cafeconomics

For the past several quarters, the U.S. Treasury has run a quiet but consequential experiment in how a government finances a multi-trillion-dollar deficit: lean hard into short-term T-bills, hold coupon (note and bond) issuance roughly flat, and hope the gap gets filled later once conditions look friendlier for locking in long-term rates. Market participants call it the "T-bill skew." The logic sounds clean in a slide deck. In practice, it's a bet — on the Fed's path, on oil, and on the world's appetite to keep rolling over America's debt every few weeks without asking for more in return. As of this week, that bet is genuinely in play, not just theoretically at risk. The strategy, in plain terms Treasury debt comes in two broad flavors: bills (maturing in a year or less) and coupons — notes and bonds (2 to 30 years). When the Treasury issues more bills relative to coupons, it's choosing to finance the government with debt that has to be refinanced constant...

For any observer of emerging markets, India’s macroeconomic scoreboard presents a staggering picture of resilience. In a global environment bogged down by geopolitical friction and energy shocks, India’s headline real GDP grew at an enviable 7.7% for the financial year 2025–26. On paper, it is a performance that commands national pride and global celebration. Yet, beneath this golden veneer lies an agonizing macroeconomic puzzle that has split the economic community down the middle. If the headline growth is rocketing past 7%, why are the captains of Indian industry refusing to build new factories? This paradox—recently brought into sharp relief by former Reserve Bank of India (RBI) Governor Dr. Raghuram Rajan—points to a profound structural divergence between Gross Fixed Capital Formation (GFCF) and headline growth. For a nation looking to cement its status as a global manufacturing superpower, understanding this mismatch is not merely an academic exercise; it is an urgent economic di...

Part 1: The Foundations — The Data Centre Life Cycle The critical challenge of managing data centre infrastructure is navigating a massive economic mismatch: the heavy physical shell (the building, generators, and concrete) is designed to last 15 to 25 years, while the logical architecture inside it (the servers, switches, and GPUs) faces economic and technological obsolescence every 3 to 5 years. To manage this contradiction, operators visualize the facility through a cyclical loop. The following comprehensive infographic illustrates the primary phases of a data centre’s life, from ground-breaking to physical destruction. Phase 1: Strategy, Site Selection & Feasibility This is the multi-year planning phase where the biggest long-term cost decisions are locked in. The focus is securing the key inputs: cheap power, available land, and ultra-fast network connectivity. Power and Grid Access : Securing hundreds of megawatts of guaranteed grid capacity, preferably near renew...

​The economics of artificial intelligence are fundamentally distinct from previous technological shifts. Traditional rules governing industrial production, and even early digital products, break down when applied to large language models. This exploration tracks a series of interconnected structural realities: how diminishing returns apply to the foundational inputs of machine learning, whether base model training can ever be truly finalized, the trajectory of synthetic data, and the political economy of user data compensation. These insights represent a record of thinking in progress for an unmapped economic landscape. ​I. Diminishing Marginal Utility in Machine Learning ​Data: A Classic Case of Diminishing Returns ​When evaluating data and compute, data represents a classic textbook resource governed by diminishing marginal utility. The first thousand training examples provide massive learning gains, whereas subsequent millions yield progressively less. Error rates in neural netw...

Part 1: From a Single Neuron to Supercomputer Clusters Artificial intelligence often feels like a monolithic, almost mystical entity. We talk about it as if it’s a singular, omniscient mind living in the cloud, but the reality is far more grounded—and honestly, far more fascinating. Stripped of the marketing hype, modern AI is an incredible feat of scaling, built by taking one microscopic, incredibly simple decision-making unit and multiplying it by the billions. To truly understand how AI works, we have to look past the sci-fi imagery and explore how a tiny digital seed grows into a global supercomputing network. The Microscopic Seed: The Artificial Neuron The entire universe of modern AI begins with a single, humble building block: the artificial neuron. Loosely inspired by the biological cells in the human brain, an artificial neuron is essentially a tiny, automated calculator designed to make a single choice. It doesn't possess wisdom; it just processes data. When information f...

Diminishing Marginal Utility (DMU) is the idea that while total utility always increases when you consume more, the marginal utility—the satisfaction added by the very last unit consumed—always falls. It’s the "extra" joy that decreases. For example, a second slice of pizza doesn't give you as much satisfaction as the first one and third slice of pizza is not as delicious as the second one and so on. The same law is applicable in production theory as well. As you increase one factor input, holding everything else constant the marginal productivity of the factor declines. But this law is not only limited to economics and it isapplicable in many other forms in our day-to-day lives, as described below. The Law of Diminishing Marginal Actress Utility . ​When a new actress debuts, the marginal utility is astronomical—the silver screen lights up, and audiences appreciate her presence like a rare cinematic event. But because commercial cinema is relentless, director...