Version 2

Milk Road AI's 'Save This' series, running through June 15, 2026, argues that Corning, Qualcomm, and Micron are undervalued AI infrastructure beneficiaries — and extended the thesis on June 15 with a Chamath Palihapitiya data point: a gigawatt of compute now costs $100 billion, up roughly 20x from the start of his Arizona data center project [12]. HBM market data showing SK Hynix at 62% share directly challenges the 'monopoly' framing Milk Road AI applied to Micron [9][10], while a separate thread argues Chinese memory capacity expansion — not demand collapse — is the most likely peak signal for the memory cycle [13].

The series distills a retail investor thesis gaining traction in 2026: that durable AI gains may sit in physical infrastructure constraints — fiber, memory, edge chips — rather than in AI platform companies. The Palihapitiya compute cost figure gives the thesis a quantitative anchor, but the emergence of a Chinese oversupply counter-argument and concrete HBM market share data add friction to the simplest versions of the Micron bull case.

Milk Road AI's 'Save This' series, published between June 8 and June 15, 2026, argues that Corning, Qualcomm, and Micron are underappreciated beneficiaries of AI infrastructure spending — each overlooked relative to Nvidia and the headline hyperscalers. Each post uses a recent news event as its anchor and frames physical infrastructure constraints, not model or platform leadership, as the durable investment story.

The Corning post (June 8) coincided with Amazon signing a multibillion-dollar, multiyear agreement to supply optical fiber, cable, and connectivity solutions for US data centers, expected to create roughly 1,000 jobs in North Carolina [1][2][3]. This followed earlier fiber supply agreements with Meta and Nvidia, contributing to a 115% year-to-date stock gain for Corning by mid-June 2026 [4]. The Qualcomm post, also June 8, treats a remark by Jensen Huang as an implicit endorsement: asked whether Nvidia would enter the smartphone chip market, Huang said no and described Qualcomm as 'doing a great job' [5]. Financial media including Motley Fool and MSN interpreted this as Huang signaling Qualcomm's unchallenged position in on-device AI inference [6][7]. The Micron post (June 14) is the most aggressive of the three, predicting a $4,000 stock price and describing Micron as 'America's monopoly on the most strategically critical material in the AI buildout' [8] — a claim directly contested by HBM market data showing SK Hynix holding 62% of the market, with Micron and Samsung contesting second place [9][10][11].

On June 15, Milk Road AI extended the series with a data point from Chamath Palihapitiya: a gigawatt of compute now costs $100 billion, compared to $4–5 billion when he began his Arizona data center project — approximately a 20x increase in a single investment cycle [12]. The post frames this cost escalation as the central explanatory figure for the AI infrastructure thesis, providing a quantitative argument for why physical infrastructure suppliers face durable demand regardless of which AI models or platforms prevail.

A counter-thesis has also appeared: Convequity argues that Chinese memory capacity expansion — not a collapse in demand — is the most probable signal that the current memory cycle has peaked [13]. If Chinese producers scale HBM-adjacent capacity faster than anticipated, the supply constraint that underpins the Micron bull case would erode. None of the bullish accounts tracked in this thread engage with this argument.