Ignore the narrative that AI investing is a simple two-way bet between hardware and software. Look at the data. Over the past 18 months, two signals have emerged from the same AI review process—a systematic analysis of 200 podcast episodes covering the entire ecosystem. One signal delivered a 180% return on Micron. The other missed a $6 billion acquisition of Cursor. The gap between these outcomes is not luck. It is a structural feature of how capital flows through AI's macro cycle.
Illusions dissolve under stress testing. The stress test here is the asymmetry between the 'safe' hardware bet and the 'volatile' software bet. Most allocators treat hardware as a bond proxy—steady demand, proven capacity. Software is treated as venture lottery. But the reality is inverted. The hardware trade is a liquidity trap disguised as a value play. The software trade, when timed right, offers asymmetric upside that most institutional frameworks systematically undervalue.
Context: The AI Liquidity Gradient
To understand why one trade succeeded and the other failed, we must map the global liquidity flows behind AI. In 2023, central banks emitted a wave of liquidity through fiscal expansion and rate cuts. That liquidity first flowed into the most capital-intensive layer of AI: infrastructure. Semiconductors, data centers, energy. Micron, as a key HBM3E supplier to NVIDIA, sat directly in that channel. Its 180% return was not a bet on technology superiority; it was a bet on the vector of money. Follow the vector, not the hype.
Cursor, on the other hand, sits in the application layer. Application-layer AI requires a different liquidity profile—not capex-driven, but adoption-driven. In 2023, the adoption curve for AI coding tools was still S-shaped. Many dismissed it as a feature, not a product. The $6 billion acquisition marked the inflection point where the market recognized that developer tools are the new operating systems. But by then, the entry price was already compressed.
Core: Deconstructing the Returns
Let's dissect the Micron trade. In my years auditing on-chain liquidity, I learned that institutional capital chases the path of least friction. Micron's HBM business exhibited clear frictionless demand: hyperscalers had to buy memory to run training clusters. No alternative existed. The margin of safety was the demand inelasticity—hyperscalers could not delay procurement. That is a classic macro short: buying a supplier with non-discretionary demand during a liquidity expansion.
The Cursor miss illustrates a different friction. The deal's $6 billion valuation was not visible to the standard discounted cash flow models used by most macro desks. Cursor was a 'growth at all costs' story, but its unit economics were masked because the market for AI coding was still nascent. Traditional value screens would flag it as overvalued relative to revenue. But the revenue multiple hides the structural shift: coding tools are becoming the primary interface to compute. The miss was a failure of framework, not of foresight.
Volume without conviction is just noise. The capital that flowed into Micron was high conviction but low information gain. The capital that ignored Cursor was high information gain but low conviction. The differential is a symptom of institutional inertia.
Contrarian: The Decoupling Myth
A popular narrative claims AI hardware and software are coupled—what benefits one eventually lifts the other. I challenge that. The coupling is asymmetric. Hardware benefits from a rising tide of macro liquidity; software benefits from a concentrated wave of micro adoption. When liquidity contracts, hardware de-rates first because it carries inventory risk and capex commitments. Software, being asset-light, can compress multiples but often retains a floor from recurring revenue.
The Micron trade worked because we were in a liquidity expansion phase. The Cursor miss happened because we underestimated the speed of adoption in a liquidity-neutral environment. The next phase—liquidity tightening—will reverse this. Software will hold value better than hardware. The floor is a trap for the impatient; those who buy hardware at these levels may be buying peak cycle.
Takeaway: Positioning for the Next Vector
The 180% gain and the missed $6 billion are not lessons about stock picking. They are lessons about timing factor exposure. In the current market—sideways, choppy, waiting for direction—the smart money is rotating from hardware plays that have already priced in the HBM boom to application-layer picks that still trade below their adoption curve.
My model suggests the next vector will be AI agents and infrastructure software that enables agentic workflows. Cursor was an early signal. The market is now pricing in a repeat. This time, do not let the conviction lag the data.