail in Chapter II.”
“In the near term, the ongoing AI investment boom raises questions about the sustainability of the current economic expansion. The five largest hyperscalers are set to spend over a trillion US dollars on AI-related capital expenditure from 2025 through 2026. These commitments are outpacing earnings and the free cash flow of these firms, leading some to issue debt to raise additional financing (Graph 11.A). This investment race may be partly driven by the perception that only a small number of players with superior technology will ultimately dominate the market shares. The intense competition raises the risk of firms over-committing resources to investment projects with still uncertain returns, leaving all firms vulnerable to disappointments in AI payoffs. Model analysis based on such contest motives highlights the downside risk of current AI exuberance. As competitive pressure drives capex higher, the net economic surplus – the total payoff less investment costs – declines for the sector as a whole and could turn negative in adverse scenarios (Graph 11.B). Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust, with potential knock-on effects on financial conditions (see below).
Another risk is that the AI boom runs into a supply side roadblock. The AI build- out has recently been facing growing bottlenecks in electricity, advanced semiconductors and grid equipment. Fast-growing demand for computing power is already pressuring electricity prices and input costs, with potential spillovers to inflation. Looking ahead, these temporary shortages may also amplify over- investment, as firms attempt to lock in future capacity through long-dated contracts that further expose them to any disappointments in demand.
Historical episodes of investment booms offer instructive parallels (Graph 11.C). The canal mania of the 1830s, the British railway mania in the 1840s, the electrification exuberance of the late 1920s (roaring 20s) and the dotcom boom of the late 90s all shared one common trait: a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify. These episodes ended with an eventual reversal in investment, inducing economy-wide recessions. The scale and pace of the current AI investment boom accompanied by expectations of large productivity payoffs bear resemblance to these precedents, highlighting potential downside risks in the near term.”
Financial vulnerabilities as amplifiers
Should inflation rise significantly or AI-led investment turn to a bust, the macroeconomic consequences could be amplified by existing financial vulnerabilities. A tightening of policy rates needed to contain inflation could precipitate a sharp pullback in asset prices after a prolonged period of exuberant risk-taking, triggering disruptive macro-financial feedback loops. A reversal of AI optimism could likewise have major financial consequences, given AI firms’ rising leverage and growing footprint in credit markets. Vulnerabilities extend to their supplier ecosystem, including engineering, procurement and construction (EPC) contractors whose balance sheets are comparatively weak, leaving them exposed to any capex pullback by hyperscalers.
“Equity valuations are elevated, particularly for firms at the core of AI development. The implied long-term earnings growth for the largest corporations sits well above recent historical benchmarks (Graph 12.A), with US stocks often trading at large premia to peers in other major markets. These implied rates often exceed even the elevated growth that some of the technology firms have delivered in their relatively short lifetimes. As these firms mature and command a larger share of the market, sustaining such high growth could become increasingly challenging.
Sentiment has also been a major driver of current valuations. Risk premia on the largest US stocks have compressed markedly since the Covid-19 pandemic, with the distribution shifting clearly to the left (Graph 12.B). This points to growing investor complacency and reduced compensation for risk-bearing. Post-pandemic exuberance has been largely broad-based across sectors and countries, coinciding with the rapid rise of AI as an investment theme following the release of generative AI tools in late 2022.
A major equity market correction could have larger macroeconomic consequences today than in the past. Household equity exposures have grown over the past few decades, both relative to total wealth (Graph 12.C) and income (Graph 3.C). A large correction in valuations could have more pronounced wealth effects and sharper consumption pullback than in the past. And with US stocks accounting for an outsized share of global equity markets – about 64% of the MSCI Global index – the wealth impact from a US-led repricing could propagate globally.
Financial stability could also be at risk in the event of an AI bust. Fixed income markets are one obvious vulnerability, given the high volumes of debt issued by hyperscalers, AI labs and EPC firms (Graph 13.A). Should hyperscalers slow or halt the aggressive pace of capex deployment, many borrowers across the supply chain could struggle to replace lost revenue and service their debt. The credit spreads of some AI firms have already begun to widen somewhat to reflect this risk (Graph 13.B), even as equity markets continue to price in significant upside gains.
The opacity of AI-sector financing compounds these vulnerabilities. Hyperscalers, chip makers and AI labs are linked through a complex web of private arrangements.
The most prominent is circular financing: chip makers and hyperscalers take equity stakes in AI labs or neocloud providers, who in turn commit to multi-year purchases of chips or computing power. Data centre construction is increasingly outsourced to third parties that lease facilities back to hyperscalers on long-dated contracts with embedded exit clauses. The terms of such deals are typically poorly disclosed, with risks of the same asset being pledged multiple times. Together, such arrangements account for a sizeable share of sector-wide financing and forward revenue (Graph 13.C).
A sharp repricing of equity risk could prompt a reassessment of corporate credit risk and lead to tighter credit conditions more broadly. Indeed, broad indices of credit spreads tend to correlate negatively with stock market returns (Graph 14.A), more so for the high-yield than the investment grade segment. While large, synchronised corrections in both markets are rare, there are notable precedents such as the Great Financial Crisis and the March 2020 dash for cash episode. A repricing of risk this time, whether triggered by higher interest rates or an AI bust, has the potential to be similarly disruptive by triggering a corporate credit freeze with wider implications for aggregate investment.
Any tightening in credit conditions could expose existing vulnerabilities in the less transparent private credit space, whose reach has expanded among middle market and small firms (Graph 14.B). Signs of stress are already visible: direct lending funds catering to retail investors have faced mounting redemption requests, forcing some to liquidate assets and return capital despite having no contractual obligation to do so. A larger shock, whether from a renewed inflation surge or a sharp AI-led repricing, could trigger a more widespread credit crunch.
The consequences could extend beyond the non-bank perimeter, given banks’ growing and opaque exposure to private credit funds, compounded by overlapping ties through insurance companies’ balance sheets. Given that the affected segment is smaller corporates that account for a large share of job creation, the real economy implications could be substantial.
The growing role of private credit also raises concentration risks. Direct lending funds, dominant players in the private credit ecosystem, have quadrupled their lending to the AI and information technology (IT) sectors in the past five years, to about 15% of their portfolios. These loans tend to be larger than those in other sectors, while their terms such as tenor and pricing remain broadly similar, raising questions about lending standards and risk pricing. Investor enthusiasm has allowed more funds to participate, increasing concentration risks as software firms draw on multiple private credit lenders simultaneously (Graph 14.C).”
3 Consequential Thinking about Consequential Matters
ChinaTalk explores the “unseen infrastructure of chipmaking”: The Chemistry…It’s a conversation on consequential matters worth thinking about - Go do explore it in full here:
Some Takeaways
“What’s the hidden ingredient behind every advanced chip? It’s not just the silicon, the machines, or the engineers — it’s the world of specialty gases that make modern semiconductor manufacturing possible. These chemicals are purified to extraordinary levels, shipped across continents, and in many cases toxic and explosive.”
“On the application side, helium is used by every semiconductor manufacturer in every fab in every location in the world, and it’s primarily used for cooling. A lot of these semiconductor manufacturing processes are quite violent, they’re quite exothermic, even though it looks very calm and silent from the outside. There’s a lot of cooling required to keep the manufacturing process at reasonable and workable temperatures, and that’s the role of helium mainly in these fabs.
In terms of why the Strait of Hormuz issue has been so significant — the Ras Laffan Qatar helium production facility now produces about 15% of the world’s capacity. It was an overnight closure of the tap, essentially, of 15% of the world’s capacity that generally operates at or around equal supply between production and demand.
This is not an example of a molecule that’s got a huge overcapacity that’s very easy to just take up slack from alternative sources and deliver to a fab. It’s also a very difficult molecule to deliver. You need to move this around in the world’s most expensive thermos flask at -269°C. That whole logistics supply chain is extremely complicated as well. Just turning off the ability to move those assets, plus the production that comes from that site in Qatar is a pretty major disruption.”
“You look at your phone, and it’s made of plastic and metal. People don’t see the inputs required to make that in these fabs — which is all gas. There are probably 120-odd different chemicals that go into a fab from different suppliers. There are probably 60 unique chemicals that go in, and they’ll all basically go in in gaseous form.
It’s a huge chemistry toolkit that’s been developed and is required to build all chips.
This is not just future advanced technology. This is every chip everywhere that needs this chemistry set.”
“Silicon gets deposited via a gas called silane. These deposition steps happen over and over again, creating a repetition in how you build these structures. You then need to knock some of that structure out using fluorine-based gases to etch and remove portions. You need doping gases to change the electrical properties of some layers. You need to build the transistors in there, which requires different types of gases and processes.
It’s an extremely complex, almost unbelievable manufacturing process, all made possible by the use of gases. This underlayer — this supply chain and chemistry set — is not very well understood as an input into this industry.”
“…split them into two categories. Any major fab needs a huge amount of what we call bulk gas. These would be the inerting gases — the gases used to create the clean environments needed to make chips. That would be nitrogen, argon — things that don’t react or essentially don’t do anything in the process apart from keeping things clean.
Then you’ve got the specialty gases. The bulk gases would typically be delivered by gas companies with an on-site air separation system. The volumes required are massive, so they dictate that you would need a plant built next to your fab that would deliver these gases 24/7. We’re talking about tens of millions of dollars of investment by the gas company.
Once they’re built — and they get built wherever the fab goes — there’s a relationship with the gas company that extends over 15 to 20 years because nobody builds a competing air separation unit next to your fab. There’s a relationship anchored around the bulk gases.
On top of that, you’ve got 60 or 70 different process gases that are typically delivered in a range of different containers. You’ve seen these larger containers rolling down the highway, cylinders, and packages smaller than cylinders. They deliver the rest of these specialized gases for deposition (putting new material onto a wafer), ion implantation (which changes the electrical characteristics), and etching (which shapes and helps create the devices you need on silicon).
The toolkit’s very big. A typical fab will be taking these materials from every geography, probably with the exception of China. If you want to talk about Taiwan as one example, they’ll be receiving their bulk gases from the on-site supplier, but the balance of materials they need to run their fabs will probably come from at least another 4 or 5 countries. It’s extremely complicated logistics and supply chains to get that balance of the other 60 to 70 materials needed.”
“Obviously, as node sizes decrease and the complexity of these devices increases, they want to ensure that every input into the manufacturing process has higher and higher purity. At the extreme is the silicon wafer at 11 nines, which is industry speak for almost impossible. Only two companies in the world currently can achieve that.
The gases range from parts per million, which is now completely standard. A one-part-per-million impurity is really just the basic requirement to be a supplier of any material into this market. Some gases are now reaching parts per billion, and some even get into parts per trillion.
Parts per trillion is very difficult to imagine. It’s basically a heartbeat in 32,000 years. For the entirety of anything that’s been known to be living on this planet, it’s a heartbeat within that scale of time. That’s the level of impurity we’re talking about. That’s how to think about how difficult this is.”
Jordan Schneider: How do you measure parts per trillion?
Carl Jackson: With extremely expensive analytical instruments that hardly anybody knows how to use properly. It gets into ultra-specialist and extremely expensive equipment that’s so sensitive it makes the whole job very difficult. Ultimately, that will be the limitation. There’s a cost implication of increasing the purity of these products, but ultimately it’s the metrology around it that will be cost-prohibitive.”
Chris Miller: How do you get these gases? It’s different for different types of gases, but if you want a gas purified to 1 part per trillion level, how does that happen?
Carl Jackson: Well, that’s the extreme, Chris. But let me give you one example. The other thing to say is quite a lot of this chemistry set is also used in industrial chemistry.
One example is HF, or hydrofluoric acid, which is made in the hundreds of thousands of tons. That material starts life in a mine — it starts life underground as a material called fluorspar. It’s literally dug out of the earth at concentrations that you would never recognize as suitable for a semiconductor fab.
It’s then made into an industrial-grade HF. At that point, it isn’t destined for a semiconductor fab. That’s used, as I say, on a scale of hundreds of thousands of tons industrially.
Then the semiconductor guys get involved. They’ll take a small part of that volume and start to process it using very different types of purification systems that have been developed around the requirements for semiconductors. You start with a typical industrial-grade HF at 3 nines purity, and that’s good enough for almost everything required on Earth that uses HF — for pharmaceuticals, for all of the different industrial applications, but not for semiconductors.
That material then needs to go from 3 nines up to 6, 7 or 8 nines, depending on the process. That’s where the semiconductor supply chain, the people involved in this, really add their magic. It’s how you get and maintain and then ship and have that high-purity product delivered all the way to the wafer through this supply chain. At that point, it becomes very specialized, and you need to rely on a very small number of people in the world who can do that.”
“At the toolmaker level, they will typically put performance guarantees around a particular recipe set and a particular purity for that recipe set that will work. When it gets into the hands of the guys who are actually making the devices, then it goes into a different world. Some customers don’t change. Some customers are pretty active in terms of demanding better materials.
But it’s slowing down. The parts per trillion thing is really extreme — it’s an outlier. I’m certain that all of this standard chemistry set isn’t going to get there. Some specialist materials may be for some ultra-specialist devices, but generally people are not going to be demanding that. It will stay in the parts per million, parts per billion.
People forget that parts per million is an unbelievably difficult thing to deliver for a material that is — for example, HF will try to eat the cylinder that it’s being delivered in.”
“One of the fascinating parts of this is that it seems like half the materials we’re talking about are either toxic or explosive or both. Why is it that you can’t make chips without all these seemingly awful chemicals? And what challenges does this impose in terms of transporting them?
Carl Jackson: Almost everything in this toolkit is lethal. It either poisons you, explodes instantly upon contact with air, or can kill you quickly or slowly. But lethal here is not a bug — it’s the specification.
If you’ve been to a fab or seen pictures or videos, you see this very quiet, controlled, seemingly delicate process going on to make these chips. People envisage it as delicate. But at the atomic level, it’s an extremely violent process. The violence requires the most violent chemicals and the most reactive chemistries.
This deposition and etching process that builds these layers — you’re trying to drill the equivalent of an elevator shaft down through 300 stories in one go, straighter than anything imaginable physically. You need something that’s extremely reactive in the process to allow you to do that.
Similarly, when you’re doing an ion implant to change the electrical characteristics of the silicon so that it actually works and burns the transistor, you’re firing boron, phosphorus, or arsenic — all of these lovely gases that you need to do some work. You’re firing these things in at 400 kilometers a second into the silicon wafer. Most of the molecule disintegrates when it hits this wafer. It’s like firing a bullet into a concrete block. That bullet needs to go exactly where you need it to in that concrete block — the depth, the angle, everything. Then it disintegrates and leaves something behind that does some work.
The process itself is extremely reactive and violent. That’s why you need these chemistries. Which is unfortunate because making them and delivering them safely is a little counterintuitive.”
“Chris Miller: What’s the most dangerous chemical used in chip manufacturing?
Carl Jackson: There are so many dangerous chemicals. The one that people fear most is HF (hydrofluoric acid), largely because there have been a few incidents with it. It’s used in fairly large volumes. What makes it particularly frightening is that it doesn’t immediately kill you — it gets into your skin and then pulls the calcium out of your bones over time, essentially killing you slowly.”
“Aqib Zakaria: Let’s discuss these potent chemicals that are sometimes delivered at mind-warping purities. Who are the companies actually delivering them? How many are there, and what differentiates one from another?
Carl Jackson: It’s best to look at this over time because the landscape is changing rapidly — almost all of it happening in China.
If we go back 10 to 15 years before China’s Big Fund initiative to localize semiconductors and their supply chains, the market was segregated by chemistry specialty. Japan historically excelled with fluorinated gases — anything with a fluorine backbone. There are probably 15 different fluorine-based chemicals and chemistries used in semiconductors, and Japan was the go-to source for these.
The US had a pretty broad range of manufacturing capabilities since that’s where this industry originated. Twenty years ago, there was near self-sufficiency in almost everything required. While that’s somewhat depleted now at the scale and scope needed, that capability existed then.
Most of these gases were typically made by a few Japanese companies, but the majority came from the majors: Linde, Air Products, and Air Liquide. These companies that built and installed air separation units also provided all the other required chemicals. It was a self-funding, self-financing model — you’d install the air separation unit as the anchor that paid for developing all the other materials.
They offered a complete portfolio to each semiconductor customer. Twenty years ago, almost everything came from one of these majors with their comprehensive portfolios.”
“Chris Miller: So you mentioned that the industry is changing. Tell us what’s new in the gas industry and the role of China.
Carl Jackson: The Big Fund is really what’s driving China’s transformation at this level. Big Fund 1 kicked off around 2014, and we could see it was different from the start. Unlike other similar initiatives — the CHIPS Act, for example — the Big Fund targeted all levels of semiconductor manufacturing.
I won’t speak to areas outside my expertise, like design and tool manufacturing infrastructure, but all of that was included in the Big Fund. You’re not just getting new fabs and new IDMs making devices — you’re getting the entire supply chain. They went after everything at once.
There was a lot of money. About $120 billion has been invested in that whole infrastructure ecosystem. It was managed regionally rather than through top-down directives saying “you do this over there, and you do that over there.” While there was an overall top-down directive, implementation was managed on a regional basis.
This led to different regions competing with each other for all the different materials in this chemistry toolkit. Take one product: NF3, which is part of the fluorine family. It’s used for cleaning — it’s like the janitor of the semiconductor world, used by everybody in every process. At the time, China didn’t have any NF3 capability and needed to import it.
Different provinces with different leadership and scopes started developing supply chains for these materials in parallel. This resulted in massive overcapacity. You had new players who had never been active in producing these materials or supply chains.
Initially, it was the whole story of China — low-quality materials, questions about trust in where and how things were being made. That took a long time to overcome.
They started out feeding themselves, but because of how the Big Fund was organized, they ended up with huge overcapacity. Now you’ve got China plus the rest of the world in terms of their ability to provide these 60 gases.
They’ve developed the capability to make almost everything in China independently and in parallel with everybody else in the world — with a capacity that could almost feed everywhere in the world from China.”
“Chris Miller: Do we see Taiwanese or South Korean firms buying gases from the Chinese supply chain?
Carl Jackson: Taiwan is 100% reliant on Chinese supply chains today — 100% reliant. Chris, you talk about missiles, but I could just change that to NF3 as an example.
If the Chinese government decided to put an export restriction on NF3, then the Taiwanese fabs would shut down.
Chris Miller: Help us understand how we should think about finding alternative sources of supply. I think back to 2022 at the start of the Russia-Ukraine war — both countries were big suppliers of neon. The Azovstal steel plant, where there was a big siege at the start of the war, was itself a neon supplier. That created a crisis for the semiconductor gas industry, but it didn’t end up impacting chip production much at all. This suggests there was flexibility in the neon market. Is that unique to neon? Is it different from NF3?
Carl Jackson: Helium is probably the best example of a supply chain that’s inelastic, mainly because of the packages. The most basic package to get gases from the supplier to the consumer is a cylinder. These cylinders cost $50, and there’s no shortage — millions circulate around the world.
The helium packages I mentioned before cost $1 million each. They’re extremely complex. You’ve got 45 days to get from production to consumption before you start to lose the product. That’s pretty inelastic.
On the other extreme, take NF3, for example. Chinese domestic consumption requirement for NF3 is about 8,000 tonnes. You have one producer in one province in China making 55,000 tonnes a year. Those guys are busy recruiting international sales and marketing teams and trying to sell their NF3 internationally — into Taiwan, South Korea, the US, anywhere that can take that material. They’ve got almost zero incremental cost of selling anything above what’s required domestically.”
“Chris Miller: Absent the subsidy and overcapacity story, is there a reason why you’d produce more NF3 in China than elsewhere? Energy costs or some sort of comparative advantage, or is it solely a story of governments paying for it?
Carl Jackson: It’s a mixture — a blend of things. NF3 is part of this fluorine family. Fluorine comes from fluorspar, and China controls 70% of the world’s fluorspar. That whole supply chain is integrated inside of China, allowing them to do everything internally at an extremely low cost.
Some of this comes down to economies of scale. Some of it involves ignoring domestic requirements and building out huge capacity for some future need. Some decisions simply don’t make economic sense from a Western point of view.”
“Aqib Zakaria: I’m curious if there are any cases of it going the other way. We’re pretty accustomed to stories like rare earth minerals and now fluorspar where everything comes from China, creating great dependency. But are there any cases among these 60 or so gases where the mine isn’t in China — maybe it’s in Peru or somewhere else — and China has to import all of it? Are there cases where the dependency goes both ways?
Carl Jackson: Helium is an interesting challenge for China. From a Chinese perspective, helium represents the biggest gap and most high-value, important material for semiconductors. The reason they don’t have enough helium is purely geological. The US is extremely lucky with great geology for extracting helium, and anyone making LNG also has great capacity for making helium.
Helium typically isn’t produced on its own — it’s a co-product from something else. Until recently, nobody mined or extracted helium specifically. You extracted LNG, and helium happened to come along with it. When helium started gaining high-value, important applications like MRI and semiconductors, people became more interested in separating and producing that helium.
China has a challenge — the amounts of helium in their LNG production are extremely small. Where they produce it is basically all in the wrong place, distributed around a couple hundred different small sites. Their LNG infrastructure isn’t conducive to making helium, so they’re still a net importer of helium.
That’s being fixed. In the Chinese way, they’ve figured out ways to produce their own helium despite these restrictions. But it’s going to take time to develop the infrastructure they need to stop being a net importer. At the moment, that’s probably their biggest vulnerability for the semiconductor supply chain. But as I said, it will get fixed — they’re busy trying to figure that out in really smart ways.
Taiwan and Korea are in a similar position where the geology doesn’t allow economically for helium production. They will always be 100% reliant on helium imports, plus almost every other chemical needed to run these fabs forever, unless they do something they haven’t yet done in 30 years. Their supply chain vulnerabilities will remain when China’s fully fitted out.”
“I haven’t seen anything at all yet about how the supply chain and the security of the supply chain and the chemistry supply chain that feed those fabs and any new fabs. All of these domestic production requirements are being announced, but without any of the supply chain underneath. If nothing changes from the US perspective, then there will be total reliance on importing almost every chemical required to build those new capacities.
On the Chinese side, they’ve done it differently. As we mentioned before, they’ve set out to put the entire ecosystem in all at once. It’s not just new fabs built by existing suppliers — it’s new companies entirely building new fabs, creating new technology, and underneath that is the full and complete supply chain to put all the materials required to build those components in, fully domesticated.
The scope and scale of the Big Fund are much larger and, while not perfect by any stretch, it’s certainly more integrated and forward-thinking than CHIPS in its current form, which looks like it’s either ignored, not quite gotten to yet, or isn’t interested in the supply chain vulnerabilities underneath those new fabs being installed.”
“I did some work on this maybe 10 years ago when we were trying to figure out the value of the products that we were selling into an iPhone. For each iPhone manufactured, how much value of gas is in it? What’s the weight of the gas in there? Because you’re trying to explain that this is made with gas, and people say, “It can’t be. It’s solid. What are you talking about?” We tried to figure out how we could contextualize it a little bit more. Then we asked how much is in a Tesla? It’s about 10%. But again, it depends on the structures and devices you’re trying to make. That’s one way to look at it.
The other way to look at it is — which one of those 60 can you live without and still manufacture your products? The answer is zero. None of them. You need all of them — exactly when you need them, exactly at the spec that you need them.
The vulnerability is there. It’s not about how much you need or what the breakdown is or what it’s worth. The question is, what do you do if one of those is missing? How do you run your fab?
It looks even worse because if you break the 60% down into 10%, you might have a line item in your materials that’s $100,000 a year for one gas that is barely used. If you’re missing $100,000 worth of gas, you can’t run your fab. The output cost — the opportunity cost or the downside — is huge in relative terms to the input gases. It’s more important that you get them there and in spec than what you pay for them ultimately.”
“Carl Jackson: Taiwan — and I say this as someone who lives there and loves the country — is arguably the single worst location you could pick for semiconductor fabs. It has no natural resources, water is an issue, and there are significant geopolitical risks. They had a brilliant industry pioneer in Morris Chang, but if you were to look at a globe and choose where to drop a semiconductor industry, Taiwan probably wouldn’t make the list. It’s in seismic zone 4 — the earth moves regularly there. Not ideal for making these kinds of devices. But it is what it is, and they face existential supply chain risks.”
4 Big Ideas
Grace Shao takes a “Hangzhou Alibaba field trip” and explores “Five aI business products” to understand “who will stand?” - Go explore with her by reading the piece in full here:
Some Takeaways
“I got into a taxi and started chatting with the uncle driving. If I hadn’t been back in five years, he told me, then too much has changed. New highway bridges, rows and rows of new development, and of course, the completed new campuses for Alibaba.
So I asked him, “Does that mean housing prices have gone back up with the AI boom?” “
Oh no, no”, he said, only a few are buying up luxury apartments, but they work for semi companies. He added that he heard kids working in AI are making 2 to 3 million RMB a year, which seemed like a crazy number to him. But those kids working in AI are so tired that they get into his car and pass out, and he has to wake them up at their stop. In awe of their talent, but also worried about their health, as one would expect a typical Chinese uncle to be.
Now, I was flabbergasted at the response from this uncle, frankly, who could barely speak Putonghua clearly, but had the sophistication to know that only the semi companies, their investors, and everyone along that supply chain have made a lot of money this year, and that the rest of tech and AI is still mostly waiting for its turn.
In some ways, that taxi driver in his 50s in Hangzhou had a more accurate understanding of the cycle than a lot of professional analysts.”
“You know how tech companies have sprawling campuses? This whole city is an Alibaba campus. The train station itself was reportedly built in part to make the Shanghai commute more convenient for Alibaba staff, and apartment complexes went up around the new developments Alibaba established. The spending is crazy, and it cuts both ways. On the one hand, it shows confidence in the company’s direction; on the other, when you speak to investors, there is a level of skepticism because they are wondering how much of that money should have been spent versus how much should have been returned to shareholders.”
“The Qwen app is extremely impressive in how it connects all the commerce activity, even though we wrote about how it hasn’t fundamentally changed how users engage with the idea of shopping yet.
But this is Alibaba’s edge if tokenomics eventually comes down. Think about it, everything can be completed within its own platform. E-commerce, same-day delivery, grocery, maps, ride hailing, bookings, payments, all integrated, agentic, agent to agent. This is something we’ve already seen OpenAI couldn’t really figure out with its reliance on third-party partners. It’s easy to push a user to DoorDash and then to Stripe after they’ve shown intent to buy, but it’s much harder to have the agents link up so the full loop, from car booking to airplane to hotel, actually completes.
In this case, Alibaba owns the rails, so the loop closes all under its control. There is good and bad to that. The good is that this activity loop could even exist; the bad is that you bear all the liability, right?
The representative I met repeatedly reminded me that any purchases or transactions still require human verification to prevent mistakes, but if we are to push for an agentic commerce future, maybe that step will eventually be removed?
Anyway, what I found is that the Qwen app is more nuanced than a checkout bot. The app is memory-based, so when I want a dress, it already knows my size, my body type, and my usual preferred style, and unlike GUI-driven agents, it doesn’t need to operate a screen because it calls directly on stored memory. Again, this is something unique to Alibaba, having the Taobao purchasing data.
It behaves like a shopping guide rather than a fulfillment engine.”
“The team’s framing, which honestly I buy, is that China already app-ified every booking and ordering need over the past fifteen years, and it is much easier to AI-fy a mature internet infrastructure than to build agentic commerce on top of fragmented rails.
The payment step stays deliberately human, so the final purchase has to be personally confirmed, except for utility goods like the weekly cat kibble or toilet paper, for which you can choose to authorize a recurring order. Shopping has been live since May, and consumer usage is reportedly growing 2x month-on-month; overall data is not publicly disclosed yet as it’s still in its relative infancy stage.
One example that really impressed me, and that I would personally pay for, is planning a day trip from Shanghai to Hangzhou. All you need to do is tell the agent the time you want to arrive and your preferred transportation, and it will plan out your time, your snacks, your coffee, and everything in between, frankly, even more meticulously than a human secretary.
But right as I was getting excited about this, it dawned on me: how can you justify the cost of token usage for such simple activities?
I would pay for an app to manage my itinerary, but how much would that token cost, and would it be worth it? I guess it really depends on how expensive your own hourly rate is.
And there is a second problem that pricing can’t solve, which is liability. If there is traffic and you miss your train, will you blame the AI or your own judgment? Who bears that risk, Alibaba, the airline partner, or you?
The bigger question here isn’t really an Alibaba question at all. It’s a question I have for the whole industry: how much is a consumer actually willing to pay for agentic support?”
“Just as Azeem’s team wrote in the recent report The State of the AI Economy, “The AI economy is bigger and faster than any technology wave before it, yet still small enough to be early.”
That is how it felt. Qwen, in many ways, felt ahead of the curve. But maybe the infrastructure cost-benefit just isn’t mature enough to justify such a use case yet? Because the stock price has not been a friendly reflection of this vision. That same report shows that AI is scaling three times faster than any prior IT wave, but much of that is from the enterprise level. Who is bearing the cost for consumer usage?”
“TBH, my real goal in going to Alibaba this time was to learn more about Accio, and it was the highlight of the day, a business I find very undervalued. I read about it a year ago, but at that time it was still merely described as a merchant support AI tool.
I explained on the ~100-investor JPM call on Friday that it really undersells it, especially now after a year of iterations. On the surface, it looks just like any other agent service for merchants, and the interface even looks like Claude. But it relies heavily on Alibaba’s sprawling supply network, which is its ultimate advantage.
Accio is less an “AI product” and more an AI UX on a twenty-year supplier data moat. This is the clearest enterprise 2B use case Alibaba can push out right now, and frankly I don’t even think it’s getting enough attention internally at this point.
The value of Accio isn’t just the AI agent. It’s that the AI sits on top of 1.5 million verified suppliers, 400M+ SKUs, and decades of transaction and rating data that Alibaba owns because it is the B2B marketplace. There are already millions of MAUs.
And there is also a reason that there is no Accio equivalent in the West, because the supplier graph simply doesn’t exist there for anyone to build on. No Western company has that asset. Even Amazon can’t do this.
For a merchant, it is essentially everything in one; it shocked me how it removed the barriers for business/ drop shipping and all the tedious search and due diligence work that is needed to start a brand.”
“It starts with supplier finding. The agent sits on the supplier graph and can find, filter, and talk to vendors for you. Autochat can speak to vendors through one interface, say, for sourcing just the lid of a bottle, run the first round of filtering against your core requirements, and then hand off to you to speak to the shortlist directly. Sometimes it is literally Accio talking to Accio on both sides of the negotiation. And the sourcing guidance is real know-how, not chatbot filler. One buyer didn’t know the measurements or the requirements for their product, and the agent guided them through the safety requirements, the weather conditions, and what those mean for raw material choices- the kind of professional judgment that lives in the network and twenty years of transaction data, frankly. It’s not more intelligence but more know-how. [Industry-specific use case: AI tools are the future is what we’ve been writing about!!]
From sourcing, it moves into pricing and margin. The agent negotiates pricing with vendors behind the scenes over multiple rounds, then benchmarks against industry comparisons and hands the merchant a rough margin before they’ve committed to anything. This is the part I keep coming back to, because it means a small merchant gets the pricing intelligence that used to require a sourcing office in Guangzhou.
Then the operational layer: supply and inventory management, plus the back-office work around it. Accio Work claims to handle VAT filings across 100+ markets, multi-round supplier negotiation, and logistics ops, though how much of that is truly AI-executed end-to-end versus AI-assisted with a human in the loop is one of my open diligence questions. On the storefront side, it covers CRM and platform management, lead generation, content management, and SEO and GEO optimization. It connects to Mailchimp, WordPress, and the like, where some opened up access to Accio, and some skills were built directly by the Accio team, so the experience is smoother, and it is opening up WhatsApp and Telegram so your vendor conversations get summarized straight back into Accio instead of dying in a chat thread.
And then distribution. Like HELLO. Plugins are already live on Amazon, Shopify, Etsy, and Shopline, and you can swap the 1688 agent for an Amazon seller toolkit depending on where you sell. The data runs both ways, too, internal Alibaba data from Alibaba.com and 1688 alongside external data from Amazon. A marketplace’s self-run toolkit only works inside its own marketplace; Accio pushes to every channel the merchant sells on.”
“It is sourcing to storefront to marketing in one conversation box.”
“This is also where the willingness-to-pay math finally works. A consumer won’t pay for twenty minutes of saved itinerary planning, but a merchant will absolutely pay for an agent that compresses sourcing, negotiation, storefront creation, and marketing into one workflow, because the ROI is measurable and the alternative is headcount.”
“While it’s much easier to get carried away by narratives about the consumer story, the real question is how they can monetize it.
The moat is the rails and the data, not just the model. Qwen commerce works because Alibaba controls Alipay, Taobao, and Amap, and Accio is defensible because of the supplier graph, while Qwen the model lags GLM, and it barely matters domestically.
Similarly, if you think about it, whether it’s the smart cockpit business Banma or Dingtalk, B2B monetizes before B2C, because businesses buy productivity and consumers buy value for money, and until token costs fall far enough, or someone solves the liability question, the consumer agent is a moat-defending feature rather than a revenue line.
And the reason why I’m so gung-ho about Accio is that the merchant use case is obviously chargeable and defended by a real edge.
And despite some concerns around the company’s heavy spending, one small detail that gives me some confidence in it is that the mgmt is now taking a page from ByteDance and running A/B teams apparently on various products - including the foundational model layer- to see who can outcompete and win out. Maybe that kind of “wolfness” or hunger is back at Alibaba?”
5 Big thinking
Paul Graham explores “When to do what you love” in this essay - it’s worth pondering for people at all stages of life…Go do it here:
https://paulgraham.com/when.html
Some Takeaways
“There’s some debate about whether it’s a good idea to “follow your passion.” In fact the question is impossible to answer with a simple yes or no. Sometimes you should and sometimes you shouldn’t, but the border between should and shouldn’t is very complicated. The only way to give a general answer is to trace it.
When people talk about this question, there’s always an implicit “instead of.” All other things being equal, why wouldn’t you work on what interests you the most? So even raising the question implies that all other things aren’t equal, and that you have to choose between working on what interests you the most and something else, like what pays the best.
And indeed if your main goal is to make money, you can’t usually afford to work on what interests you the most. People pay you for doing what they want, not what you want. But there’s an obvious exception: when you both want the same thing. For example, if you love football, and you’re good enough at it, you can get paid a lot to play it.
Of course the odds are against you in a case like football, because so many other people like playing it too. This is not to say you shouldn’t try though. It depends how much ability you have and how hard you’re willing to work.
The odds are better when you have strange tastes: when you like something that pays well and that few other people like.”
“There are even some people who have a genuine intellectual interest in making money. This is distinct from mere greed. They just can’t help noticing when something is mispriced, and can’t help doing something about it. It’s like a puzzle for them.”
“In fact there’s an edge case here so spectacular that it turns all the preceding advice on its head. If you want to make a really huge amount of money — hundreds of millions or even billions of dollars — it turns out to be very useful to work on what interests you the most. The reason is not the extra motivation you get from doing this, but that the way to make a really large amount of money is to start a startup, and working on what interests you is an excellent way to discoverstartup ideas.
Many if not most of the biggest startups began as projects the founders were doing for fun. Apple, Google, and Facebook all began that way. Why is this pattern so common?
Because the best ideas tend to be such outliers that you’d overlook them if you were consciously looking for ways to make money.”
“So there’s something like a midwit peak for making money. If you don’t need to make much, you can work on whatever you’re most interested in; if you want to become moderately rich, you can’t usually afford to; but if you want to become super rich, and you’re young and good at technology, working on what you’re most interested in becomes a good idea again.”
“When you can’t decide which path to take, it’s almost always due to ignorance. In fact you’re usually suffering from three kinds of ignorance simultaneously: you don’t know what makes you happy, what the various kinds of work are really like, or how well you could do them.”
“What do you do in the face of uncertainty? Get more certainty. And probably the best way to do that is to try working on things you’re interested in. That will get you more information about how interested you are in them, how good you are at them, and how much scope they offer for ambition.
Don’t wait. Don’t wait till the end of college to figure out what to work on. Don’t even wait for internships during college. You don’t necessarily need a job doing x in order to work on x; often you can just start doing it in some form yourself. And since figuring out what to work on is a problem that could take years to solve, the sooner you start, the better.
One useful trick for judging different kinds of work is to look at who your colleagues will be. You’ll become like whoever you work with. Do you want to become like these people?”
“If you choose a kind of work mainly for how well it pays, you’ll be surrounded by other people who chose it for the same reason, and that will make it even more soul-sucking than it seems from the outside. Whereas if you choose work you’re genuinely interested in, you’ll be surrounded mostly by other people who are genuinely interested in it, and that will make it extra inspiring.”
“The less sure you are about what to do, the more important it is to choose options that give you more options in the future. I call this “staying upwind.” If you’re unsure whether to major in math or economics, for example, choose math; math is upwind of economics in the sense that it will be easier to switch later from math to economics than from economics to math.”
“There’s a lot of selection bias in advice about whether to “follow your passion,” and this is the reason. Most such advice comes from people who are famously successful, and if you ask someone who’s famously successful how to do what they did, most will tell you that you have to work on what you’re most interested in. And this is in fact true.
That doesn’t mean it’s the right advice for everyone. Not everyone can do great work, or wants to. But if you do want to, the complicated question of whether or not to work on what interests you the most becomes simple. The answer is yes.
The root of great work is a sort of ambitious curiosity, and you can’t manufacture that.”
6 Depth is the new luxury
Have a Great weekend when You get to that stage,
Sune