I don’t know how much of this tradition is in earnest and how much is self-exoticisation to promote tourism (another pillar of Icelandic GDP), but it doesn’t really matter for the point I’m hoping to make. 

Such mystical overlays are familiar to me from the Middle Eastern stories of djinn I listened to as a child, and more recently from my experiences in the field of quantitative trading. And that’s where it starts to get interesting. 

In systematic trading, where you expect to find cold logic, data and algorithms, you instead encounter halls of mirrors, voyages into the psyche and, above all, wizards. 

Jack Schwager: Do you remember your worst losing week?

Mark Weinstein: I haven’t had any losing weeks during that time, but I have had some losing days. 

Jack Schwager: That is an incredible statement. How can you be sure that you are not simply forgetting about a few weeks when you lost money trading?

Mark Weinstein: The reason I am sure is that I remember all my losses.

When I first started being involved in trading and managing other people’s money in 2018, I read Jack Schrager’s Market Wizards: Interviews with Top Traders – a book which by then was already forty years old. I quickly it followed it up with two more books in the series, finding some kind of catharsis in the blunt, non-corporate, hindsight-blessed recollections of other people who had been in the same position as me. I didn't spend much time wondering whether those narratives were technically true.

Mark Weinstein, quoted above, was a real estate broker who started trading with great success in the 1970s and who claimed in the book that 99% of his thousands of trades were profitable. This claim has since been met with skepticism.

I don't see why Weinstein would fib. I was and still am inclined to take him at his word. This is not the same thing, however, as taking his word at face value – at least, not in the way that others have. For example, I’m not sure that he ever claimed to be a wizard, despite the context in which his claim was published. 

This brings us to one of three key concepts that I think Market Wizards helps a reader to decipher and appreciate, so long as it is read alongside other books that offset its imbalances:

#1: Skill is something we see regardless of whether it’s there

Market Wizards is an unwittingly perfect illustration of survivorship bias, luck-attribution bias and a number of other behavioural biases that constantly bedevil investing. 

If you plot a distribution of profits of a sample of traders with similar strategies and abilities, you’ll get something like a normal curve. Except, it won’t exactly be a normal, bell-shaped curve – it will have fat tails on either side. Purely as a consequence of luck, or randomness, there’ll be many traders in the left-hand tail who get wiped out by losses and never heard from again, while those in the right-hand tail get rich and get interviewed. The chart will be symmetrical, but the storytelling won’t. 

Market Wizards is therefore most valuable when read alongside, and used as an illustration of, a book like Nassim Nicholas Taleb’s Fooled by Randomness. You get the sting and then the antidote. You end up understanding just how hard it is to stop being biased against luck, even after you’ve learned that such a bias lives inside us all. 

#2: Profitable trading exploits market inefficiencies, and history is how you learn to spot them

If you start with a classic book like Reminiscences of a Stock Operator, written in the 1920s, and then go through the Market Wizards series, you'll end up with a trading history that covers a good chunk of the 20th Century. Focus on certain details and you'll see that this is largely a history of how market inefficiency has evolved over time.

You’ll read interviews with traders who started out in shady boiler rooms, who gained an advantage by positioning themselves upstream on the flow of information, and who double-timed as both traders (trading for themselves) and brokers (trading for others). 

Stuff like that shouldn’t happen in regulated markets, but it happened back then and it still happens now. In fact, once you learn to recognise it, you quickly suspect that it happens more in the 21st Century than it ever did before – thanks to conflicts of interest between modern exchanges and their customers, bot-driven wash trading and other high-tech versions of the same old tricks. 

#3: Strip away the mystique and what's left is trend-following, which actually works

Let's take a look at one last quote:

Jack Schwager: I would assume, given the consistency of your success as a stock investor for over twenty-five years, that you don’t think very much of the random walk theory.

William O’Neill: The stock market is neither efficient nor random. It is not efficient because there are too many poorly conceived opinions; it is not random because strong investor emotions can create trends.

The magic word here is “trends,” because I suspect that many if not most of the successful traders interviewed by Schwager were involved in trend-following strategies to some degree, even if this is barely touched upon in the books. 

In other words, those traders were capitalising on the well-reported tendency of stocks that have gone up in price to keep going up in price. This has been one of the most sustained and most mined market inefficiencies in history. It has fed many of the biggest quant funds – including those that have given the impression of having highly sophisticated algorithms, whereas in fact those algorithms could largely be written on the back of an envelope. Trend is huldefolk: something that should not exist, yet it does. 

So, perhaps the biggest benefit of reading Market Wizards is that it’ll lead fledgling traders to the phenomenon of trend-following as possibly the only realistic way of getting anywhere close to the results those wizards describe.

Perhaps that sounds unrealistic. But this is the age of AI, and what better use of an AI coding tool than to 'vibe-code' your own trend-following trading bot? All that’s needed is a reliable source of price history (such as the free data available from Yahoo Finance), a source of trend strategies (Robert Carver has published some good ones), plus an API-friendly exchange to run the bot on, and finally some initial capital. Pretty soon you’ll be able to target a reasonable ratio of return to risk. 

Just don’t be shocked if, before you reach that destination, the faeries and elves lead you on a merry dance of overfitting, data mining and in-sample forecasting. Don't be surprised when you meet all the biases you were sure you'd be able to avoid – the conviction that you're too smart for them being the most stubborn one of all. And whatever happens, don't expect a 99% win rate. 

It’s a strange book-pairing perhaps, but it led me to a fundamental question: Why aren’t more British companies participating in the AI chip boom?

After all, computing is a very British topic. It's a montage that typically begins with daguerrotypes of Babbage and Lovelace in the 1830s, followed by black and white photos of Alan Turing and the Enigma machine a century later. We tend to skip a few decades to give people a chance to pop to the loo, before fading back in to colour footage of the 1981 BBC Micro, Acorn Computers and the UK's one truly big commercial success story: ARM, which is now publicly owned with a $250bn market cap (through Nasdaq not the LSE, mind you). 

But none of this stuff takes up much space in Chip War, and for good reason.

Chris Miller is an American economic historian and a Russia specialist who zooms in on different decades and different locations to tell a more commercial story: from the development of the transistor at Bell Labs in 1947, through to the first mainstream applications in the 1950s and 60s, supported by US military investment (mainly for chips to be used in missile guidance) during the Cold War, finally culminating in the early 1990s with the rapid growth of US chip companies like Intel, AMD and joined later by NVIDIA (together worth around $7.5 trillion today). 

The story then switches to Asia, as American executives went on the hunt for cheaper, non-unionised labour to make their chips. This is one of the best parts of the book, covering how small countries like South Korea and Taiwan, which were even more impoverished after WWII than the UK was, turned potentially superficial and exploitative contracts into long-term domestic wins such as TSMC ($1.9 trillion market cap) and SK Hynix ($1 trillion), not to mention more general electronics companies like Sony and Samsung. 

Britain is largely forgotten at this point. But insofar as Chris Miller provides an answer to my initial question, he gives it indirectly: the book repeatedly returns to the theme of risk-taking, and specifically the need for risk-taking by individuals to be matched by government risk-taking through long-term industrial policy commitments that survive political turnover. 

One of the individuals covered in the book is Intel’s Andy Grove, born in Hungary as András Gróf, who found in the United States not only safety from the Nazis, but also proactive support for his entrepreneurial drive – including a significant share of the $1bn (in modern dollar value) distributed to Intel and other chip companies by the US Defense Department. Such government help has continued into recent years via grants under the CHIPS Act (some of which have been converted into a shareholding in Intel, as I’ve covered here.) 

Miller points out clear parallels with the journey of Morris Chang, who fled China as a child during the Japanese invasion, studied in the US and eventually settled in Taiwan. Like Grove, the uncertainties Chang experienced as a refugee may have fostered his ability to cope with risk. But what really mattered after that was the support of the Taiwanese government, which effectively became Chang’s anchor investor when he pitched the idea of building TSMC. 

Contrast these examples against the story of Inmos – a British company that is understandably not covered at all in Chip War. It was founded with the equivalent of around $300m of public money in the 1970s – a significant sum – and built a chip fabrication plant in Wales. It was then abandoned when Margaret Thatcher came to power, and ultimately folded into what is now the Swiss-Dutch company STMicroelectronics, whose share price has surged 170% in the past six months (a statistic I mention in a "look what you could have won" sort of way).  

Committed British support for its homegrown chip industry is today barely worth $1.3bn spread over a decade. This is orders of magnitude lower than what others have been spending, as shown in the CSIS chart above, which is based on BCG and SIA data. Even when you look at spend per capita, the gap is enormous: $18 in the UK versus over $100 in the US, EU and China.

This is hardly a recipe for change. If anything, the data should point investor attention more towards the EU than to the UK. However, returning to Ackroyd’s Foundation, which I’d recommend to anyone who's interested in island stories, I wouldn’t be too disparaging about the UK’s technological future, especially with major geopolitical schisms looming elsewhere. What the country has evidently lacked in sustained political will and industrial spending, it might make up for with other, deeper sorts of continuity – not least the sort of legal, linguistic and intellectual openness that produced Lovelace and Turing in the first place. 

In the mid-1700s, East Prussian demand for timber led to untidy German forests being cut down and replaced with monocultural rows of fast-growing Norway spruce. Provincial magistrates rewrote the law so that gathering fallen wood from the new plantations was treated as theft. A little-known newspaper editor called Karl Marx protested angrily at this "legal lie", but he was ignored. No trespassers could be allowed to interfere with the the new, standardised trees — known as normalbaume.

Unfortunately for the Prussians, their economic bonanza lasted exactly one normalbaum generation. A century later, when the second cohort of spruces was due to be harvested, the loss of understorey, fungi, and other critical biodiversity caused the trees to die where they stood. Yields collapsed by 30%.

In his 1998 book Seeing Like a State, the political scientist and well-travelled anthropologist James Scott used this story to illustrate his central argument: that ignorance is inherent to top-down state control, and that such control tends to be self-undermining because efforts to make the real world more orderly and legible end up destroying the things that kept it functioning. It was Scott's idea of legibility — the simplification of reality to make it easier to exploit — that first drew me to his work, and led me to ask whether it bears any relationship to the kind of world-reading that is regarded as a central feature of AI.

Scott juxtaposes legibility with an ancient Greek terms that he uses to describe practical, local, hands-on knowledge: metis. Metis is what a forest dweller has about their own neighbourhood, what an experienced craftsperson has about their materials, or what a long-time resident has about the rhythms of their street. It is knowledge that cannot be written down, captured in a dashboard, or aggregated into a model. For readers who identify more with mushroom pickers than with magistrates, the legibility/metis distinction at first seems hopeful: it suggests that even the most ambitious controlling project will be checked by the irreducible texture of ordinary life.

The hope doesn't last. Later sections of Scott's book document case after case in which legibility crushes metis — the industrial revolution being one of legibility's greatest wins, with artisans demoted into factory workers, their specialist skills forgotten or bequeathed to machines.

It is this later part of the book that feels most relevant now, as we enter an AI era that Scott, who died in 2024, sadly cannot analyse for us. If we accept that AI is more than automation and more than a passive utility — in other words, that is more like a form of agency — then it can vastly enhance the data-gathering and data-interpreting capacities of any state or state-like actor with sufficient compute.

With AI agents reading sensors in soil, air, tree bark, and human communication networks, there is nothing to stop a state from understanding a forest better than a family that has lived in it for generations. Imagine the film Das Leben der Anderen, except with sixty-three Stasi for every citizen rather than the other way around. Perhaps those AI agents would wield metis as expertly as ordinary people do — and use it to arrange them into neater rows.

The residual optimism in Scott's framework, transposed to AI, is this: a sufficiently authoritarian deployment of AI may itself become a legibility project that yields spectacular returns for one generation and then fails, because it cannot fully capture the human realities used to train it. The most capture-resistant of those realities, in Scott's analysis, was always civil society. He spent years living in Malaysian villages and wrote extensively about civil resistance — the titles of his other books (Weapons of the Weak, The Art of Not Being Governed) tell us plenty about his disposition.

If Scott were here today, he might even argue that individual access to AI — preserved and used to circumvent rather than serve top-down control — could generate a surplus of time and energy that strengthens the civil fabric, rather than erodes it, allowing people to remain just as untidy and illegible as ever.

I listened to Rogoff being interviewed when his book launched last year and decided it was worth going straight to the source. I particularly wanted to understand the so-called "convenience premium" that podcasters kept asking him about. (In a nutshell, that's the estimated $120bn bonus the US government gets every year as a result of being able to borrow at a lower interest rate than other governments, because its dollar debt is so darned convenient for lenders to hold or use as collateral.)

As it turned out, the convenience premium was the least interesting part of the book. The richest material is in the first half, when Rogoff risks wandering off topic to give us journal-like impressions of his global travels as an international chess player, academic and eventually as chief economist of the IMF. Three of those digressions were especially useful to my under-construction understanding of chokepoints in the global financial system — particularly the role of non-state money (Bitcoin), modern political currencies like the euro, and supranational institutions like the IMF.

1. Bitcoin has accumulated intrinsic value for reasons that aren't fully appreciated

Rogoff is no crypto evangelist, but he has a refreshingly pragmatic view of Bitcoin. His early travels as a chess player gave him a sharp ear for what he calls the "underground economy" — the cash-and-quiet transactions that have sat outside formal banking. For decades that economy ran on hundred-dollar bills. It now runs increasingly on Tether, USDC, and Bitcoin – with Bitcoin being the option that is least susceptible to weaponisation risks that affect dollar stablecoins and the dollar itself. For Rogoff, that's enough to argue crypto is not "going to zero", the way so many other commentators assume, even as he warns that its unregulated status makes it ripe for crises.

Where I part company with Rogoff is on what a crypto crisis would actually do to Bitcoin specifically. He treats "crypto" as a single category, but a run on USDC or USDT — possibly triggered by fears about their traditional financial collateral or the chain of institutions underwriting them — would likely increase Bitcoin's price, not decrease it. Stablecoin holders rushing to preserve wealth would move into Bitcoin via decentralised exchanges, which is exactly the asset's failure-mode value proposition.

2. The euro is the most underrated success story in modern monetary history

Rogoff readily admits he expected the euro to fail, and to fail miserably. Instead, against the institutional incoherence of the eurozone and a generation of crises, it has become the world's second currency and the only fiat currency with any plausible claim to challenge the dollar – and this has been achieved through the sustained force of political will. Despite all the debate in the UK about Brexit and European integration, I've remained largely oblivious to the scale of this achievement.

The chart above shows there's little sign of the euro gaining ground against the dollar right now. Nevertheless, reading Rogoff's chapters made me more determined to maintain some euro exposure in my own savings, not only because I might want work and live on the "mainland" in the future (I spent some blissful years in Bordeaux during Covid and would love to go back there), but also because I think it's a smart diversification in principle, just in case GBP continues to tank. (In the early days of the euro, you could get 1.5 of them to the pound – now you get only 1.15.)

3. The IMF explains why the US still bothers with multilateralism

With the US exiting so many international treaties and agreements, it has struck me as a conundrum that it has remained so active and committed within the International Monetary Fund. Reading Rogoff's book, it dawned on me that I had very little idea of how the IMF really worked.

This is what I learned: The IMF runs on an internal reserve asset called the Special Drawing Right (SDR), which lets selected crisis-borrowers access something close to rich-country interest rates. It is the rich countries that decide which poor countries get IMF support and which don't, which makes sense since its the rich countries offering the help. Except, IMF lenders like the US aren't actually putting their own wealth at risk in making these loans, so much as sharing limited access to their own borrowing rates, which are of course cheap thanks to the convenience premium mentioned earlier. Rogoff's insider explanation makes it clear that the IMF remains a relatively low-cost, high-leverage extension of dollar power. Defunding it would cost the US far more influence than it would save in dues. Rogoff makes it clear that he doesn't even like the IMF's lending mechanism and would rather just see rescue packages given as aid – which really says it all.

Shih’s theory, which has since been observed across multiple industries and geographies, holds that more value in manufacturing tends to accrue to IP- and brand-owning companies at the edges of the value chain, than to pure fabricators in the centre. 

Applying this idea to the semiconductor industry, you expect to find valuable design companies like NVIDIA and ARM at one end of the smile and consumer-centric giants like Apple at the other. 

In the middle of the trough sits TSMC. 

Is the market hoping Intel will join it? If so, that’s a tough journey with an even tougher destination. 

As the world’s dominant chip fabricator and one of the most successful manufacturers of all time, TSMC is continually forced to take on the risk of huge capex investment (expected to exceed $52bn in 2026, more than 3x Intel’s planned spend). This investment is funded by profit margins that are 20+ points worse than those of its fabless customers (one extreme metric to illustrate this: TSMC shareholders end up with a 40% return on equity, versus NVIDIA’s 115%). To survive the infamous cyclicality of its industry, TSMC must also maintain a war chest of over $25bn in idle cash, equivalent to half a year of capex that cannot be put to work. 

Geostrategically, this all makes sense. That’s why Taiwan props up TSMC and why the US has more recently subsidised Intel. The Trump administration’s conversion of that subsidy into a $9bn equity stake was a smart decision not because that holding is now worth $36bn on paper, but because it gives American industry a shot at an AI future that is less at risk from China. 

Commercially, however, the rationale for Intel trying to emulate TSMC is less clear. TSMC has always been careful never to compete with its own customers. Intel can’t make that promise. Instead, it’s building a plausible firewall between its “Product” and “Foundry” subsidiaries, to the point of having separate ERP systems (by end of 2027) and giving its Product arm the freedom to use TSMC chips if it wishes. But even if customers like NVIDIA or AMD are ready to trust the firewall with their IP, they’re not likely to give a competitor extra cash flow unless it’s absolutely necessary. 

For this reason and others, analysts have envisioned a full spin out of Intel Foundry, similar to how AMD spun out GlobalFoundries. But the US government has already positioned itself to block this. The Treasury's warrants effectively veto any sale of more than 50% of the foundry, which means the structural fix that would unlock customer trust is also the one Intel can no longer take. 

This is perhaps the biggest long-term risk that Intel investors must face: misaligned interests with their co-shareholder, the US government. Investors seem to regard government involvement as a net positive and perhaps even a price floor, whereas it could just as easily be a blocker to strategic agility and growth. 

Ultimately, this could result in the US CHIPS Act facilitating a transfer of wealth from US taxpayers, through Intel, to the fabless designers who ultimately capture the rents from bleeding-edge silicon. That might be a defensible policy outcome in a world at serious risk from US-China schism, but it’s not an investment thesis. 

Recommended Reading: 

The Globotics Upheaval: Globalisation, Robotics and the Future of Work – written by Richard Baldwin, who happens to be a recent exponent of the smile curve theory. 

Chip War, by Chris Miller – more history than analysis, but helps to make sense of the industry as a whole.  

Power and Prediction: The Disruptive Economics of Artificial Intelligence, by Agrawal, Gans and Goldfarb - a more general look at where AI is likely to cause value to accrue, beyond the usual suspects. 

Disclaimer: Published for informational and educational purposes only. Nothing here constitutes investment advice, a recommendation, or a solicitation to buy or sell any asset. I invest only my own capital, with a focus on AI and semiconductors, and may hold positions in assets discussed — only ever as a small part of a broadly diversified portfolio.