One simple signal of leadership quality: how often they say “we” when sharing credit, and “I” when taking responsibility. Warren Buffett (his shareholder letters), Eisenhower (his D-Day failure note), and JFK (owning the Bay of Pigs) come to mind. Who are other great examples?

Hyperscalers are the ultimate toll collectors, and open-source models are their most profitable traffic.

A contrarian view on the impact of AI on jobs: AI will drive the largest increase in productivity per worker the world has ever seen. And basic economics says that if the marginal productivity of a worker goes up, you'll want to hire more workers, and pay them more, to capture new and existing profit pools. The common argument, that AI will do our jobs and leave fewer of them, implies we're living at the end of time, with no more problems left to be solved. But there will always be new problems, and companies are problem-solving machines. As long as traits like judgment, trust, empathy, creativity, and accountability matter, much of that new work will need people. The economic pie is not finite, and every stage of progress has created more jobs to leverage it. In fact, I suspect we'll see an explosion in business formation, given it's never been a better time in the history of the world to be an entrepreneur. History is a good guide here. Cars put most carriage makers, stable hands, and horseshoers out of work. But hundreds of companies entered the auto business in a single decade, followed by dealers, gas stations, repair shops, parts suppliers, and motels, each creating jobs that hadn't existed before. More recently, the internet upended bookstores, video and video game stores, and Yellow Pages publishers, among others. But it also collapsed the cost of starting a company. Solo and small online businesses proliferated, along with jobs like web developer and social media manager that didn't exist 30 years ago. Existing jobs can grow, too. When ATMs emerged in the 1970s, bank tellers looked doomed. Instead, ATMs cut the cost of running a branch, banks opened far more of them, and teller jobs roughly doubled. Mobile banking has since shrunk teller roles, but total bank employment has held steady. The work changed. In the 1960s, journalists wrote about the demise of the workforce because of the mainframe computer. Instead, it drove productivity and became a tool to leverage human potential. Later, spreadsheets automated bookkeeping, but analysis became so cheap that companies wanted far more of it, and accountant and analyst jobs grew. It's natural to fall into "this time is different" thinking, and maybe this time will be. But history suggests otherwise. So, what are the risks? The honest answer requires humility, because the future is uncertain. The biggest risk to my argument is that AI is fundamentally different. Past technologies mostly replaced muscle and routine tasks, and people moved up to work that required thinking. AI goes after thinking itself. If it can do most cognitive work, the old pattern may not hold. I don't think we're there, and the human side of work is harder to automate than knowledge, but it's the argument I take most seriously. Another risk is that productivity gains don't automatically flow to workers. Labor's share of US business output hit a record low of ~53% this year, vs. ~63% for most of the postwar era. More of the gains have gone to those who own stuff vs. those who do stuff. But a smaller slice of a much bigger pie can still leave workers better off, especially if they own some of the stuff too. We'll also need a bridge to the jobs of tomorrow. Jobs with a right or wrong answer will likely get replaced, and jobs that require judgment will be augmented. We've done this before: in 1900, ~40% of US workers were in agriculture. Today it's ~2%, and that shift helped build the modern middle class. I've always found it interesting that the bear case usually sounds smarter than the bull case. Part of it is that the bear can point to specific jobs that will be lost, while the bull has to point to jobs that don't exist yet. Losses are easy to see. Gains that haven't been invented are hard to imagine. As the saying goes, pessimists sound smart, optimists make money. AI will eliminate many jobs, but I suspect it will create far more.

Gold has more than doubled over the last few years. What's actually driving it? I value gold at 1/T, where T = trust in the financial system. Distrust comes from 3 areas: fiscal excess, institutional fragility, and currency debasement. We are currently seeing all 3 play out. Fiscal excess: national debt has reached ~$40 trillion, with debt held by the public >100% of GDP. The 2026 deficit is approaching ~$2 trillion (~6% of GDP) which is the kind of number you'd expect to see during a recession or major war, not during an economic expansion. And interest on the debt has crossed $1 trillion / year, making it the fastest-growing line in the budget. This is an unsustainable spiral: debt creates interest expense, and paying that interest expense requires taking on more debt. Institutional fragility: the credibility of the institutions backing the dollar is coming into question. When the US froze Russia's dollar reserves in 2022, it showed the world that dollar holdings can be seized for political reasons. Gold doesn't have this issue, which is a primary reason why central banks are buying it. At the same time, the lines between the Treasury and Fed are blurring, which is a condition that has historically preceded currency debasement (see Weimar Germany, where the central bank simply funded the government's deficits until the currency collapsed). Currency debasement: look no further than the money supply. M2 (a measure of money supply) has grown from ~$4.6 trillion in 2000 to >$23 trillion today. This is a ~5x increase, compounding ~6-7% / year vs. above-ground gold supply, which grows ~2% / year. So, the supply of dollars is growing ~3x faster than the supply of gold, which is why gold holds its value over time while paper money loses it. What makes this so dangerous is that money isn't being destroyed all at once - it's being quietly diluted, year after year. So, how do you actually see this happening in real time? Through the dollar price of gold, not a basket of other currencies (which are all being debased as well). You can also see it in real / hard assets like real estate and commodities, as well as stocks, which are essentially claims on real businesses with real assets and pricing power. I'm a long-term bull on the United States economy. But trust is what underpins the dollar - and fiscal, institutional, and monetary trust are being tested. Higher gold prices are just Mr. Market's way of pricing that in.

The world only ends once, and if it doesn't, you want to own stocks.

Getting short term moves right makes trading feel easy. Getting them wrong makes it feel hard. And hard is what it is - not because of the analysis, but because there are too many random / unknowable macro variables that are subject to change at any given time. It’s hard because it requires luck. And luck isn’t an investment strategy.

What's going on with interest rates and why does it matter? First of all, interest rates matter to the global economy because they set the price of money, and almost every economic decision is made with that price in mind. Borrowing vs. saving vs. spending, asset prices, currency values and capital flows, government finances, etc. Rates have been moving higher (particularly at the long end) for several reasons (most of which are related): Massive and growing global fiscal deficits. Higher deficits mean more bond issuance to fund those deficits. More bond issuance leads to greater supply. Greater supply without a corresponding increase in demand leads to a lower bond price. Bond prices move inversely with yields - so lower bond prices = higher yields. The most important factor (by far) is the general idea of investors demanding more return (term premium) for the risk of doing business with the United States of America. Our deficit to GDP ratio is ~6% right now and likely to go higher. This is where you'd expect deficit to GDP to be in a recession, not under the backdrop of a (close to) fully employed workforce (you could argue that the employment picture has been aided by government spending / deficits). And in a recession (when tax receipts fall and spending rises to stimulate growth) deficit to GDP typically moves materially higher. So, at some point, government revenue (tax receipts) is eaten up by interest expense, and we need to issue more debt in order to fund our obligations. The US has had the luxury of being the world's reserve currency, which means (among other things) that when global growth slows, investors buy US treasuries / dollars as protection (flight to safety). When investors buy treasuries, prices go up and interest rates go down, lowering borrowing costs. The existential risk is that someday, global growth slows and investors decide to sell treasuries (given our current fiscal position), driving borrowing costs up at just the time we need them to go down. This is the debt spiral / doom loop, and while I don't actually expect this outcome, higher rates on treasuries are Mr. Market's way of discounting a greater chance of this happening. Likely the end of central bank quantitative easing (QE - the Fed buying treasuries on its own balance sheet and creating new money) under Kevin Warsh. In fact, Warsh has signaled that he intends to make balance sheet reduction a primary tool. So, if Warsh lets bonds mature and run off, that removes the Fed as a buyer and forces the private market to absorb more of the existing supply, which leads to lower prices and higher yields. Quasi tightening. What Bessent did this week with treasury buybacks was not QE. This was the treasury issuing other securities to fund the buybacks. Simply rearranging the maturity schedule, with no new money entering the system. By taking out long bonds and funding with shorter issuance, Bessent pulls duration off the market in an attempt to ease the term premium pressure that is driving long yields up. Also, nominal GDP growth (real GDP growth + inflation) is ~5-6%, so it makes sense that the long bond is in a similar range. Nominal GDP is basically the economy's rate of return. Buying a long bond (like the 30-year) means you are lending money to the government for a long period of time. As an investor, it wouldn't make sense to accept a fixed yield below how fast the economy is growing (despite this being the case for several years post GFC) if you could earn more investing in the growing economy (like the stock market). Investing involves understanding the opportunity costs of every decision. On the other hand, there are reasonable arguments for lower rates as well: Productivity. AI should give us the economic holy grail, which is to do more with less, and grow without a corresponding inflationary impulse. This will drive down the cost of goods and services - and this deflationary / disinflationary impulse should drag rates down with it. This is key to Kevin Warsh's belief that we can have an environment of high employment and low inflation against a backdrop of productive economic growth. Demographics. The population is aging. As people age, they save more. Higher savings rates is a deflationary phenomenon (see economies like China and Japan). We are likely closer to late cycle than early cycle. And when you are late cycle, the bias toward growth and inflation (assuming its demand driven) is decelerating vs. accelerating. I don't think anyone can accurately predict the macro with any degree of consistency - it's a fool's errand, given there are too many random / unknowable variables that are subject to change at any given time. However, it's important to contextualize material moves in macro variables, particularly when that variable sets the price of money.

AI infrastructure investment has been the marginal driver of US economic growth. As today's largest and most consequential capital cycle (others being power & grid infrastructure, defense & rearmament, reshoring & advanced manufacturing), it's impossible to understand the broader macroeconomic landscape without understanding the drivers of and risks to AI capex. So, what are those drivers? First and foremost, intelligence is the largest TAM the world has ever seen - every consumer and every business is a potential customer - this is the 4th industrial revolution and companies are investing to capture the massive profit pools that intelligence unlocks. Demand for intelligence / tokens continues to exceed supply, at a time when inference workloads are growing exponentially. @GavinSBaker made an interesting point (one of many) on @InvestLikeBest that only ~250-500k people use agentic AI today, yet the industry is already acutely constrained. If a fraction of a percent of humanity can strain capacity, compute demand has enormous room to grow as adoption scales. Scaling laws (more compute = better model performance) continue to hold. First through pretraining, now through inference-time compute. So, if models keep improving, why slow investment? Competition. This is viewed as an arms race, not discretionary spend, and hyperscalers seem more afraid of being structurally short on compute than overspending. And from a geopolitical standpoint, economic and military strength will always be downstream of technological innovation and industrial capacity. And what are the risks? Hyperscale free cash flow is deteriorating, pushing these companies toward external financing. This is more of a cautionary signal than it is a negative catalyst or near-term risk. Near to intermediate term ROI for AI's end customers continues to be a key question. So far, improved profit margins are showing up for the sellers (Mag 7) but not the buyers (S&P 493). Torsten Slok of Apollo made the point that in business, profit margins are frequently higher for the owner of the end-customer relationship. But that is not the case for AI, where profit margins are higher the further you get from the end user (S&P 500 operating margins are low double digits, while Nvidia's are ~60%). And while we've seen certain industries like software engineering and professional services drive efficiencies, this has not been the case across the board, with many businesses paying up for AI without offsetting productivity gains - a net inflationary phenomenon. The point: profits associated with the AI infrastructure cycle (for now) are largely being funded by investors rather than earned from customers. The history of these capital cycles is that shortages eventually turn into gluts. While hyperscale ROI is incredible (currently <1 year payback on GPU/server hardware), they are making decisions on long lived assets and capital projects based on spot pricing (understanding that spot could fall meaningfully and ROIs would still be attractive). We saw this in rails, telecom / fiber, shale, and others - demand eventually arrived, but it's a risk to whoever's balance sheet is holding the asset when the price gaps down. Regulations like data center moratoriums that would prohibit this spending from taking place. It's also important to contextualize what happened in July and understand why the AI infrastructure complex sold off so aggressively. A couple of causes, with the move exacerbated by a leverage unwind. Kimi K3 (Chinese open-weight model) scared people into thinking open source would undercut frontier lab pricing power / revenue. The market's question: if the labs slow spending because competition is cannibalizing their economic opportunity, who pays for the compute and associated infrastructure? I believe in Jevons Paradox: as inference gets cheaper, usage outruns the price drop, leading to more absolute dollars being spent on AI infrastructure. So interestingly, open-source competition (in my opinion) is actually long term bullish for the downstream AI ecosystem. Mr. Market moves based on rate of change (accelerating & decelerating vs. absolute levels). Stocks were down partly due to a rate of change deceleration in hyperscale capex increases (a function of the law of large numbers). So, what happens next? I suspect the status quo will continue until 1) we see a meaningful / obvious shift in consumer or enterprise monetization and productivity relative to spending, leading the frontier labs and others to scale back spend (curious if anyone has thoughts on what this might look like) 2) credit markets seize up making it more expensive to finance spending plans. I don't see either of these happening near term. Also, this is not how cycles end - I always go back to the Sir John Templeton quote: "Bull markets are born on pessimism, grow on skepticism, mature on optimism, and die on euphoria." By definition, euphoria is the absence of skepticism, and skepticism about AI capex is everywhere right now.

Trump says companies should report earnings every six months. I tend to agree. Reduces short term focus and the need to "hit numbers" sometimes at the expense of longer-term interests. Also lower admin costs associated with preparing and auditing reports - this is highly resource intensive. From an investors standpoint, it likely attracts more patient investors vs. speculative traders. Counterargument is that less frequent reporting could reduce transparency, increase information asymmetry, and make it harder for investors to monitor performance. It may also lead to larger market reactions when reports are released, as more information is packed into fewer updates. bloomberg.com/news/articles/…

Gold continues to do its job, up ~40% YTD in anticipation of rate cuts and continued deficit spending. As I've said, gold is more currency than commodity, and it's telling you something about the value of the US Dollar...

I enjoyed listening to the latest @BG2Pod episode on China. I spent a few weeks in China during the fall of 2023 and came back with a perspective very similar to that of @bgurley. While I believe the US system, culture of risk taking, and (generally) decentralized market approach to solving problems is optimal, that doesn't mean there isn't a lot to learn from our Chinese counterparts. Here's what a prominent Chinese businessman said to Gurley: "Every founder and every VC in China studies the west at a nauseating level. They listen to all the podcasts, they read everything they possibly can, they study any speech, they look at the financials - and he said, the west doesn’t do that of China." Maybe we should. He also said of @danwwang's book Breakneck: "the vast majority of the Politburo (the principal policymaking committee of a communist party) are engineers - this is the ruling party within China. While the vast majority of people in Washington DC are former lawyers. He uses this lens to suggest this is why they are great at building things and why they aren’t so great at social things.” Bill also discusses his visits with various tech companies, specifically EV manufacturers like @BYDCompany. I very much agree that the combination of low cost and high quality make these products extremely competitive on the global stage. One of Bill's takeaways after this meeting was that: "the US has too many people in government who's job is to block things." This is the bureaucratic red tape that is common in systems like our own - a certain amount is necessary, but there's a point at which it becomes value destructive - and I believe we are past that point. I'm a US optimist and believe we will always be the best house on the block, but I think it also makes sense to remain open to the idea that other societies may have arrows that we could benefit from putting in our quiver.

I generally subscribe to the traditional belief that lower interest rates stimulate inflation by encouraging borrowing and spending, which increases demand for goods and services, and puts upward pressure on prices. However, I also believe that the opposite can be true for heavily indebted economies with growing deficits (like the US). The argument is simple: lower interest rates = lower debt service costs (interest expense) = lower deficits (interest expense will comprise >50% of the US deficit in 2025) = less money printing = lower inflation. This is a dynamic I continue to think about and would be curious what others think.

Axon $AXON continues to be a problem solving machine. The bull case has always included their ability to compound bookings growth ~30% for a long period of time. Remember, its the T (time - the exponent) in the compounding formula that does the heavy lifting. This will be yet another year of ~30% bookings growth...

Strong opinions, weakly held

AI capex remains robust and the companies downstream of that capex can sleep well at night. @satyanadella mentioned last night that demand for data centers / compute continues to exceed supply. Here's a ridiculous stat: Azure generated $75B in revenue over the last fiscal year, and last quarter that business grew 39% y/y... This is a 400bps acceleration from the prior quarter, following Google $GOOGL Cloud's acceleration last week. AWS $AMZN tonight. Increasing AI capex feels more justified in the context of accelerating cloud and ad revenue streams $META (ad revenue +22% y/y). This is the R in ROI.

The consumer remains resilient $V $COP $JPM $CITI and continues to spend on travel and experiences $BKNG $RCL $UAL $DAL $RYAAY, AI infrastructure spending is healthy $VRT $GEV $GOOGL, and M&A activity looks ready to explode $GTLS / $BKR, $NSC / $UNP, $PANW / $CYBR. Uncertainty around goods $UPS $NSC $WHR and autos $NXPI $TXN $GM $STM. Ad spend (typically first to get cut as business confidence drops) squishy $WPP but more evidence needed.

Remember that in 2019, the last full year before the pandemic, the US government spent ~$4.45 trillion. In 2024 the federal government collected ~$4.9 trillion in revenue. This means that if we brought government spending back to pre-Covid levels, we'd be running a budget surplus. This is meant to be an encouraging sign, suggesting that its possible to get our fiscal house in order.

Listening to Jensen $NVDA at @theallinpod AI summit reminds me how dynamic a leader he is (see quoted post below). A few interesting notes from the conversation: "There’s going to be a whole industry of just producing tokens. And this is going to be the new infrastructure - just as we have the energy production infrastructure and the internet infrastructure - now we need to build out the AI infrastructure. My sense is that we’re a few hundred billion dollars into a multi trillion dollar infrastructure buildout (per year). And the reason for that is because you want the new infrastructure which increases revenue and drives your costs down." On physical world AI: "Everything in the world that moves will be autonomous someday. And that someday is probably around the corner."